Build & Sell with Codex (5+ Hour Course)

0:00 I'm about to take you from a complete beginner to a pro AI builder with Codeex. And you don't have to have a technical background at all. My name is Nate. I have no technical background.

0:08 And I started using AI about two years ago. And in that time, I had an AI consultancy that we scaled to over $100,000 a month. And then I exited that business. And now across my different

0:16 business units, we generate an average of over $400,000 per month because we all use AI so well. And none of us are actually like coders or anything like that. So in this course, I'm going to

0:24 show you guys how to install Codex. I'm going to talk about the core concepts. I'm going to talk about how you turn it into your AI operating system and your second brain. We're going to go over

0:32 things like skills. We're going to build some together. We're going to go over how you create deliverables that are branded and feel like it's coming from you and feel like it's coming from your

0:38 company. We'll go over images. We'll go over design websites using the browser using Codex to edit videos, using codeex to build automations. We'll even address the whole codeex versus cloud code

0:48 debate. And by the end, not only will you be a pro AI builder, but you'll also know exactly how you could go off and sell this kind of stuff to clients. So, there are going to be timestamps down

0:56 below. Feel free to skip around to what interests you. Save this video so you can come back to it for later. And now, let's not waste any time and just get straight into this one. All right, so

1:02 we're going to go ahead and get started with installation. You're going to open up Google and you're going to type in cloud code. Wait a minute. That was so 3 months ago. No, I'm just kidding. But it

1:10 is funny how Chad GBT also pops up right here. So, you're going to type in Chadbt or Codeex. I'll show you what that looks like. I will just type in codeex install. And then you see right here,

1:17 ChadBt Codeex because Codex is a product of OpenAI and Chad GBT. So, this is what it looks like. What you're going to want to do is just download this for your operating system. So right here I would

1:26 click on download for Windows and then I would go through the installer to get the codeex desktop app. Now this actually will be called the chatbt desktop app but we're going to be using

1:33 codec. Now if you do use VS code already or you use cursor or something like that and you'd rather run codec through the IDE extension or through the CLI then you certainly can do that. But I love

1:42 the desktop app. I think that they're bringing it in really really nicely with their browser use and their sites and routines and everything like that. And the rest of this course we're going to

1:50 be using the Codeex desktop app. So, if you want it to all look the same, then just go ahead and download the app real quick. Now, once you're inside here, it's going to look like this. You're

1:57 going to have two different tabs up top right away. You'll see chat and work, but then on the lefth hand side, you will switch over here from chatbt, which is chat and work to codeex. And if you

2:06 do have a chatbt subscription already, like you're paying 20 bucks a month or 100 or 200, that's the same exact subscription. So, you won't have to make a new account, but if you are currently

2:13 on a completely free account for OpenAI or for Chatbt, then I would at least get on the $20 a month plan right now. And if you start to blow through your usage, then just go ahead and upgrade. Because

2:22 the way that this works is you can see in the bottom left, you can see that I have 94% left and I'm currently on the $200 a month plan for Codex. And so if I use this up within the week, then I

2:31 would just basically have to wait till it resets. Or I could pay per usage API billing, which is just a little bit more expensive than being on the subscription. The subscription of Codeex

2:39 actually gets you about $14,000 of inference if you use the entire subscription. So, it's it's really really cheap for being on the subscription, paying $200 a month, and

2:48 I'm getting about $14,000 of usage out of it. And I know that 200 bucks a month or 100 bucks a month may seem expensive relative to subscriptions, but we're not paying for Netflix here. We're not

2:57 paying for entertainment. We're paying for productivity. We're essentially treating this as the cheapest hire you will ever make. $200 a month for the output that multiple hires would give

3:04 you if you start to use it right, which is why you're investing time in this course right here. Anyways, before we get into this, I wanted to address one question, which is okay, why would I use

3:13 codeex over something like work? Because work is kind of marketed as like for knowledge work for nontechnical people, blah blah blah. I think it's good that they're doing that, but ultimately it's

3:23 one of those things where codecs can do everything that work can do and it can do more. So, why would you not just invest the time learning codecs because it's basically the same, but then one

3:32 day you might need some of the extended functionality that Codeex has and you can just do it right there because you've already learned Codex. I'm not technical at all. I don't know how to

3:39 write code, but I've been able to do some very, very incredible things with codecs and with cloud code and tools like that, all with my natural language. And this entire course, you're going to

3:46 see it's literally just natural language. There's just a few things that you need to understand in order to get the most out of it. So, if you're just starting or if you've been using work

3:52 for a while, just trust me, go ahead and switch over to Codeex. You're going to thank yourself one day. And trust me, it's not anything more difficult or intimidating at all compared to work.

4:00 You're just getting more power. Awesome. So, now that that's out of the way, there's a lot of things that we're going to be looking at in this app. And I don't want to just run you through in a

4:08 really boring way and click on every single thing and explain what it means. I'm going to basically be doing that throughout this entire course as we get familiar with everything. So the first

4:15 thing that I actually want to do is kind of start with like a 30,000 foot view of everything that's possible in Codeex and how simple it all is when we really bring it together. So we're going to

4:22 start with me explaining to you all of the core concepts that you actually need to know in order to use Codex and in order to understand what it can do. And I'm going to explain the stuff extremely

4:30 simply with some examples. So let's hop into this next segment. Today I'm going over the 18 core Codeex concepts that you actually need to know in order to start using it right away and getting

4:39 value from it. It doesn't matter if you're not technical at all or you've never used Codex before. By the end of this video you'll understand exactly how this thing actually works so that you

4:45 can start using it right away. So let's not waste any time and just get straight into this one. All right, so I've split these 18 core concepts into four different parts and they get cooler as

4:52 we continue to go on. So part one we're going to start off here is foundations. So we're going to kick off with concept number one which is projects. Now, a lot of you guys, I'm assuming, have used

5:01 chat GBT or cloud in the web before where you can talk to an LLM and it can, you know, give you an answer. But the problem you face is that every time you're talking to it or you start a new

5:09 chat, you have to kind of familiarize it with you and your business and what you're working on that week or that quarter. So, with projects, you can basically keep everything organized into

5:17 one project so that as you keep working inside of it, it only gets smarter about knowing who you are. So, right here on the lefth hand side, you can see that I've got a few projects going on. I've

5:25 got Herk 2, I've got Hyperframes, I've got Trading Challenge, I've got AIS demo, and I typically am always just working inside of my Herk 2 project. My Herk 2 project is basically what I call

5:33 my AI operating system. And what a project is is actually very simple. It is just a collection of folders and files. So right here, this is inside of my file explorer on my computer and it's

5:41 called Herk 2. And this is exactly the project that I have open in Codeex. When I say right here, I'm inside of Codex, right? This is the file path that I'm working in, which is if I open this up,

5:51 my Herk 2 project. So what that means is this project has rules about me. It has, you know, credentials. It has all my projects. It has all my YouTube videos. This thing has so much knowledge about

6:01 me. So now when I come into the chat and I say, "Hey, can you just give me a rundown of what we've been working on this past month?" This thing basically gets familiarized with my project. And

6:10 then it looks at everything that we've been working on together. It even looks through all of my previous chats that I've had with Codeex. So here you can see that it looked back from the past

6:18 month and it said, "Here's what we've worked on. We've been improving your Herk 2 and your AISOs. We've been working on YouTube content and strategy, AI model testing and demos, apps and

6:27 product experiments, websites, branding, business operations. So, this is no longer just an AI chatbot. This is now a co-founder. This is a personal assistant. Which brings us on really

6:37 nicely into concept number two, which is something called an agents.md. And don't let that MD intimidate you here. If I go over to the right hand side and I open up my files, you'll

6:47 notice if I scroll down, I have something right here called an agents.mmd. And when I open this up, it's literally just a markdown file. That's what the MD means. And it's just

6:55 rules. Like this is a very simple thing that you all can read and it's not technical at all. So this is basically like the rules for your project. So you remember inside of my Herk 2 project, I

7:04 have this agent on MD. Well, if I was inside of my hyperframes editor project, there would be a different agents.md. So this thing basically just sets the ground rules for the project that you're

7:12 working in. So right here you can see that this one says, "Hey, you are Nate Herk's AI operating system. Your job is to help him spend less time on operations so he can focus on making

7:19 YouTube videos." And then I give it things like core operating rules for how to work with me. I give it some stuff about how to be safe on my desktop. And then the real bulk of my agent MD is

7:28 what I call routing. So this is a routing map. It basically means Okay, so yeah, Nate's project is huge. There are hundreds, probably thousands of files and folders inside of this project. So

7:37 let me just explain to you how this actually works. If you need business stuff, you go to the wiki. If you need corporate structure, you go to the corporate structure section inside the

7:45 wiki. If you need his voice, you go here. If you need course knowledge, you go here. If you want active projects, you go here. So everything that I'm doing, we're routing it back somewhere

7:53 so that my agents.mmd file tells this project where everything is so that every time I open up a new chat, like I said, I don't have to reexlain things. And basically the way that this works,

8:01 if we picture this being a chat thread, you know, you come over here and you say, "Hey, you say, "Hey, Codex, I need you to help me do this." And before Codex actually even like reads this

8:10 message, what it does is it first reads the agents.mmd. So I'm just going to put a here and that stands for agents.mmd. But codeex will basically read this and then it will read your message and then

8:20 it will respond with something like, you know, hey, how can I help you today? So the kind of stuff that you want to put inside your agents.mmd is the type of stuff that you want codeex to know about

8:28 you every single time you're talking to it in that project. And now moving on to concept number three, we have something called the agent loop. So you'll hear about something like codecs or claude

8:38 code or Hermes agent. These are all called agent harnesses. And these harnesses have their own agent loops. And real quick, if you're wondering about codecs versus like work chatbt

8:48 work, I always use codecs. The majority of my work is not building products or building software, but I only use codecs because it's just the most powerful. And once you get through this video, you'll

8:57 realize how easy it is to use. But anyways, the reason why this is called a harness is because they have this thing called the agentic loop, which at the highest level basically just means when

9:05 you ask the agent a question, it has a bunch of tools at its disposal and it basically thinks and it reasons and it uses tools and then it responds to you and then it thinks and then it reasons

9:15 and it takes action. It uses tools, responds to you and that's basically the agent loop. And what's really awesome about that is you can see it in action. So right here you can see that this said

9:23 worked for 25 seconds. And if I open this up, we can basically see what it did, right? Right? So like it hid this response and it gave us our output, but it had a reasoning loop inside. It first

9:32 said, "Okay, I'll check our recent tasks and project nodes, then I'll pull together the main themes from the past month." And when it did that, it ran all these commands. It ran this, which you

9:41 don't have to know what this means, but basically what this did is it looked inside something right here called a memory.md. So very similar to the agents.md, but this is about memories

9:49 rather than just like rules. It then looked at other chats. And so we can basically watch it do things. So, to show you guys a real example of this, hey, could you just check my most recent

9:58 YouTube video that I uploaded and give me three comments that you thought were funny from that video. Now, as I shoot this off, what I want you guys to pay attention to is what

10:07 we're watching. So, first it said thinking. Now, it says, "I'll find your latest upload, read through the comments, and pick three that made me laugh." And now we can see what it's

10:14 doing. So, it's looking through the memory. It's looking at these things. It's running these commands. And these commands are basically right here. It's reading, you know, files to get

10:23 established and to get oriented with what it needs to do. And so, you can actually learn a lot about how codeex works and how to best work with it by when you ask it a question, you just

10:32 kind of watch what it does. So, you can see what it had to do was it had to run a command in order to actually go talk to YouTube, look at the data, and pull it back. And now we can see three

10:42 comments. We have this one, never seen anyone so excited to lose money. Trading with AI is very easy. I just prompt Claude only take profitable trades. And if I want to lose money, I can do it

10:50 myself faster and better. And what you'll notice is how quick that all happened. Now, I do have my Astra on fast mode, but I will talk about that later because that is a core concept

10:58 that's coming up. By the way, guys, I've got this completely free SOP for you about getting your first Am. It's going to go over the exact steps that has been proven for hundreds of our AIS Plus

11:07 members to get their first paid gigs. It goes over the one sentence service pitch that can get you started today. Why your first client should cost you money. The five-minute video that answers, can this

11:15 person actually deliver before you've actually received any money, what to do when you have zero case studies. There's so many good things in here that are going to help you out. Even if you

11:23 already do have clients, I would recommend grabbing this because like I said, it's yours completely free. So, if you want to grab this, there's a link for it down in the description. Let's

11:30 get back to the video. All right, and moving on to the last concept of part one. It's a slash goal. And this is honestly one of my favorite things about AI ever. It's the ability to do

11:41 something called a slashgoal. And right here you can see that that says set a goal to keep pursuing. Now AI agents are super goal oriented which is super cool. So let's say you have Mr. AI agent right

11:51 here. And you basically say hey here is your goal. You know this is what you need to do and you give it this objective goal. What happens is it will basically just keep working and keep

12:00 working and keep working until it actually hits that goal. And it's cool because even if you kind of like give them a roadblock, they are going to be so goal-oriented that they're going to

12:09 keep retrying different things and they're going to just basically keep going until they're able to succeed and hit that goal for you. Now, obviously the more objective a goal, the easier it

12:18 is to prove that. So, if you say, "Hey, can you, you know, work on this until you've pulled in 257 sources and then write the report, that's an objective goal, right? 257 sources and written

12:29 report." But you can also set goals that are a little bit more emotional. Like your goal could maybe be more like, you know, until you're fully satisfied or until you've, you know, verified this

12:38 over and over and you feel good about it. And in that case, you're still going to, you know, get there and it will still decide when it's done. But as you get more objective with your goal, the

12:48 goal prompts are just more effective. But let me show you one goal that I actually did that you guys have already been witnessing in this specific example is I went to my Hyperframes project. And

12:57 hyperframes basically lets your coding agents build HTML and then render it as video. So essentially it creates video. But what I did here is I shot it off a/goal prompt. You can see this one says

13:06 sent as goal. So basically what I did is I said I need you to help me create some motion graphics for a YouTube video. So all the motion graphics you've seen so far were from this exact prompt right

13:15 here. I'm going over 18 different Codex concepts. I want, you know, an intro. I want transition cards. I want all this. So I basically defined that I wanted one intro scene and that I wanted 18 cards.

13:24 And then you can see that it said I finished and I visually checked. meaning it actually took screenshots to make sure everything looked good before it decided to be done. And now we have all

13:31 these animations. But what's cool about the goal prompts is that models nowadays are getting so good, the AI models are getting so good that it's actually better to give them a goal and then step

13:41 out of their way. If you really just try to like micromanage them, you're kind of just like muzzling their capabilities. So if the situation isn't super super risky, give them a goal and you can get

13:49 out of the way. Now, there's one quick thing that I felt the need to address, which is basically the question, "But Nate, what if we've been using cloud code for a long time?" Well, I'm really

13:57 glad that you asked. It's actually really simple. Really, the main difference is the agents.mmd, thecodex, and the aagents. Because in cloud code, you basically have like your cloud. MD,

14:06 and you have your cloud. So, when I first started switching over from claude code to codex, all I did was I said, "Hey, codeex, take a look at this herku project. I've been building this up

14:15 using cloud code for the past couple months. I need you to help me get this codeex ready." And basically what that means is I made a copy of my cloud.mmd and called it agents.mmd. So I have two

14:24 of those now. One one aent.mmd one claude but they're basically the same file. And then I made a copy of all my cloud skills and put them in the agents folder and I made a copy of all of my

14:32 like claude settings files and I put them in my agents. And the cool thing is you don't have to do any of that manually. You literally just say hey codeex look at the documentation analyze

14:40 my project and make this codeex accessible. Make it codeex ready. So anyways, if you were feeling a little bit of doubt about all that kind of stuff, that's the way I did it. And now

14:49 you have a bunch of local files and folders that you can use on any agent harness whenever. It's very flexible. It's very tool agnostic, and that's what you want to be building at the end of

14:56 the day. So anyways, had to address that real quick. Let's get back to the video. All right, awesome. So let's move on to part two of these core concepts, which is environments. So the first piece of

15:05 environments is number five, which is local versus cloud. So there's a couple things to talk about here. The first thing is if someone says, "Oh, is this running locally or are these local

15:15 files?" All that means is does that exist only on the machine that you're currently using? Whether you're on a MacBook right now or you're on a desktop PC like I am right now. When you say

15:23 local, it just means it's only accessible by you. So right here, if I open up my files, these are basically local files. Now, they're syncing with one drive when I want them to. So that

15:31 would kind of push them into the cloud, but otherwise it's completely local. So like if you're working on a word doc on your computer and you forgot to save it and then email it to yourself on so you

15:41 could work on it on a different laptop that was a problem because it was locally on that other machine. But something like Google Drive is always in the cloud obviously. But this also goes

15:48 beyond just files. This goes to like things that are produced. So right here you can see that it actually created me this little site for all these motion graphics. So if I open up this site you

15:57 can see that I can view all of these videos. I can click into the different ones and it threw this together for me really nice and it served this to me on a local host. So right here you can see

16:06 that this URL is 127.0.0.1 and the port of this local host is 8008. Now realistically that's not super important. You don't really need to understand what exactly this means. But

16:17 this is just a local host. But this is a little confusing because it looks like this is a web address that if I, you know, emailed to you, you could open this up and have the site. But what

16:25 would happen is this would show on your screen or it would show on your computer as nothing actually lives here. So like if I copied this and I pasted this into a new browser but I changed the number

16:34 to 7. I don't think I have anything running on port 8087 on my local machine even though you might. So nothing actually loads up. And it's quite funny. I remember seeing

16:44 these these tweets where I think it was a joke but it was like uh someone was pretending they're a beginner and they were like wow cloud code and codeex it's so cool. Look what I built in one day.

16:53 and then they, you know, attached a bunch of local host addresses and obviously no one else could open that up because they're not on that local device. So anyways, that is the

17:01 difference between local and cloud. Now the other piece here is that when you're spinning up these new chats, you can kind of have these be either local, meaning right here, you know, I'm

17:09 working in my local files and folders. And that's great because codeex can actually like edit these files, right? It can delete things, it can make new ones, it can edit them, it can move them

17:18 around, it can go to my downloads, it can go to my desktop, it can do anything on my computer really. or you could actually have these work in the cloud. And what you'll notice here is when you

17:25 choose cloud, you no longer have the ability down here to actually choose the model you were using and the permissions because this sets up sort of like a cloud sandbox environment. Now, I'll be

17:35 honest, I hardly ever do this because when I'm working locally, as long as I keep my machine on, as long as I keep my PC on, it feels pretty cloud-like. And here's what I mean by that. Right here,

17:45 you can see that I'm mirroring my phone and I'm in the Chad GBT app. And if I go over here and I click on this right here which says remote, this lets me see all of my actual codeex chats that are right

17:56 here. So if I go to Herk 2 and I go to summarize past month, this is exactly what we were just looking at right here. You can see this is literally the same conversation. It has those three chats

18:05 and I could come in here and I could say hi and this gets sent in my actual codeex right here on my laptop or sorry on my desktop as well as my phone. So, as long as my machine's on, even though

18:14 this is running locally and I'm out on a walk, I'm out at dinner. I could even be in a different state and I could control my desktop from my phone right here. So, let's say for some reason I left a file

18:24 locally on my computer, I could just say, "Hey, can you email that file to me?" And then I could use it on my laptop. So, anyways, just wanted to show off that remote capability. And that's

18:32 going to bring us into concept number six, which is work trees. Now, honestly, this is something where if you're not really building software or, you know, production apps, this isn't going to

18:41 matter to you too much. Like I don't hardly ever use work trees, but it is good to know about since you will see these in the interface. So right here you can see that if I was to go to um

18:51 the local setting, it says that you can work in a new local work tree which says create a copy of Herk 2 to work in parallel. So basically because I'm working in Herk 2, if I was going to do

19:00 something like risky or if I wanted to try and experiment with something or for example, if I was building software and I wanted to try to change the way that the onboarding flow worked, but I wanted

19:08 to like not interrupt the main branch, especially if other people on my team are working on that main branch, I could use a work tree to basically create a copy and I could test in there. And then

19:18 what you can do later is you can merge it back into the main. So it's basically just a duplicate that lets you test. But like I said, there's hardly any times for knowledge work that I ever use work

19:26 trees. And if I ever am building software and stuff, a lot of times Codex will say, "Hey, by the way, you might want to use this. You might want to do this in a new work tree." So, I wanted

19:32 to call it out because it's important because you might see things like this master branch and you might see other work trees that might have been spun up automatically. But I will say it's not

19:39 something that I'm actively thinking about or think that you need to be actively thinking about in order to get the most out of codeex. So, wanted to bring that up. But let's move on to

19:46 number seven, which is codeex. Now, this is really interesting. We talked about how when you're inside of your project, you have things like an agent.mmd. Now, if I go to my files, there's also

19:56 something in here called a codeex. So, if I open this up, what you'll see in here is I have some agents and I have some workflows, but I also have a config. ML file, which is basically like

20:07 some local settings. And there are basically going to be two different places where a codeex file lives. And that's either going to be inside of your project. So like in this example, this

20:16 is my codeex inside of my Herk 2. But there's also going to be one which is user or global, which means this lives under your actual user. And these are settings that hold things like memories

20:27 and sessions and automations. These are things that will apply to your global codec. So whenever you're in Codex, regardless of what your project you're actually in, the docex will have

20:37 personal settings and app data and stuff like that. Now once again, I think this is something it's important to understand where that lives and what it does. But this isn't a folder that I'm

20:44 actively thinking about and actively having to like maintain. But I did want to call that out because sometimes if you need to have a certain setting applied or something like that, that is

20:53 where it lives. Now, the cool thing about this is if you're ever confused, you say, "Hey Codeex, what did you do there?" or "Do we need to add this to my codeex or do we need to move this to a

21:01 different folder?" You can just ask it questions because it can search through its own documentation and understand how to do things like correctly. But I do think that understanding the difference

21:09 between project scope things and user scope things that's definitely important especially as we get into the agents which is going to come up later. And that is a great segue into part three of

21:18 the core concepts which is control and customization. So number eight here is about different AI models. So when you are starting a new chat you can choose the model you want to use. You can also

21:30 change the model in between messages. So, I could shoot off a message on Astra and then I could, you know, change this to something else and then I could shoot off another message and I can keep

21:37 changing if I want. Now, if you're watching this video way later, maybe the model names look different, but essentially the theory is the same. You have different models to choose from and

21:45 they all have different prices and strengths. So, like right now, GBD6 Astra is just the absolute strongest, but it's going to cost the most. It's going to eat the most of your weekly

21:53 limit. Right here, you can see that I'm at 100% because it just got reset. But as you chat with Astro, it will drain your weekly usage faster than you know 5.6 Soul would, which would drain your

22:03 usage limit faster than 5.6 Tarot would. Now, right here, I just want you to look at the API pricing so I can explain to you how these models are actually billing you. So, they build by tokens,

22:13 and they usually build by 1 million tokens. Meaning, on Astra, for every 1 million input tokens, it will cost you $10. And for every 1 million output tokens, it will cost you 50 bucks. And

22:23 here you can see the different pricing for these different models. And you can see how it gets cheaper as you go down. Now, this is API pricing, which means if you were building an automation that was

22:32 programmatically using these models, that's how much it would cost you. But this usage inside of Codex, that's not programmatic. This is usage limit or subscription limit. So, when you pay 20

22:42 or 100 or 200 bucks a month for Codeex, this is what you're actually your usage limit is eating up is that subscription. And you're getting way more inference. Like, this is a $200 a month

22:50 subscription. And if I use all of my weekly usage every single week, I'm getting about $14,000 worth of inference every month and I'm only paying 200 bucks for that. So the subscriptions are

23:02 much cheaper than if you were using the API pricing. But if you ever do go over your weekly limit, you'd have to buy extra codeex credits. I don't exactly know why they price it like this, but

23:10 you know, you can see they just give you codeex credits instead. So let's say I went over my weekly limit. I could go here and I could click on buy credits. This would bring up this screen which

23:19 lets you buy credits by like the thousands or something like that. And that is what I wanted to show you guys. So like 2500 credits, 5,000 credits, 25,000 credits. And this is how it would

23:29 charge you based on those tokens. Now, if you've never heard of tokens before, I know that might be a little confusing. Think about it as roughly four characters or roughly 3/4s of a word.

23:38 And there's an OpenAI tokenizer right here where if I paste in some text, it will show you how many tokens this would be. So this paragraph, for example, was about 240 characters or sorry 274

23:48 characters and about 56 tokens. And it's interesting because sometimes punctuation is a token, right? Like this period is a token, this comma is a token, this word's a token, um this

23:57 word's a token. And this is basically just kind of breaking down what it might look like for these tokens. But anyways, you can see this was 56. And what else you might notice is that I earlier said

24:06 input tokens or output tokens. Input tokens are whatever you're feeding in. So your prompts or if it's reading like a PDF or a word doc those are input because they're going into the model and

24:15 output tokens are anything that it outputs. So even its reasoning here is output tokens. This entire thread is output tokens and that's why the output tokens are more expensive. Okay. So

24:25 let's move on to number nine which is about effort. So within each model not only do you choose the model but you can also choose the effort level. So right here I can move Astro down to medium or

24:34 I can move it down to low or I can move it up to extra high or if I go like ultra I have to allow full access because that can get you know pretty autonomous. But anyways the point being

24:44 as you move up on these effort levels you get more quality like you get more intelligence but you also get more cost. it's more expensive to run it at higher effort levels. And that's why when you

24:55 see these benchmarks, when there are new models released, you'll see the model, like for example, Cloud Fable 5, you can see this mark was on low and this one was on medium and this one was on high.

25:05 So, it shows how one model can behave differently and also cost a different amount based on the effort level that it was being used and tested at. Now, it actually is pretty nice to play with

25:14 these because sometimes Astra High is just so much more powerful than what you need to do. Like if you were using GBD6 Astra High to help you write an email, that's overkill and you're paying way

25:23 more than you should realistically to write an email. I could probably come down here and use 56 Terra and be just fine with that and it's a lot cheaper. So definitely try to play with the model

25:30 and play with the effort based on your task. But a lot of times if you're just doing a lot of knowledge work, I would just chuck it on sole and I'd maybe just go to medium and just call it a day

25:38 there. Now the other thing about effort is you can do something called fast mode. This obviously makes the model respond quicker and do things quicker. And it it's 1.5 speed, but it costs you

25:45 more usage, so it will eat your weekly subscription faster as well. All right, and moving on to number 10, we have permissions. Basically, right down here, you can see that this is orange and it

25:54 says full access. This is basically Codeex's yolo mode, which means that it's not going to stop and ask you a bunch of permissions and questions. Can I do this? Can I do this? If you go on

26:03 ask for approval, it's going to always ask to edit external files and always ask to use the internet. So that's if you want to sit here and really sort of like manage it and you're wanting to

26:11 make sure everything's safe, you know, maybe start off like this so you can just understand, oh, okay, it's asking me if it can run this command or it's asking me if it can open up my browser

26:18 and then you just get familiar with what it looks like for Codex to actually run that agentic loop. You could also go on approveof for me. So it's only going to ask you if actions are detected as

26:27 unsafe, like maybe deletes or maybe certain sorts of like API calls or things like that. So that's how these permissions work. And if you really want to get granular about, oh, I want to

26:35 like make sure these certain actions never happen, you can ask Codex about that and it can work things into those config files like we talked about earlier with the codeex that it can

26:43 certainly like block things out. But now let's move on to number 11, which I think is my favorite one, which is skills. Skills are so so important. They're basically reusable workflows.

26:52 They're recipes that let you do something once and then teach Codex how to do it the same way every single time. So now you can just run all these skills. So like I said, a skill is

27:01 basically just a recipe. So, let's say you come in here and you make, you know, a chocolate chip pancake and you followed a recipe to make that pancake. You would basically say, "Okay, if I

27:10 ever want to make chocolate chip pancakes again, I'm going to use this recipe because I know that I'm going to get a good result." But let's say that you burnt these pancakes and you said,

27:17 "Okay, well, what I need to do is make a quick update in this recipe to say, hey, you know, cook it up 30 seconds less on each side." And the next time you run the skill, you basically see, okay, are

27:25 these better? Are these good? Do I need to make any more feedback in here? Or now, is this skill just the way I like it? And then your agents are able to use these skills. So here's an example where

27:34 I basically gave it a YouTube video. So you can see that this is literally just a local file path of a YouTube video. And I said that I needed a video description, timestamps, and a LinkedIn

27:41 post. It went ahead and it ran its agentic loop and it gave me a description, timestamps, and a LinkedIn post. And then I said, "Okay, now I need you to turn this into an X article." And

27:49 put the draft into my X as a draft. And it knew exactly how to do this with a thumbnail, with a title. It also put screenshots from my actual video and it sprinkled them throughout my whole X

27:59 article. And this is because it used a bunch of skills to be able to know how to do this. So right here, if I ask it, what skills did you use? It used my YouTube description skill, my timestamp

28:07 skill, my LinkedIn post skill, my X article skill, my X article from video skill, my format X article skill. It used all of these skills because I've taught it how to do something good once.

28:15 And then every time I run these skills, I just give feedback. Hey, that was good. You know, I liked this, but I didn't like this. Update the skill. And now what's cool is you can just invoke

28:23 these either with natural language. It will be smart enough to do it with natural language, or you can do it as a slash command. So, I have one in here that I can call like the audit. And you

28:30 can see this will run my audit or I have one in here called the 3D brain. And this runs the 3D brain. Man, I think I planned this out so good. This is going to bring us on to number 12, which is

28:40 agents. So, we talked about codeex. And agents is basically the exact same. It just holds different things. So, right here, you can see that the agents holds your skills. And these once again can

28:50 either be project level skills or they can be user scope skills. So, they can be global skills. So, if I come into this chat right here, which is inside of my um you know, Herk 2 project, and I go

29:00 to open up the files, and I go to my agents folder, which is right here, we have a folder called skills, and this is a bunch of skills, as you can see. Now, if I open up an actual skill, so for

29:11 example, let me go to my YouTube description skill. This is an actual skill.md. So, once again, MD just means markdown file. So, what we have here is some metadata, and this is what Codeex

29:20 will read to understand, okay, do I need to invoke this skill or not? So, it has the name and it has basically when to use the skill. And then all the skill is is just instructions. It's, hey, for

29:29 Nate's YouTube descriptions, keep them short, two to four sentences, paragraph form, no bullet points, no m dashes. Here's the process. You read the video outline. You identify the core topic.

29:38 You write a short paragraph. And that's the skill. It's a very, very simple prompt. But if I go to my ex article one, they can also be way more in-depth. So, this one is way longer. There's a

29:46 lot of other rules and a lot of other criteria. There's way more steps. There's way more things to do. And also what's cool is skills can reference other things. Skills can reference other

29:55 skills. They can reference other agents. They can reference Python files or context files. So in this example, this skill references my voice guidelines like my LinkedIn style guide as well as

30:04 my ex article style guide. But that is where your skills live inside of the agents folder. And then moving on to number 13. Here we have plugins. And this is where a lot of the magic

30:13 actually happens because you can super easily come over here in the Codex desktop app. You can go to plugins and you can connect a bunch of things. You can see all the stuff that I have

30:21 installed like Alpaca, Gmail, ClickUp, Clay, Google Drive, GitHub, Canva, and the list goes on and on. You can also search for different plugins. So, let's say you were really interested in

30:30 connecting something like, I don't know, let's see, do they have a LinkedIn plugin? They have a LinkedIn plugin where we can find the right profession, we can grow our business with ads. We

30:38 have all these other sorts of plugins, too. So, if you want to try to connect your different apps, just come in here and see if you can connect them using these plugins. Maybe they have

30:46 Composeio, they don't have Composeio, and if they don't have a plugin here, you can still connect it using Aenv using an API key. But plugins are so much easier because you can just

30:54 basically sign in once. They've got ones for creativity, for developer tools, for business and operations, for data analytics, communication, and they're basically always growing this plug-in

31:03 library. All right, so that now brings us on to part four, which is tools and scale. Okay, we've got five left here. Number 14, we have the browser. Now, this is definitely one of my favorite

31:14 things about Codeex. If I open up the tab over here, we can open up a browser, which is really cool because I can basically just control this. So, like I could go to school.com. And what you'll

31:23 notice is when I log into school.com over here, I'm just going to zoom out a little bit. You can see right here that I'm already logged into my school account. So, I could manage my different

31:30 communities, which also means that Codeex can help me manage my communities from here. Because when you save these login inside of the Codeex browser, it then saves them. So, if you say, "Hey,

31:38 could you go to that account and pull the report for me?" or hey could you go into school and could you make this post for me it can do it through the browser if there's not a plugin or if there's

31:47 not like an API or MCP server and the browser use of codeex and specifically with GBT6 Astra is the best that I've ever used ever I'll play a clip right here where I asked it to basically open

31:57 up Canva and paint me and draw me using the tools in there and I gave it just a picture of me and it was able to replicate it in a way that I thought was really really good. So, silly example,

32:06 but it definitely demonstrates the vision and the browser use capability of this agent. All right. So, number 15, we have sites. So, if you go over here to explore and you click on sites, this is

32:15 very, very cool because it basically lets you put things out there onto the cloud. Remember earlier we talked about local versus cloud. Codex can build something for you and it can be a local

32:25 host, but then you say, "Hey, can you just put that on a site real quick so that you could have your team log in or so that you could share it with, you know, other people and make it publicly

32:31 accessible." Like right here, you can see this London and Paris one is everyone whoever has this URL could go ahead and open this up. Whereas this one is just me. So it would literally have

32:40 to be logged into my account in order to be able to open up this site. And what else is amazing about that is you can open up the analytics for this site. So you can basically host things. You can

32:49 connect your own domain and this is basically going to replace something like a Versel or like a I don't know a Squarespace, wherever you wanted to host your sites normally. You can now just do

32:57 it right here from the Codex desktop app which is pretty cool. You can also even connect a database. So on the back end if you had to store like user login data or you know permissions or settings or

33:07 you know a database of customer records whatever it is you could actually store that here too and it's really cool. All you need to do to actually use a site is you just basically have to do either say

33:16 hey can you turn that into a site or you could do slash site and then it will basically know that you can you know host a site or build a site. So great touch here from codeex. All right so

33:24 number 16 we have sub aents. So, sub agents are really cool because it lets you delegate work out to a bunch of different Codex agents. So, right here in this main chat, you can see that I'm

33:34 talking to GBT6 Astra, which is a really, really smart and intelligent, but also expensive model. So, let's say I wanted to do a bunch of research, but I didn't want to waste Astra's brain on

33:43 that. Or let's say I wanted to do like a bunch of different testing and just do a lot of stuff at once. Astra here that I'm talking to can delegate workout to tons of sub aents. Hey, I need you to

33:52 search through X and search through YouTube and also dig through my own comments. So, just find, you know, what are people talking about right now in the AI space. Is there any big news or

34:00 drama? I don't want you to do this research though. I want you to basically just delegate a bunch of sub agents out to do the work and I want all of these sub agents to be using 5.6 Terra as the

34:09 model. So, shoot off those agents and let me know what they find. So, this is really cool because I can specify that I want those sub agents, those little like researchers and workers to use different

34:19 models. So right here it's using its agentic loop and it's going to spin up these agents. And whenever you see these little colorful things, these are different agents that it's spun up. So

34:27 right here, this one's called X. And when I open up X, it is actually this sub agent working right here. I can watch this sub agent go through its agentic loop and I can watch it like

34:37 pull in data and everything like that. So this is the same thing as over here, but this one is just a different model and it's not the main session that you're currently talking to. You can see

34:45 we have two others that started working. We have this one called YouTube Pulse. And once again, it's going through the same exact agentic loop. We have this one called X drama. We have this one

34:52 called audience comments. And we can basically watch all of these work, but we don't have to manage them because once these agents are done, the main session here will say, "Okay, cool. All

35:01 five are done. They reported back to me. Here's what they found." And this is really good for gathering different perspectives or researching out a bunch of things in parallel. And these

35:09 intelligent models are really, really good at delegation. So sub agents are definitely something that you need to be using. So, while these sub agents are fanned out and they're working, let's

35:17 move on to number 17, which is scheduled tasks. Scheduled tasks are awesome because this basically lets you have your codeex working autonomously for you. So, on the lefth hand side, if you

35:27 click on scheduled, you can basically ask codecs to schedule things for you, reminders, updates, or run full sessions. Because all a scheduled task basically is is it injects a prompt into

35:37 a codec session like this. So, it uses your same local files, it uses your same skills, it uses your model, it uses everything. thing the agentic loop it just does it automatically for you. So

35:47 in here I you see that I could create one and I could set it up manually. I could have it run locally or I have could have it run on the cloud. I could also choose what project to work in. So

35:55 Herk 2 or you know Hyperframes editor. I can also choose what chat. So every one of these routines could be a new chat or it could be a new chat for this task or it could be even in an existing thread

36:05 that we already have going. And then you can basically choose when to schedule it. So, it could be daily, it could be hourly, it could be, you know, weekly, or you could get it really, really

36:14 custom in here. And the cool thing is you can describe everything in here manually, or you could basically just ask Codex to make one for you. So, for example, I have these like seven or

36:23 eight different ones that are running for um trading. They're literally trading $10,000 of my real money. And I have this set up as scheduled tasks so that they can basically all work

36:32 together, talk to each other, check the market, make trades, things like that. And of course, you guessed it, these run in a different project right here called trading challenge. And they all run

36:40 inside of this challenge thread. You can tell that this has scheduled tasks in it because there's a little clock icon next to that thread. And moving on to the very last concept we have for today,

36:49 number 18, this is voice mode. Now, everything that you've learned so far, skills, sub aents, projects, agents, all of this can be controlled by using your voice to have a real conversation with

37:01 Codeex. With your voice, you can manage different codec threads. You can delegate different agents. You can run skills. You can set up schedule tasks. You can build sites. You can do anything

37:09 here with your voice. And all you have to do is come down to the bottom left and click on start new voice chat. So, forgive me here as I'm awkwardly like hunching down because my headphones

37:17 don't reach far enough. But watch this. Hey, Codex. So, I need you to help me out real quick inside of my Herk 2 project. Can you start a new thread to make a thumbnail? I'm making this

37:26 YouTube video about 18 CEX concepts. So, if you could make me a thumbnail for my YouTube channel where you see me sitting at a desk working on a laptop and I'm pointing to the right and there's like

37:36 18 different little AI agents. Um, yeah, just start start with that. And then once you have finished that thumbnail, could you send it to the thread that is currently working inside of my her 2 as

37:47 well and it is doing research on YouTube and X and um looking at my channel and stuff like that. Could you just send that thumbnail to that thread? >> All right, let me check on that. Okay,

37:56 I've got your thumbnail brief queued up with you with the laptop pointing toward the 18s using your existing photos for reference. And that found your research thread. So, I'll send the finished

38:05 thumbnail there. Now, do you guys see how this is all coming together? It started a work tree chat. It was doing the agentic loop. It also created a goal down here. It literally created its own

38:14 goal because it knew that we had this, you know, task for it and it started working. And you know, the voice chat's still going. So, I don't know what it's going to think of this, but I just

38:22 wanted to call that out because all of what we just talked about is all coming together because even you can see over here, it started making that actual prompt. Like, it shot off this prompt

38:31 right here and it started off this new thread. And now this is working in a work tree and it's going to create that thumbnail for us and then it's going to send it over into this actual chat and

38:41 we'll see that actually happen. But as you keep navigating, like you can go to the web and you can go to, you know, other things and you can keep this voice working. So as you're just working on

38:50 your laptop, it'll show you what's going on and you can keep talking to it. And you can also do all this on your phone as well because of the whole remote thing. You can use voice mode when

38:58 you're on your phone, too. Okay, so that finished up and you can see what happens is it actually can send tasks or messages to each other, right? So what happened was this was the original

39:07 thread that it created to make the image and it made this and then it sent it to this thread and said, "Hey Nate wanted me to send this to you. Here it is." So with just your voice, I know that was a

39:16 very simple example, but you're able to coordinate a bunch of different codec threads, manage them, multitask while you're on the go, while you're flipping between different apps. Very cool. All

39:25 right, so now that you guys have learned the core concepts of codecs, we're going to start to actually get in there and play around with some of this stuff. So, what I want you guys to do here is think

39:34 about the way that we're actually going to start using codecs. Don't think of it as a tool that helps you build like AI workflows. Think of it as your new operating system. Basically, what I mean

39:43 by that is you should be reaching for codeex whenever you want to do a new task. And that's the whole mindset shift behind becoming AI native, which is actually the book I wrote right here. As

39:51 you can see, this book is all mindset oriented. It's basically just about how do you talk to AI? How do you use it? And how do you change the default to, oh, okay, I'm just going to open up

39:59 Google Chrome and open up this new tab and just do the job. the old way, the manual way that I'm used to. How do you shift to, okay, I'm going to open up codecs and I'm going to use this as my

40:07 operating system. I'm going to figure out how can I build a skill for this? How can I use the browser used to do this? How can I turn everything that I'm doing into faster processes because I've

40:16 built systems around it? And that doesn't mean we're giving up a lot of the judgment or the control. That still sits with us, the human. It just means that we're able to move faster and we're

40:23 able to do way more. And the way that that truly happens is when you build your AIOS and your second brain, meaning you're not just opening up a new chat. you know, a new chatbt tab every single

40:32 time you have to do something. You're building a system where every single time that you talk to it, it gets smarter. Every single time that you build a deliverable, it saves it and it

40:40 remembers it. You're essentially turning AI from, you know, what we used to call like a personal assistant or like an executive assistant. Not anymore. We're turning this into a co-founder. We're

40:47 turning this into like a business partner that actually can work with you and keep up with you and remind you of things rather than like forgetting things or hallucinating things. So,

40:55 we're going to jump into this next section and I'm going to teach you guys how to build your own AI operating system. And just a quick heads up, guys, throughout this course, I'm going to be

41:01 playing different clips and different little segments that I might have recorded at a different time of day or, you know, last week or something. So, if I look a little bit different in some of

41:08 these, that's why. But I have very intentionally laid out this entire course in the order that I think makes the most sense to follow it. So, hopefully that'll make sense. But yeah,

41:15 let's get into the next piece and start building out your AIOS. GBT6 Astra is the most powerful AI model I've ever used. So, I turned it into my AI operating system and my second brain.

41:24 And this is just a quick visual, but right now what you're seeing is my actual company. This is my second brain. This is everything that I know and everything about my business. And

41:32 because I have everything actually stored and usable, that means so does GPT6 Astra. So when I say AIOS, what I mean by that is an AI operating system where now I can basically do anything I

41:42 need to from right inside of this interface with Codeex with GPT6 Astra because now it doesn't feel like I'm just talking to a chatbot. It feels like I'm talking to a co-founder, someone

41:49 that actually knows everything that's going on in my life and my business. So today I'm going to show you guys how you set up an AI operating system from scratch, but if you already have one

41:57 going, how you can keep improving it. And I'm going to give away a bunch of free skills that help you set up exactly this and help you set up other things like auditing your AIOS, leveling it up,

42:05 and just setting up the structure. Right? Now, what you'll notice here is that this isn't just a random dump of knowledge. Right here in blue, this is my business wiki. So it's like AIS

42:12 coaching, our corporate structure, our Ammation Society Plus. And when I click in, we can see all of these things in these different relationships. I can also see how stale this is, so I might

42:21 need to update it. and it calls out all these areas where like this number is old. I need to update this. But it also shows me all the different connections where this specific node is linked into

42:30 every other area of my business. So like I said, that was business wiki. I also have meetings that I've added in here, which takes up a big section. I've got video knowledge. So every single YouTube

42:38 video I've made, it has all that knowledge and it relates it back to everything in my business and in my meetings. Same thing with the clawed memory right here as you can see. Same

42:45 thing with codeex memory. I've also got a bunch of projects and then I have skills and agents and this forms together my herk brain and I've even gone as far as to take this exact brain

42:55 and put it in its own OS dashboard where I can basically have my calendar over here. Today's a Sunday so I don't have many meetings and this is fully functional, fully synced with my

43:03 calendar. We've got these three stats, YouTube subs, AIS, AIS plus and these are fully synced. I can come in here and manage my brain. I can search for different things. I can look through

43:11 different things. I also have my communication fully synced. So Slack, ClickUp and email and these are functional. I can respond to things from here. I can also see different meetings

43:17 that I have coming up as well as past meetings with the AI summaries already generated. And then I can get community pulse. So top YouTube comments, top AIS plus posts, what people are stuck on, as

43:25 well as industry pulse from things like YouTube and X. And all of this of course was made with GBT6 Astra. But really the idea of an AIO OS or an operating system is my framework that I call the four

43:35 C's. So context, connections, capabilities, and cadence. And the way that I think about this is I think about these first two things as the second brain piece. So I think of this as your

43:44 second brain and then over here we have the actual AIOS and this is the capabilities and the cadence. So let me real quick explain what I mean by that. What's the actual difference? So the way

43:53 that you get this thing to actually feel like it knows you is by giving it context. So I mean things about you and your background and your business and your goals and your priorities and

44:01 things like that. I think of context as the things that are in your life that really don't change too much. So quarterly goals or yearly goals or what your business does, your avatar pain

44:10 points, things like that. and the connections. I think of things as the tools that you use every day because the data in there is always relevant and it's always changing every day. So, your

44:18 email, different data is always in there. Your conversation, Slack, ClickUp, your project management, your financial information, giving your AIOS the ability to know all the context, but

44:27 also reach for things just in time when it needs it. That's how you build a really, really powerful second brain that helps you actually ask your AIOS questions or say something and rather

44:35 than getting a vague or generic response, it knows what to do. So, for example, when I asked Codex to make me this AIS sizzle reel, I didn't tell it everything I wanted. I just said, "Hey,

44:43 make me a sizzle reel for AI Automation Society." And it knew all this context, and it also knew that it could go to my school community and look things up in order to make this. In fact, what you'll

44:51 notice is it actually went so far as to go to the community, take pictures from it, like take screenshots itself, and add those in here because it knew that this is what I was talking about. So, I

45:00 have these three little tests that I like to run, which is basically if a teammate messages you with a question and you realize that your AIOS would answer better and faster with the exact

45:09 sources, then that's a good sign that your AIOS really is where it should be. Another one is context switching reduction. One thing that you should challenge yourself to do after you've

45:18 gotten this set up is to try to force yourself to do everything from here, from this interface. rather than opening up all those new tabs and doing it the old way. Try doing it the new way of

45:27 using the AI as your operating system. And then number three is the knowledge leaves your head. So stop trying to remember everything. You don't have to rehearse what you decided last quarter

45:36 or what your customer said in that meeting because you trust the retrieval of asking your AIOS because it can find it in your Slack threads. It can find it in your email, find it in your meetings,

45:44 or it just knows it. Because if you start to build new skills and capabilities and you start to put cadence into play, which means agents that run autonomously and automations

45:52 and things like that, these things are not nearly as valuable without the first two C's because otherwise these skills are very generic. And the automations are just like I said, generic. So these

46:01 four things really come one after the other, context, connections, capabilities, and cadence. And they all work together very beautifully. And that's how we get our AI operating

46:09 system. So in order to get started or to scale all this up and get all the resources, I need you to go to my free school community. The link for that is down in the description. Like I said,

46:16 it's completely free. And you're going to go to classroom. You're going to click on all YouTube resources. And you're going to grab the AISOS resource pack. This thing is going to come with

46:23 five skills. It'll come with an onboard skill to get you set up. It'll set up your whole project structure, your files and your folders. Then you'll have an audit skill to do frequent checks on

46:31 your AOS. You'll have a link skill, which helps you route to different files when you need to. You'll have a level up, which you can run after your audits that tell you, hey, this is probably

46:39 some stuff you should do based on this audit. And then you have the 3D brain skill which helps you turn your knowledge into this actual 3D brain like you guys saw right over here. Now, yes,

46:48 it is these four things. And when you start to run that resource pack, it'll onboard you and ask you questions and it will start to fill in this information which is great. But then what comes next

46:56 is something that's super super important and that is called the agents.md. And so if you've been using claude, this is basically the claw. MD, but now this

47:04 is the one that Codeex is more familiar with reading and it's the exact same thing. So, if you are coming over from Codeex, literally what I would do is I would just copy your claw.md and then

47:13 just name it agents.mmd. It's the exact same thing. And just to show you guys what mine looks like, if I go into here and I go to my files, you can see that it's very simple. It's you are Nate

47:21 Herki operating system. Here's your job. Agents.mmd and cloud.mmd contain the same stuff. And then I basically give it a few core operating rules like to be concise and I use bullet points and

47:30 don't use m dashes and you know use the Oxford comma. But then I go into a routing map and that's the majority of my agents.mmd. I think of it as you know core rules but then I think of it as

47:39 routing meaning can my agent understand where everything lives because if I open up my Herk 2 project which is just like we said a combination of folders and files and that's all this is displaying

47:50 is folders and files and it's showing the relationship between all of that but the point I'm trying to make is that this is a lot right there's a lot of folders and files and I can go into my

47:58 projects and there's a lot more in here and then I can go into things like you know my YouTube videos in here and it's even more but because all of this is laid out in a way that to me is

48:06 intuitive That also means that my agents can crawl it a little bit better. But the more that I have information in here like routing logic, like, hey, if you need business advice or team stuff or

48:15 OTAAS, you look in the wiki. If you need corporate structure or entities, you look in this section of the wiki. If you want to look at Nate's voice and style, you look here. If you want to look at

48:23 course knowledge, you look here. If you want to look X, go Y. If you want to look A, go B. And you can see there's a ton of different rules in here, but what's important is that the agent knows

48:30 this every single time it loads up so that when you ask for something, it knows where to look. Because that's essentially how the agents.mmd file works. You know, you say, "Hi, codeex."

48:39 And before it even will read your message, it reads this first. So, it reads this and then it reads your message and then it will respond with like, "Hey, how can I help you today?"

48:47 You know, that's kind of what it will say. But it always reads this context first, which is why it's so important. All right. So, if you're a complete beginner, how do you actually get set up

48:54 with all this? Well, let me show you. First thing you're going to want to do is you're going to want to go to somewhere on your computer. Let's just say it's your desktop. And you're going

49:01 to create a new folder. And so for example, this is where I am using my Herk 2 project. But for now, I'm just going to call this AISOS demo. So you can call this whatever you

49:12 want. Now once that folder has been created locally, which just means on your computer, you're going to open that up in Codeex. So at this point, if you don't have the Codex desktop app,

49:21 download that. And then if you aren't on a subscription, just get on subscription. But now what we're going to do is we need to open up this OS. So what that means is you would come in

49:28 here and you would click on new project. You would create a local project and then you would give it a name and then choose this folder. So this is once again going to be AIS demo and I'm going

49:37 to open up that folder. If I go to my desktop and I just have to find this folder we just made right here. Select that and then go ahead and click create project. So this now means whenever we

49:47 are doing things in here, we're working inside of this actual local folder which if I open this up, there's nothing in here right now. But you will see as we start this whole onboarding, this folder

49:56 will fill up. So, what I'm going to do is I'm going to go to the resource pack and I'm just going to copy this actual URL. And I will now paste that into the chat and say, "Hey, I want you to

50:06 install this and just go ahead and run the onboard so I can start giving you some information and building my AIOS." So, now that this is installed, it's going to start onboarding you. And you

50:14 can see that it's going to say, "I will save your answer as you go." So, obviously, you guys should take some time and give this some pretty specific information, but I'm just going to blow

50:22 through this real quick. So, my name is Nate and I run an AI education community and we have a certification program and we have live events and things like that. And what's going to happen is it's

50:31 going to start creating folders and files for you. If you click on this button and you go to files, you can see that it's already created a bunch of stuff in here. None of this was in here.

50:39 If you guys remember, this was all completely empty, but now it's starting to set up the actual structure. But really, the two most important things to call out right now are the context

50:46 folder. This is where it's going to start to create some markdown files of business context and also the agents.mmd. I remember I showed you guys mine and this is going to start getting

50:54 built out over time with you guys right in here. Then it asks you to paste some of your recent writing so that it can start to build some skills around how you write. So what I'm going to do real

51:02 quick is just basically fly through these seven questions and then you guys do this as well and then come back. All right. So now that I've done those seven questions, if I open up a new thread, I

51:09 can actually say, "Hi, who am I?" And it will check my profile first because now it can actually look at all that and understand that. So if I open up the files once again, we can now see that if

51:19 I go to my context, we have one called about business, we have one called about me, and we have one called priorities. So if I open this up, you can see that this has information. If I open this one

51:28 up, this one has information, too. And over time, these will continuously get more filled with context. And this is context, like I said earlier when I was talking about these four C's, context

51:37 that we want to basically be more evergreen, right? Information that's not changing very often. Whereas these are more the connections where you have information that changes very often. And

51:47 that's when you're going to start filling it up by looking at things like this. You're going to start looking at um what bookmarks do you have? What desktop apps do you have? What apps do

51:56 you use the most on the weekly basis? And I like to think about it as far as like revenue, customers, calendar, coms, tasks, meetings, and knowledge. And then once you've taken some time to like list

52:06 them out here, just say, "Hey, Codeex, I want to connect to school. How can I do that? I want to connect to Stripe. I want to connect to YouTube. how can I do that? And it will tell you, oh, you need

52:14 to get this API key or you need to use this MCP server or maybe you need to use browser use. And it will help you figure out the best way to connect to all these tools. But what else you'll notice is

52:21 that it's creating the things that we need in case you ever want to go over to cloud code like acloud folder as well as a cloud.mmd. So it's really helping you make this your own AIOS, your own IP

52:31 where you're not locked into one ecosystem. So anyways, I just wanted to call that out. But the agents and the cloudmd file are basically identical. Right now we have like the operating

52:39 system, we have some skills, we have where things live, but right now this is pretty empty. So obviously we need to get this filled up a little bit more. You can also see that in the references

52:47 section we have a voice. So if I go to my references and I go to the voice MD, we can see that this is stuff about how Nate talks based on the information that I've given it so far. So anyways, now

52:57 you can see we're starting to get some stuff filled up here. But now what you'll do is once you have connected some things, you're going to go ahead and run the audit. So you can do /

53:04 audit. This will automatically be installed because you use the resource guide. And this basically checks the AIOS. It looks at the four C's. It gives you a score. And it will also look at

53:13 your agents MD or your cloud NMD and make sure everything's synced up like I said. And what else is cool about this is this creates another folder in your AIOS called audits. You can see there's

53:22 nothing here called audits. But what's going to happen is it's going to give you a report and then it's going to store your audit. So every single time you run this, let's say you want to run

53:30 it every Monday or every two weeks, it will store your scores every time so that you can see how you're actually improving your system, you know, month over month. So I know if you're new to

53:39 Codeex and everything, this might feel a little bit intimidating, but it's really not too bad. And the cool thing about this is you can just ask questions if you ever get confused. Hey, why'd you do

53:46 that? Or where should I put this? Or, you know, how do you feel about all this? You can treat this thing like your best friend who's really smart, who's also a mentor rather than just like a

53:56 chatbot. Okay, but you can now see that this audit is done. And obviously, we scored a 30 out of 100 because we literally just built this and there's not much in here. But it'll show you

54:03 across those four C's what needs improvement. And you can keep working on that stuff. And over here, you can see that we have a new folder called audits. We have this markdown file which shows

54:11 us that today, September 7th, you ran this audit. And here was basically the conclusion of that audit so that you can improve on it. Now, what you can do from there is you can go ahead and run

54:20 another skill which comes in the kit, which is called a level up. So, this one will basically look at this audit report and then it will help suggest areas for you to improve it. It'll help you think

54:28 about different connections and different ways that you could, you know, build in some more capability and build in some more cadence so that this thing keeps growing and keeps growing. So,

54:35 here you can see it said, I'd start with turning team updates into one clear action list. It targets your stated bottleneck coordination, taking time away from learning AI and making videos.

54:43 So, two candidates here, which are team action list and certification improvement backlog. So, I would then think, okay, cool. So how do we actually start driving towards this? You know,

54:54 what do you need to see? What connections do we need to set up? And then how do we build these automations? And so you can just keep running this loop of auditing, leveling up, building,

55:02 auditing, leveling up, building. And this is the type of stuff that Astra is so good at because it's so so smart. It really feels like it's got some really incredible general intelligence,

55:12 especially as you give it more and more data because it's really good at building automations. It can oneshot a lot of this stuff and it will help you build all these routines. So automating

55:21 some of these little tasks is no problem. The real issue is getting everything from your head into the system because the more you give it, the more context you give it, the better it

55:30 gets obviously. So yes, it can give you these recommendations, but now what I want you guys to think about is how you can give it more data. So I do have this other skill here called Grill Me. It was

55:38 inspired by Matt PCO's Grill Me and I will actually put this in the AIS OS kit. So if you install it based on this video, you'll have this here. But I didn't do it right now. So anyways, then

55:48 you could use the grill me and say, I need you to use this grill me skill to just, you know, find out as much as you can about my current priorities in my business. And what this does is it will

55:58 basically do similar to the onboard where it asks you questions, but this is a separate skill where it just relentlessly grills you over and over. And what's cool about this is you just

56:07 keep having it grill you on different topics because every time you start one of these interviews, it will create a new file in your AISOs and it will store all this interview. So every single time

56:17 that you come in here for 20 minutes, 30 minutes, and you just get interviewed by AI, it's going to store all that. So you're just building up your second brain over time, and it's really, really

56:25 cool. As you can see, it said, I'm going to save every answer in business priorities capture. It created a new folder right here. And now we have this markdown file where as I start to answer

56:33 these questions, it's going to store everything. So that's something that I would recommend going through with business priorities, your team, you know, your goals, and just like a bunch

56:41 of different topics about your business and about your life. Now, the next thing you need to do, and it's how you actually get something more like this where you actually have like all of

56:48 these relationships forming between these different things that you want to put into your AISOs or into your second brain. This gets really important because the relationships is what makes

56:57 this actually connect to everything and what makes it feel more holistic rather than just like a data dump. So, the way that you get to this point with all these relationships is we're going to

57:05 use Carpathy's LLM wiki. This basically just has AI crawl through data sources and find those connections for you. So, it's really, really cool. And guess what? It's really easy to use because

57:14 all you have to do is you can either copy this URL, which I'll put in the description of this video. Or you could just come in here and you could copy this. Like you could literally just go

57:22 like this, copy it, and then come into your actual AISOs. You can start a new chat and then say, "Hey, based on what you currently know about me, I want you to build me a new LLM wiki, and I want

57:32 you to do so on all my business contacts and everything using this Carpathy wiki method." And then you can literally just paste in that prompt, and boom, it's going to crawl through everything. It's

57:42 going to create you a new wiki vault. And you can obviously be more specific like I have one vault for my YouTube videos. I have one vault for my business knowledge. I have one vault for my

57:50 meeting transcripts. And so I started to separate them out over time. But just to start, you could do, hey, this one's just for business or whatever you need to do. And it will help you find those

57:58 relationships. And then once you've done a bunch of grillies, once you've built out these L&M wikis, then you're going to go in here and you're going to run that 3D brain skill. And that is the one

58:07 that's going to take all your knowledge and it's going to turn it into something like this. And this obviously will continue to sync and stay optimized and continue to improve for you. And it will

58:16 probably build that out into something called a local host. But what's really cool is you could ask it to turn that into a site. As you can see, Codex basically lets you build things and

58:24 store it on their domain. So then after you build it, you could have it be accessible to different laptops or to your phone or whatever you want. Or if you wanted to keep it local, you can do

58:32 that as well. And if you also wanted to connect your own custom domain, you can do that, too. So really, the possibilities here are endless. And it's just about doing that loop, like I said,

58:39 of constantly scaling up your four C's and doing that loop of the audit level up, audit level up. Now, the last thing I wanted to address was the whole idea of GBD6 Astra driving everything, right?

58:51 Because Astra is an incredibly powerful model, but I also wanted to say so is 5.6 Soul and so is 5.6 Terra. And for the majority of this knowledge work that you're doing here in your AOS, like

59:02 asking, oh, can you find this doc or can you help me create this spreadsheet? A lot of that using Astra is probably overkill. Like Astra is one of the most it is the most capable model that I've

59:12 ever played with. So what I would recommend is build out some of the stuff. You know, get familiar with how this model works. But then when you start to try to, you know, use different

59:20 skills and when you start to try to connect to different things, I would recommend trying with 5.6 Soul because Astro is going to crawl through your usage limit, which is the weekly limit.

59:30 It's going to eat at that more because it's a more expensive model than 5.6 Soul is. And once you really start to get comfortable here, you can play with different things too, like the effort

59:38 level. You can make it go on lights, which reduces some of the capability, but it also makes it cheaper. And then you could also go to 5.6 soul, and you could play with the effort level over

59:46 here as well. There's a lot of cool things to dig into once you really start to connect a lot of stuff to your codeex system. Once again, I just think it's important to call out that all you're

59:55 building here is you're building files and folders. And what's cool about that is as you build out your files and folders, it will grow into something that's just amazing. So now I have my

1:00:04 herk 2 which like I said has everything in here. And the cool thing is if I wanted to build a Hermes agent I can connect that to this. If I go back to cloud code I connect it to this.

1:00:12 Whatever is the new tool, whatever is the new AI model, I don't have to worry about switching over. I don't have to worry about being vendor locked because I own all of this stuff right here and

1:00:21 every agent can crawl through it and use it. And that's really what we're building here. We're not just building a codec skill or a codeex hub or a little like 3D brain. We're building IP. We are

1:00:30 building a knowledge base that is going to be the most valuable thing because now we can work through it because there's so much context and knowledge in here. So I really want you guys to

1:00:38 seriously get in there and start doing a bunch of grill me and just building up your routing and your context. So remember free school community linked in the description. You'll find all the

1:00:46 resources that you need right in here. So you've got the foundation set up. Obviously like I mentioned this is never a finished product. I've been building my own AIOS for about 9 months now and

1:00:55 every day I make a change to it. I always make improvements. I always clean things up. And that will always just be the case. You're always going to keep improving that thing. But obviously, you

1:01:02 can build things along the way that help you just become way more fast, way more efficient, way more powerful. And those are called skills. So, let's get into this next section where I teach you guys

1:01:12 how to build skills better than basically anyone out there. And this is my proven skill building and skill using method. So, let's hop right in. Today, I've got this proven six-step process

1:01:21 for building codec skills better than 99% of people. So, let's not waste any time and just get straight into this one. All right. All right, so these are the six steps that I'm going to go over.

1:01:28 I'm going to explain each of these and tell you why they're so important and then we're going to get into Codex and actually try some of the stuff out. And real quick before we get started, I'm

1:01:34 going to talk about what actually is a skill. So, if you already know what a skill is, then just go ahead and skip past this part. But for those of you who need a refresher, here we go. So, I'm

1:01:41 going to use my chocolate chip pancake analogy. Let's say we have this chef and this chef makes the most amazing chocolate chip pancakes and you want to know how to make those. This is

1:01:49 essentially the output that you're looking for. The way that you would be able to copy this chef is by looking at the recipe that the chef used or the recipe that the chef made. So the chef

1:01:58 gives you the recipe, this is you. And now you're able to pretty much make the exact same output, the really, you know, popular famous chocolate chip pancakes because you followed the recipe. So this

1:02:08 recipe is basically the skill. This is essentially the skill.md file, which means a markdown file. It's just a simple language. It just means that inside of the skill file there are like

1:02:17 pound signs and asterisks to indicate like bullet points and headers and things like that. So it's it's just natural language. But then the agent is basically able to take this skill file

1:02:26 and just use it so that if you say, "Hey, Mr. AI agent, you know, make me those chocolate chip pancakes." It wouldn't have to be like, "Okay, well, let me just do some research on how to

1:02:33 make them, and I don't know exactly what kind of pancakes Nate wants. I don't know how big they should be." So what I'll do is I'll just follow this skill. I'll follow the recipe. And now I get

1:02:40 the same output that Nate's looking for and I get it consistently the same every single time because it's all documented for me right here. And these can be really simple. It can be a simple prompt

1:02:49 like, hey, you know, help me turn this email into something that's more professional or something that sounds like me and maybe that's like my Nate email skill. But it could also be

1:02:57 complicated processes like doing research on the market and analyzing, you know, 50 stocks and telling you which one to buy. So really, it's whenever you want to basically like

1:03:05 codify some sort of process that you do and so that you can delegate that process to an agent and then you turn it into a skill and now your agent can use those skills. All right, so now that

1:03:14 that is out of the way, let's start with number one up here where we have reverse engineer. So the whole idea with reverse engineering your skills is basically that you want to start with an output.

1:03:25 You want to start with what is the definition of done? What are you actually looking for? Because let's say you ask your agent here for chicken, for example, and you actually want like

1:03:35 chicken parmesan on a bed of pasta, but because you just said chicken, the agent might interpret that a little bit differently and make you a chicken sandwich. And then next time you ask for

1:03:44 chicken, it might make you, you know, chicken thighs. It doesn't actually know specifically what you want. So if you start with an output and you have, you know, essentially this chicken parmesan

1:03:52 and you say, "Okay, let's reverse engineer this food and see what went into it, how long we cooked it, how did we get here?" And that's how you build the recipe. So, for example, if you

1:04:00 wanted to build a skill for the end of the week report, you've got certain columns in your Excel sheet. You've got certain calculations that were made. It's way easier to give the agent that

1:04:09 Excel sheet to give it the final deliverable and say, "Hey, this is an output that is really good and this is what I want to build a skill for so that you understand how to take some raw

1:04:18 input and turn it into this output that I have already told you that I like and this is what we're looking for every time." And then it can basically walk you backwards through that process.

1:04:25 Okay, what data did you look at? where did you get it from? How did you calculate it? Where did you format it? You answer those questions and then you have a version of a skill that already

1:04:33 knows sort of like the north star that it's building towards. I think it's so much easier to run the process, get the output and say, "Okay, let's turn that into a skill." Rather than saying, "Hey,

1:04:41 build me a skill for building a YouTube dashboard." And then you might have this vision and your agent might have a completely different vision and then you're going to get frustrated when it

1:04:50 delivers you a chicken sandwich and you actually wanted chicken parm. Now remember, all of these concepts I'm going to, you know, bring back together when we actually hop into Codex and I

1:04:57 show you some stuff. But let's just keep moving down the list because I think this foundation is important. So number two, we have one specific job and one specific trigger. So remember how I said

1:05:05 that these skills are markdown files. Skill.md. What happens in the skill.md is you've got a YAML description. It's called YAML front matter. And then you've got the actual skill

1:05:15 instructions, which is the meat of the skill. So right here is one of the skills that I'm going to keep coming back to in this video because it's it's one of my favorite skills. And you can

1:05:23 see up here this metadata, this is the YAML front matter. So if I view the source, it looks like this. It's actually just separated by these three dashes. So this would be the YAML front

1:05:31 matter. And then everything below it would be the actual instructions of the skill. Now this is markdown. And when I said markdown is very simple. It is. You can see here are two pound signs. And

1:05:41 that just indicates a header. We can see that we've got these little dashes which are bullet points. And so when markdown is actually rendered, it looks more something like this. And you can see

1:05:49 it's just formatted better. So this is the way that I like to look at it. But if you click on view source inside of the Codex app, you can see the raw markdown file. But anyways, the metadata

1:05:58 up here is the YAML front matter and that tells us what is the name of the skill and when do you explicitly use it. And there are other fields that can be populated in here like in this one we

1:06:06 have an argument hint. And basically this X article skill is used when I ask my agent to turn an YouTube video into a long form X post. And so sometimes it'll feed in this argument which is my

1:06:18 YouTube video URL that it will take, it will download, it will transcribe it, it will take screenshots of it, all that sort of, you know, all the stuff. And then once the agent has basically said,

1:06:26 "Okay, cool. Nate wants to turn this YouTube video into an X article. So, okay, cool. I found this skill that I need to use. Now that I know this is the skill I'm going to invoke, let me read

1:06:35 the entire description of the skill. So now I understand exactly what to actually do." And you can see this skill is pretty long. There's a lot of instructions in here. And this wasn't

1:06:43 like me on one try turning this into a skill. This is probably a skill that's been iterated on 25 or more times. And we'll talk more about that in a bit. But the whole idea is that you want a skill

1:06:53 to do one very specific job. You don't want to have a skill that's like run the marketing team. You want to have a skill that breaks that marketing team, all of those processes into individual steps.

1:07:01 So in my book right up here, becoming AI native, I talk about this and I call it the function breakdown. I call it the the tree. Basically, your job is a tree and you've got different trunks with

1:07:11 different like, you know, bullet points in your job description and then each of those trunks have branches and each of those branches have ultimately leaves and you want to turn all of those little

1:07:18 leaves or those little tasks into skills because you can chain skills together over time. But the point being, if you have a skill with one very specific job, you can give it one very specific

1:07:28 trigger, which means that your agents are going to be able to execute them more consistently, invoke them automatically. It just helps you separate out what your agents are doing

1:07:36 in a much better way than if you have a skill that's like 30 pages long and it's supposed to do so many different things. So think about breaking down processes at the task level and turning each of

1:07:45 those tasks into skills. All right, moving on to number three. We have thinking about the freedom level. So what I mean by this is when you're automating things, you basically have to

1:07:55 figure out is this a deterministic automation or a non-deterministic automation? Basically, is this something that's predictable and we know exactly what comes in and we know exactly what

1:08:04 happens and we know exactly what comes out or is it more of a non-deterministic AI agent, more of a nondeterministic black box where we know sort of what the input's going to look like. We sort of

1:08:14 know what's going to happen inside the process, but we do kind of know like what the output should be. We still know the definition of good and the definition of done. You can think about

1:08:21 it like, can I automate this using rules? Hard rules, hard logic. Is x greater than 10? Is x equal to 500? or is it something that needs a lot of judgment? Analyze this. Turn this into

1:08:32 an email. Turn this into a spreadsheet. And it's really important to think about where does your skill live on that spectrum? Because if you have a very deterministic skill, like for example,

1:08:41 it just has to take a Excel sheet and it just has to populate the cells into the CRM or something and that's basically just data processing transferring. That's a very deterministic process. And

1:08:52 if you write the skill in a very nondeterministic way, there's a lot more room for error. There's a lot more room for AI to interpret something wrong and do something wrong. And that's the case

1:09:01 where you want the skill to be like step one do this, step two, do this. Exactly, exactly. Exactly. Do this. It's very specific. But if you've got something nondeterministic, like for example, my X

1:09:10 article skill, we don't want to be so specific. We want the AI to be able to use its judgment and to be able to use research and thinking to generate a unique output because every YouTube

1:09:20 video is different, which means every X article is going to be different, which means all the screenshots that it needs to take are going to be different. I couldn't say something like screenshot

1:09:26 the video at the 1 minute mark and then the two-minute mark and then the 5minute mark and put those into the article at line 400, line 600, line 700 because then the articles would come out feeling

1:09:35 generic and they probably wouldn't even make sense and it just wouldn't be a very good output. So think about the freedom level of the process and when you're doing this manually, do you

1:09:43 follow the same set of instructions every time or are you constantly using judgment and are you constantly doing different things within the process to ultimately get to that definition of

1:09:52 good? By the way, guys, I've got this completely free SOP for you about getting your first a automation client. It's going to go over the exact steps that has been proven for hundreds of our

1:10:00 AIS Plus members to get their first paid gigs. It goes over the one sentence service pitch that can get you started today, why your first client should cost you money, the 5-minute video that

1:10:08 answers, can this person actually deliver before you've actually received any money, what to do when you have zero case studies. There's so many good things in here that are going to help

1:10:16 you out. Even if you already do have clients, I would recommend grabbing this because, like I said, it's yours completely free. So, if you want to grab this, there's a link for it down in the

1:10:23 description. Let's get back to the video. All right, moving on to number four, we have verification. And this is probably the most important piece of building skills. So, the whole idea of

1:10:34 building a skill is that now you can trust that your agent's going to give you an output more consistently that meets your standard. But what you want to do in there is you don't want to be

1:10:43 constantly being handed the agents V1. You want to be handed an output that's already good to go that you can give it a quick skim, approve it all, and then shoot it off. And the cool thing about

1:10:53 agents building stuff to ultimately achieve some sort of northstar is that you can also have agents check its own work, verify its own work, or you know, other agents verify different agents

1:11:03 work. So every single skill that I build works in some sort of verification loop, meaning the agent who builds the thing, delivers this output, and then we either have the same agent or different sub

1:11:12 agents come in and verify that output and then provide feedback. And this verification loop can go on multiple times. Sometimes it's just, hey, here's the V2 and it's approved. Here's the V7.

1:11:21 I've even had agents verify over and over and give me like a V15 or 16. And that basically just ensures that your time isn't being wasted verifying and sending feedback when you can have

1:11:30 agents do that for you. Now, when it comes to verification, there is a difference between objective checks and subjective checks. So, let me explain what I mean by that. An objective check

1:11:38 is something that can literally be proven. Sort of like the non-deterministic versus deterministic thing. Objective means okay um I wanted you to do research and then the

1:11:46 verification was to pull in 500 sources cut that down to the best 250 and then every single fact that's in the article I want that to be double checked by second agent and that's objective we can

1:11:57 literally prove 500 sources cut down to 250 every single fact double checked by a different agent that's objective that is a rule does x equal y is a greater than b but then you have some things

1:12:07 that are subjective like verifying the video right if it's it's creating a video for you or if it's creating a website for you. A lot of these subjective checks are more like, oh, you

1:12:16 know, does everything look good? Is everything inbounds? Is it loading in a certain way? When you don't have an actual hard metric to align it to, you have to sort of be a little bit more

1:12:24 subjective on your verification. And what you're thinking about doing is turning that agent into an LLM as a judge, which basically means, okay, so if AI is going to be applying its

1:12:34 approval, and it has to use judgment to approve, then how can you tell the agent exactly what you're looking for? What does good typically look like or feel like even though there's not a hard

1:12:43 rule? So, for example, with the X article skill, I've got some objective checks in there, like there are 10 screenshots from the video. But then there's other subjective checks like how

1:12:51 do you make it flow in a way that makes sense? How do you make it sound like Nate? How do you make sure that the images that are put into the the article, the screenshots, are cropped?

1:13:00 There's other subjective checks that we have to let it do things where it like opens up the app and it scrolls through and it reads through and it screenshots and it verifies. But a lot of those

1:13:08 checks, like I said, are more subjective. And so that's where iterating on these skills takes a lot of time because you're constantly running the skill, analyzing the output, giving

1:13:16 feedback, running the skill again, analyzing the output, giving feedback. And you're basically just like training the skill where every single time you use it, it gets better. But the thing

1:13:23 about verification is there's always a way. You might be thinking to yourself, well, I don't really know how I would have agents verify that process. Agents can do anything. They can use your

1:13:31 computer. They can use your browser. They can look at things. They can listen to things. They can do anything. So just think about it like this. If you assigned that task to a human, what

1:13:38 would you do to basically give it the stamp of approval? And then just explain that to the agent and that's your verification. That doesn't mean it's going to always come out 100%, but it's

1:13:47 going to be more like 95% rather than giving you something that started off at like 75 or 80. Okay. And moving on to number five, we've got walk it down. And when I say walk it down, I mean walking

1:13:56 it down the model list. So whether you're using Cloud Code or Codeex, we all know that different models have different strengths. And we all know that models that have the strongest

1:14:05 strengths are the most expensive. So for example, Astra is more expensive than Soul, but more capable than Soul. Soul is more expensive than Terra, but more capable than Terra. So the idea is you

1:14:14 build a skill and you're getting this good output. And maybe you were testing that with Astra. Okay, well let's run that same skill on Soul. Are we getting the same result? If we're getting the

1:14:22 same result and it's cheaper, okay, let's test it on Terra now. Are we getting the same result? Is it cheaper? Okay, let's test it on Luna. There's no reason to be running a skill with Astra

1:14:30 if you could run it with Luna and get the exact same output. But it is important to test it because not every skill fits that. I've had some skills where I run them on Haiku or Luna

1:14:39 because they're just super simple. For example, when I give my agent a YouTube video and I say, "Hey, I need a YouTube video description and timestamps that I can do with Luna, and doing it with

1:14:47 Astra is overkill. But for this ex article thing, I typically like to use at least Soul or Astra because it just has a better understanding of screenshotting things and putting it in

1:14:56 the right spot and cropping it down and even blurring things out. It's so much better with the browser use with Soul and Astra. And then once you've landed on a model that you like, you can take

1:15:04 that one step further and you can walk it down the effort level. So start it off on maybe high and if it's not good enough, then move it up a little bit. And if it is good, then move it down to

1:15:12 medium and just keep finding basically the simplest model or the lightest and cheapest model that still executes at the level of quality that you're looking for. And then moving on here to number

1:15:21 six, we have the bike method, which is I've kind of already alluded to it multiple times throughout this video so far, but really it's just the idea to me that your skill is never done. Every

1:15:30 single time you run the skill, you're going to improve it. You're going to say, "Hey, here's what I really liked. Here's what I didn't like. Update the skill so that next time it's better."

1:15:37 Almost every single time I run a skill, I give it feedback and I tell it to update. And if you think about this, like you're teaching a kid to ride a bike, that's what a skill really should

1:15:45 feel like. You start off very cautious. You're watching everything. You're guiding them. You you've got your hand on the the the steering wheel. Steering wheel, I meant like handlebars. And

1:15:54 you're you're right there. And then what happens is you give feedback. You say like, "Okay, that was good, but you know, you were leaning a bit too far to the left. Make sure your weight's more

1:16:00 in the center." And then you keep going. Eventually, you take off the training wheels because the skill is getting better and you're getting more trust in the skill. and then you give more

1:16:07 feedback again and then eventually you take off the elbow pads, right? And then you keep giving feedback. But that doesn't mean that you just take off the kid's helmet and let them bike on the

1:16:14 highway. You're not going to do something reckless like that. You're also not just going to shoot them off down the road and then go inside and take a nap. You're still going to be in

1:16:21 in some way watching or having certain guardrails in place in order to make you feel comfortable and in order for you to have more trust in the actual skill itself. And I truly believe in the idea

1:16:30 that there's no such thing as a finished product unless your skill is super super deterministic and it runs perfect every time and you don't have to worry about it. But when the skills are more on the

1:16:37 judgment side of the spectrum, you know, as we talked about up here, the more your skills live over here, the less I believe that you can actually have a finished skill because like I said,

1:16:46 every time things might happen a little bit differently and every time your process might change or new models might drop and you're just constantly going to be giving feedback every time you use

1:16:53 it. Okay, so let's hop into Codeex here and we can take a look at how this actually works. Now when you are inside of your project and if you're inside of your AIOS which is usually what I'm

1:17:03 building in the codec skills will live in a folder called the aagents and then inside of the aagents there's a skill or sorry not a skill there's another folder calledskills and if you're doing this in

1:17:12 cloud code the skills live in aclaude and then in a folder called skills now these can be transferable I basically just tell codeex or claude hey see all the skills in there and bring them over

1:17:22 here too so just I have duplicates of them but that's where they actually live inside of your um project unless you have them at a global level which just means they're living locally across all

1:17:33 of your codeex projects rather than just inside of your AOS. So that's where they live and you can always just say hey you know Codex I have this skill I don't remember where it is could you show me

1:17:41 the file path and it will find it and you can move it if you want but anyways that's where they live and you can see I actually just ran this ex article skill on a video so it says your ex article

1:17:49 draft is saved. I'll click into it and I actually did obviously already publish this one but I didn't even have to change anything. I basically just clicked in here. I read it all and I

1:17:57 published it. And you can see that as it's going through, it already formatted everything and it used the video. Of course, it took screenshots of things and it like put them in the right spot

1:18:06 to line up with what was actually being said in the article. And it even added this little um I don't know what you want to call it, like a spotlight. It added that. I didn't actually do that,

1:18:15 but it will basically crop things out and it will add things like that into my actual article because I've told it in the skill to do things like that. And you can see that for this specific

1:18:24 video, this ran on GBD6 Astra. So, what I'm going to do real quick is I'm going to walk this down. I'm going to give this same prompt to Soul and Terra and tell it to make the X article and it's

1:18:34 going to use the skill. And we'll see how those outputs compare to Astra's output because this is a pretty complex skill. It has to transcribe the video. It has to open up the article. It has to

1:18:43 make the thumbnail. It has to format everything and it has to screenshot everything and put it in and then actually like drag it around. So there's a lot of browser use and there is a ton

1:18:51 of judgment inside of this. Like I said, this is not a deterministic skill. So you can see right now I have Soul, Terra, and Luna all running. It's been about 10 minutes, so I'll just check

1:19:00 with you guys when this is done. But I just wanted to show you how all of them because of the skill, because of the verification, are viewing images. You can see how many times Luna has viewed

1:19:08 these images here because it's taking screenshots and it's trying to understand, okay, which of these should be worked into the article and where and why. Same thing down here with Soul.

1:19:17 Obviously, this one is a super super visualheavy sort of skill, but because I've worked that in there, all of them are viewing images no matter what. And so, it'll be really interesting to see

1:19:25 how they decide to place them and, you know, what images they actually choose to put in there because X limits you to a certain amount of images per article. So, I'm going to let these run. We'll

1:19:33 compare the results and we'll kind of talk about how walking it down has changed my perspective on this specific skill. All right, so you can see now that we have Luna and Terra have

1:19:42 finished up. Soul is still working. So, we'll start reviewing these two and then we'll hop over to Soul. Hopefully, it'll be done by the time we're looking through these. So, Luna took 28 minutes

1:19:50 and 33 seconds. Let's open up the draft and see what we are working with. So, we have um the thumbnail in there. We have I designed a one person million-dollar business with Claude. Um the title is

1:20:01 the same as like the opening line, which I don't love. There's also this weird spacing was right there. I don't know if it tried to like center it, but I do notice that. Anyways, we have sort of

1:20:10 like the TLDDR. We have these different headers here. I'm yet to see an image. Here comes the first screenshot. So, this is talking about um the three different types of ideas to compare. And

1:20:20 in the screenshot, we do see all three. So, that matches up pretty well. Um we see agent report card as well as down here. So, really what I'm checking for right now is that it all sounds like me.

1:20:31 So, this is kind of trained on like my LinkedIn style guide and my YouTube style guide and like the way that I speak, but I'm also making sure that the images look good. So, the test suite is

1:20:39 a golden data set. We know what the correct answer or correct action should be. And we see an evaluation right here. The first run scored 88. That's what we see. Um, approved the new policy. Same

1:20:50 16 tests. Ran it again. That looks decent. I clicked create report and we don't really see the report here. So, that's a little bit off, I would say, but it's not too bad. I'm actually

1:21:02 impressed that Luna is doing it this well. This is customer support. This is kind of just the home dashboard. And then down here we see searching for companies using clay. It did do a nice

1:21:14 little spotlight animation here when we're looking at the actual pricing. And then we've got the three Ps down here. Okay, so this is not too bad. We'll have to compare, but I would say that I was

1:21:22 impressed by Luna's ability to screenshot, add spotlights, put it in the right spots. So that is Luna. That also took 28 minutes and 30 seconds. Okay, so we have Terara. And by the way,

1:21:32 all of these are running on high. So we have Terara here. took 26 minutes. So about 2 minutes, 2 and a half minutes faster than Luna. Let's open up this draft. See what we got. So same exact um

1:21:43 thumbnail. We have an image right away. How I built a $1 million AI business model with Claude. This image isn't great. Like to start off the article like that, that's not a very strong way

1:21:54 to start it off cuz it looks like it's cut off like that. So that's bad. Um we go through the three filters. Oage doesn't look very good either. I would

1:22:03 already say right now that um Luna did a better job. I mean, this isn't a bad Also, this is a zoomed in. So, it took a screenshot of the video and then it zoomed in on the score going from 88 to

1:22:13 94. So, that's not terrible. Read complaint check company fit recommend small trial. What you'll notice here is it actually, if I can open this up, it redacted sensitive details. So, it put

1:22:25 this block over it and said sensitive details redacted. So, I think that that's pretty cool. I like when it does things like that. It did a little spotlight effect here. It redacted more

1:22:34 sensitive details in this one. As you can see, it did it again here. So, it's it did it again here. It It shows so many times when it needed to redact sensitive details. It did it again. It

1:22:42 did it again. Okay, but this is obviously an issue. It didn't verify this good enough. Like, you can see that there's literally four images in a row down here. This one looks bad. I think

1:22:51 that, you know, you you you have to use the browser to like drag these around. And I think that it just gave up or it stopped and thought it was done. So, so far I would say that Luna did a better

1:23:00 job than Terara. And Terra was, you know, it's a more capable model apparently than Luna. It's also more expensive. So, right now, if I was choosing between these, I would choose

1:23:08 Luna. But now, Soul just finished up and Soul took 38 minutes. So, normally when I do run this, I run this on Astra. I usually run this on Astra low, and I run it on fast mode to get it done faster.

1:23:18 But, let's see. This did it. Let me open this up. Okay, so here is uh Soul's version. I built a $1 million business with Claude. We have the same sort of like TLDDR, the three filters. We have a

1:23:29 picture of the three filters right here. So, that's not bad. Choosing the product. Okay, it's zoomed in here. These are all zoomed in, by the way. So, it's it's it's using more, I guess,

1:23:37 vision intelligence to choose what to actually display in the image. You can see it redacted here the demo receipt. So, not too bad. It zoomed in a little bit here on the report, although we

1:23:48 don't see the score 88. What it what it highlighted here was 14 tests pass and two failed instead of showing us the 88. So, just pointing that out. It also zoomed in down here on some analysis of

1:23:59 that report. It did a little spotlight effect here to show the 94. It redacted more of these receipts. It did another spotlight here. It shows sales and support without losing control. So,

1:24:09 that's not terrible. We've got this customer support image as well. We've got more internal source ID redacted. Another spotlight effect on the money. And yeah, so this one isn't too bad

1:24:20 either. I would say honestly like I'm wondering if Luna's was better. And that's the thing, like these are so different and sometimes I have tutorial videos, sometimes I have more of this

1:24:28 style where they're edited. So, it's very hard to just say, "Okay, I've run one test and I know that now I'm going to go with Luna instead of like Astra or something for for a specific example."

1:24:36 Now, I will tell you guys, I've done this multiple times on multiple different YouTube videos and Astra just seems to be the best. So, I do stick with Astra on low, but I do think it's

1:24:43 worth running them through a few different examples and and understanding. And that's why I said that there's really no such thing as a finished product, especially the more

1:24:50 non-deterministic you get on that whole spectrum. And what you can see about all of these with Luna, it gave us a QA report, it gave us a visual plan, and it gave us selected frames. Terra gave us a

1:25:01 QA report as well. And Soul gave us a QA report. And like I said, that's the most important part is just them proving that they've verified stuff. So, it verified the title, the body, 76 non-mpty text

1:25:12 blocks, um 11 inline screenshots, seven H2 headings, visual review and privacy, editorial coverage, artifacts. It's proving to me that it's done. And it's proving to me that it did its

1:25:22 verification. And trust me, if it didn't do this QA report and it didn't do the verification, this skill would be so much worse as far as actually getting it to place the screenshots in the right

1:25:31 spot. So, for this specific example of the skill, let me talk about how I did all of these things. I took an article that I did manually and I did a few of these manually where I literally went

1:25:38 through the YouTube video, screenshotted things, put it in the right spot and I gave it some of those to look at. So, first was like, hey, analyze all these outputs. I want to turn this into a

1:25:45 skill. So, help me understand like what's good about these and and how do you choose the screenshots and how do you place them and things like that. So, I reverse engineered and then obviously

1:25:53 this is one very specific job. It was to turn a YouTube video into an X article. It's not just to write X articles in general. We've got I've got a different skill for that. This is literally to

1:26:01 turn a YouTube video into an X article, which means the trigger is very specific as well. So here, all I said was turn this YouTube video into an X article. And it automatically was able to invoke

1:26:10 the X article skill as you can see. But if I wanted to do a slash command and do X article, I can invoke the skill like that as well. But typically, if I say, hey, I want this YouTube video to be

1:26:20 turned into an X article, it just invokes the skill automatically because it's one specific use case and it's one specific trigger. We talked about the freedom level in here a lot. This is

1:26:27 obviously a super super non-eterministic process. We talked about the verification. We did the walking it down live. And as far as the bike method, like I said, guys, I would basically

1:26:35 come into here and I would look at what Tara did that I didn't like. And I would basically be like, "Okay, Tara, so in your article, you had some really weird screenshots in the beginning. It looked

1:26:44 like things were cropped out." The whole idea is that people can look at the screenshot and they won't need context of what happened a few seconds before that frame or a few seconds after. And

1:26:52 you have multiple examples in here like the first screenshot you did as well as the second screenshot that you did that just don't look very good. You also have things at the end like these random four

1:27:00 images that don't feel like they're placed right. So I feel like your verification wasn't good enough, but also the third the third image in this set of or sorry the second image in this

1:27:09 set of three at the bottom. It's like halfway through an animation. So that was first of all a bad screenshot to take and second of all a bad place to put it. So that's some of the feedback

1:27:17 that I have for you. Now if you could analyze this, understand what I mean by this and then update the skill so that you don't make these mistakes in the future and so that you can highlight

1:27:24 some of the important things that I just mentioned blah blah blah. And then I would just shoot that off and it would go through, it would look at the feedback and it would take all of that

1:27:31 information and it would update the skill. And that is how you start to sort of see that bike method coming into play every single run, give good feedback and also give bad feedback and tell it to

1:27:40 update the skill. Otherwise, all of those changes that you have to make manually, you're going to have to make manually again the next time. So anyways guys, those were the six tips. Those

1:27:48 were all of them ran on a real example of one of the skills that I use pretty much every day. Every time I make a YouTube video. So I appreciate you guys watching through all that. And also

1:27:56 guys, real quick, just wanted to tell you about my second channel where I just kind of like document these things that I'm doing at AI events or these different trips I'm taking to meet up

1:28:03 with AI creators or speaking at things. And if you want to check out sort of like what else I'm into besides just like making tutorials, then head over to the second channel. I'll put the link in

1:28:11 the description and I hope to see you guys over there. I hope you enjoy this type of content as well. All right, now that you guys understand how skills work and how you use them, let's talk about a

1:28:18 few specific examples of how you can start to build skills that are very, very specific to you and they feel like they're coming from you and they feel like they're coming from your business

1:28:26 when you're creating deliverables or and or when you want your team to create deliverables that sound like you guys and that they all look the same and things like that. So, let's hop into

1:28:34 this next section. Okay, so we've built our AIOS, we've got that all set up, we've talked about skills, now let's start to talk about some specific use cases of using your business and skills.

1:28:44 So, what I wanted to talk about here is how we can start to create some branded deliverables like sheets or docs or slideshows or, you know, even videos and websites and all that kind of good

1:28:53 stuff. But one thing that you want to get set up is kind of like your brand guidelines, your brand rules, logos, color schemes, typography, things of that nature. So, right here we're inside

1:29:03 of the AIS OS that we have set up. And you can see that I'm said, hey, what's my company in this AIOS? It's up at AAI. It's a Automation Society. And if I go in here, we have all of our files and

1:29:12 folders set up. So, if you are ever in a situation where you want to add some context or you want to do something with your AIS or sorry, not your ASOS, your AIOS, and you don't really know exactly

1:29:22 where things should live and you feel like you're going to do something wrong. The first thing I want to say is I don't think there is such thing as wrong. There's not like one universal standard

1:29:28 because it's AI. AI is able to search through things and understand your rules and every single person's AIOS is going to be different, but no one's is like wrong. The only way it's wrong is if

1:29:36 it's going to hallucinate things or if it can't find things and if there's way too much data, things like that. So, if you're noticing that it's lying or that it can't find things or that it's making

1:29:45 up details, that's where you probably did something wrong and you just need to sort of run an audit, analyze things, and clean some things up. And once again, you can do that with natural

1:29:52 language. But let's say I wanted to add some like logos and some color schemes and things like that. I would say, "Hey, so I want to add some logos. I want to give you some more brand guidelines so

1:30:01 that in the future when we're making things like docs or any deliverables, you can make sure it feels branded to me." meaning like my voice but also you know physical styling when it comes to

1:30:11 these deliverables. So I want to have some sort of folder inside of the AIOS called brand guidelines. Where should we put that based on our current folder structure

1:30:22 and we will go ahead and shoot that off. So it made the suggestion and it suggested what we could put in there and I'm just going to say yep go ahead and set that up. And now you can see that it

1:30:29 created that folder right here brand assets and we have some things in there like a few markdown files some spaces for logos and things. And so I opened up this in my actual folder right here. You

1:30:38 can see this is the U brand assets and I clicked into logos. I clicked into uppai and then I opened up another file explorer where I actually just want to drag in some logos. So I'm going to copy

1:30:47 these three and I'm going to copy them and I'm just going to come and paste them right in here. So now we have three upi logos. And then I could also do the same thing. I'll copy these four and I

1:30:58 will paste these into the AI automation society side over here. So now we just have a few logos set up. And now you can see that those are showing up here in our actual AIOS. Now what I would

1:31:07 recommend doing from here is yes, it's good to have logos in there. But I probably want to brainstorm a little bit on what are our brand guidelines. If you already have that, like if you already

1:31:16 have your colors and your typography, like the fonts that you use, then just go ahead and give that information to codeex right here and it'll put it in there. But if you need to brainstorm on

1:31:24 it, then just brainstorm on it a little bit. And what else you can do is have it generate you a brand guidelines doc. So watch this. You can see that I just dropped in some logos for UPAI. Now, in

1:31:34 there, I have a few different logos. I've got like a white one and a light blue one. The light blue one is the main logo. That's like our primary logo, and that's what I want you to always add on

1:31:42 our deliverables. Like in the docs, I want you to have it as a header. In the slides, I want you to have it like in the bottom left or something. And the colors are pretty much that same way.

1:31:50 You know, we like to use white and we like to use that light blue. And I always want our deliverables to feel professional and they should feel trusted. And so we really like to

1:31:59 utilize that light blue and those light colors. I don't exactly know what type of font that I like to use, but I don't want it to feel like too techy. I want it to feel very modern and clean. So

1:32:08 could you just go ahead and generate me a brand guidelines sheet, you know, like an actual image that shows um the way that we like to have our buttons, the way that we like to have um logos,

1:32:18 colors, things like that. And that's actually how I've made all of these different images that I have where we see like our brand guidelines with, you know, Nate Herk. It has like these

1:32:27 colors. it has these sorts of buttons and all this stuff like that so that as we're generating new things in the future, I can always just point my AI towards that file and that's like the

1:32:35 source of truth. And I like to usually have one that's a markdown file so it's easier for the AI to read and you can see the actual hex codes and everything like that and the font names, but then I

1:32:43 also like to have one that's like an image in case you want to send that to your team or send that to a sponsor or something like that. And so while that is making that, I'll show you guys some

1:32:50 other examples and I'll also talk about like why this is so important. And real quick, you can see that inside of the readme of the brand assets folder, here is what it generated. It says, "Hey,

1:32:58 this folder is the visual source of truth for UPAI and AIS deliverables. It keeps approved assets and usage rules findable, so future work uses the brand instead of guessing for any document,

1:33:06 PDF, presentation, report, whatever. Read this, read this, read here, go look at the templates, go look at the logos, do all this." And so it basically is just creating this read me so that your

1:33:16 agent always understands why does this folder exist and where should it look for different things like fonts and templates and examples and colors and whatever else you want to give it in

1:33:24 here. So you can see right now it's using its generate image tool to actually make that for us. And then we can turn that into a markdown file that's a little bit more structured. And

1:33:30 there you go. You can see that this came out super super quick for us. I can click into this. We see we have the primary logo. We have the reverse. We have a primary action button. We have a

1:33:37 secondary button. We have things about placements like docs and like slides. We have typography that it shows. We have a color palette right here as well with different actual hex codes that we could

1:33:46 put into Canva or to whatever we need to put them in. And obviously I could mark this up and I could make comments. If I don't like this and I wanted to comment on that or if I wanted to change

1:33:53 anything. That's what's so cool about the desktop app is you can do so many things right from inside here. You can even remove backgrounds. So you can play around with this until you get it the

1:34:01 way you want it and then say, "Hey, can you also turn that into, you know, like a markdown version for me or a text version for me so that different agents can read that easier." But look how easy

1:34:08 I was able to turn one logo into a formal brand guidelines doc. And now that you have that, what you can do is you can generate deliverables with this in mind or websites or videos or

1:34:17 whatever it is. But also you work that into the skills. And now I have skills like this. This is a student resource guide skill. So whenever I have a YouTube video or I read an article or I

1:34:25 just want to do a brain dump and I say, "Hey, can you turn this into a student resource guide?" It does all the stuff. It puts the AIS header up here. It uses this font. It uses these sorts of colors

1:34:34 and headers. It links my YouTube channel at the top. You can see this link right here to my YouTube channel. It puts these horizontal bars. It does all of this in the way that I love it. And then

1:34:41 at the very bottom, what it does is it also links people to AI automation society plus. And this is because I gave it the skill with my brand guidelines and with also, you know, of course, my

1:34:51 rules for building the student resource guide. I also have one for writing up at AAI memos. It has a UP at AAI header up here. It has this sort of like um I I guess I was about to say YAML front

1:35:01 matter as if this is a skill. Whatever this whatever you call this, like a little header. Once again, it has like different colors. It has these the way that I structure this. It even has the

1:35:09 footer here, the confidential internal UPAI communication footer. It has all of this now because once again, I gave it brand guidelines, but also I told it how I like to write our internal memos. It

1:35:18 also works with things like sheets. This is one that I run whenever I want to do like a YouTube report, whether it's quarterly or whether it's threequarters. Look at all this stuff. It has colors on

1:35:26 the tabs down here, so I can easily navigate between all these tabs. It has, you know, colors inside of this. You can see that the font isn't just the basic font. It's using Monsterat for the font

1:35:34 because that's on our brand guideline doc. It put the logo up here. It has big hero text up here. And then as I go through, I'm getting data visualized for me as well. I'm getting charts. I'm

1:35:43 getting different things like this because this is the way that I told it. I like to see it. And not only is it giving me all of this data with all of these links and all of these, you know,

1:35:50 interesting insights, but the whole thing feels branded to me. It feels like something I would send to my team and they'd be like, "Yeah, this is an upai sheet." And that's what I want you guys

1:35:59 to realize is that every single time that you're generating outputs with your AIOS, you can have it be exactly the way you want it. And you can give the brand guidelines and you can give the skills

1:36:08 to your team. So now everyone is generating things that look exactly the same. So that it all feels like it's coming from your brand voice so that it all looks consistent. And then when you

1:36:16 guys start to push out things like articles or blog posts or LinkedIn posts, it always also has the brand feel as well. And it doesn't just have to be Google Sheets. It can be PDFs or HTML

1:36:24 docs or whatever you want. But here's another example with a slideshow. This is a slide deck where we can see we have the branding up here. Once again, this all just feels consistent because it's

1:36:33 using the same color schemes and it's building it in a way that feels like it's an UPAI doc. And obviously, there's some other things that I put into the skill here like having these sorts of

1:36:40 visualizations, having, you know, swim lanes or um calendars or having actual data being visualized here with charts and things like that. And what's cool about this one is it also generated this

1:36:50 Excel sheet or Google sheet to go with it because it had to actually generate these visualizations. And it did this in Google Sheet and then it brought all these over and put them into the actual

1:36:58 slide deck that it built which I was just on the other tab for. And you can see that was actually right here. It was the mock data slide deck. We have all of these different things with the

1:37:06 branding. It also created um a different workbook which was the Google sheet to make all of those charts. And like I said for all of these we talked about how with skills we like to sort of

1:37:16 reverse engineer them. So what I did was I basically used codeex to help me get to a place where I liked an output. So let's say the student resource guide. I went back and forth. Now I have an

1:37:25 output that I like and now I turn this into a skill. And in that skill I basically say, "Hey, by the way, include all of these things from the brand assets. Include the header up top.

1:37:33 Include the logo down there. Include these colors, whatever." And the reason I really like to do it inside of Google is because we work out of Google Drive. So it also knows when it's creating

1:37:41 deliverables to put it in the share drive, to put it inside of the education folder or the marketing folder or the AIS folder or wherever it needs to go. It knows to just throw it right there so

1:37:50 that my team can access it automatically. Because when you're building things locally like this um HTML report for example, there's not really a problem with that. It's just

1:37:57 less editable because now if I wanted to edit this, I would have to manipulate the code and I would have to say, "Hey, you know, Codex, can you change this line or can you make this a different

1:38:05 color?" Whereas now in here, if I wanted to make a quick edit, I could just come in here and make a quick edit. And then I can also send this to my team right away. With this, I would have to like

1:38:13 download this and send it to them over ClickUp or email, they'd have to download it, open it up. If they wanted to make a change, they'd have to make a change and then send it back to me. So,

1:38:20 I just like to work with live cloud data, you know, cloud deliverables like Google Sheets, Google Docs. I really just like the Google ecosystem, honestly. And it's pretty cool because

1:38:28 you can obviously just come into your codecs, you can go up here to the plugins, and you can just connect the Google Drive plugin, and it lets you do pretty much everything through Drive,

1:38:36 Docs, Sheets, or Slides. And I know this is starting to sound like a Google ad, but trust me, it's not. But here is another example of an HTML skill. And this is what I call an HTML explainer.

1:38:46 So whether this is a deliverable or whether this is just something that I want, whenever I want to turn a concept into something that's a bit more visual and in a way that explains it to me or

1:38:53 to my students or to my team, I use this HTML explainer skill. And this one isn't as branded, but you can still see that in the top left, we have the logo. And you can still see that it kind of fits

1:39:02 the color scheme a little bit. But the whole point of this one is that this creates HTML. So it's explaining some sort of concept to me, and it does it in this way where it's very visual. It's

1:39:11 not very wordy. It's very easy to understand and it just lays it out as if it was trying to teach me a lesson. And it, like I said, it uses visualizations really well because that's the way that

1:39:20 I like to learn. I like to actually see examples and I like to see flowcharts and little symbols and things like that. So, this is an HTML explainer skill. So, whenever I need something explained to

1:39:29 me or I want to explain something to my team, I just go ahead and use this skill. And this also created a different version for us. One with like a transparent background and then one like

1:39:36 this. So, we could send that around and like I said, make some tweaks. And now you have some brand guidelines. And that was just a visual brand guideline doc. This is something else that you can

1:39:45 start to work in is like how does your company actually talk? Do you guys have slogans? Do you guys have mods? Do you guys have big hero stats that you always want your whole team to know when

1:39:54 they're writing emails or deliverables? For us, we have this thing called an AI phrase kill list, which is things that I hate that AI says a lot. And I've got a ton of things on here and this is shared

1:40:03 with my team so that then when they're generating emails or content or you know copy I don't want them to have their output sounding like AI. So this is one of the other things that we work into

1:40:13 our like team brand guidelines not just from a visual perspective but also from like a what does our voice sound like? Who is our avatar we're speaking to? All of those things that you want everyone

1:40:21 on the team to have sort of like a unified view on. That's also stuff that you can put into your AIOS. You can turn into skills and then you can share with your team. So, I would encourage you to

1:40:30 think about stuff like that. How do you know that something came from your brand? And more importantly, what kind of stuff do you want to make sure doesn't come from your brand? So, for

1:40:37 me, that was things like m dashes and things like saying, you know, and honestly, and honestly, it's not just X, it's Y, not a tool, not a feature, a revolution. And speaking of visual

1:40:47 elements of a brand, we know that codecs can make us images because chatgbt image 2.5 or whatever the image model is from OpenAI by the time you're watching this, it's inside of codec. So if I ever say,

1:40:58 "Hey, make me a picture, make me a thumbnail," blah blah blah. It can do it from here, which is amazing. And I really think OpenAI's image model is the best right now. But what it can't do

1:41:06 right now is generate video or it can't use other AI models. If you wanted to do a different, you know, maybe Gemini's model is really good, nano banana, or maybe you want to do like a sea dance

1:41:14 video or a cling video. Well, let me talk about next how can you connect codeex to a different tool that will let you generate images and videos with tons of different models. So let's jump into

1:41:25 the next segment. One of the great things about Codex is that right inside of it, it can make images for you. As you can see right here, I gave it these three images and I said, "Hey, I need a

1:41:32 YouTube thumbnail." I described what I want and then I got it right here. But if I asked it to turn this into an animated video, it wouldn't be able to do it with its native capabilities. It

1:41:40 would have to find some other form. It would have to build code. Like here, it wants to use hyperframes. So, typically what you'd do is you'd connect it to something like Higsfield that has a

1:41:47 bunch of video and image models like Cance 2.5 and all these other ones like Cling and Omni Flash and Miniax. The problem with Higsfield was that it was way more expensive because you had to

1:41:58 get on a subscription. If we look at the monthly, if you wanted to use Seedance 2.5, you had to be on a 60 bucks a month plan and you still had a limited number of credits. But Higfield just actually

1:42:07 dropped an API which means that you can actually just pay per usage. Meaning you can connect Codeex to Higsfield API and then you're not actually paying every single month. You're only paying for the

1:42:17 times that you actually make requests. So today I'm just going to show you real quick how easy this is to set up. So let's get right into it. So this is a completely separate account. What you're

1:42:24 going to do is you're going to go to Google and you're going to type in Higsfield API. And then you're pretty much going to click on this first link right here. And it's going to take you

1:42:31 to this API dashboard where you'll be able to create your own key and create your own account. And when you sign up, you're going to get a ton of discounts because they just dropped this, which is

1:42:39 15% off sale models, up to 50% off your favorites, and you get 15 bucks right away in your account to play with. So create an account, and then when you're inside of that, it's going to look like

1:42:47 this. What you're going to want to do is you're going to go to either the quick start on this right hand side and you can copy this prompt if you want, but you don't have to. It's it's a lot

1:42:55 easier than that. But you do have to make an API key. So you'll click on create API key. You'll type in a name. So I'll just call this one demo and you'll create your key. And then all you

1:43:02 have to do is copy this key right here and then hop into CEX and I'll show you what to do next. Real quick before we hop into CEX, I just wanted to show you guys how to make sure you take full

1:43:10 advantage of this launch because this is a 7-day limited offer. So, these things up here that we see, 15% off all sale models. That is just when you sign up, you'll already have that discount. In

1:43:21 order to get up to 50% off your picks, you have to come into the pricing section and it says choose three models with max discount up to 50% off. And then you would just come in here and

1:43:29 choose two video models and one image model. And you can see all of the prices over here. As you can see, with 30% off, we would be getting it for this price compared to the original current price.

1:43:38 And then the last one is getting 15 free dollars to play with on your account. All you have to do here is connect a credit card and use an eligible business domain email. And then once you've

1:43:48 claimed all of that stuff, you now have that API key. You now have everything set up and you can run 30 generations at a time. So let's hop back into codeex and I can show you how to get this all

1:43:56 set up. So you're going to open up a new chat and you're just going to say, "Hey, inside of myv file, can you just create me a placeholder for Higsfield API? I'm going to drop you in a Higsfield API

1:44:06 key." And then you can just figure out how that all works so we can start running some image and video generations. and then just shoot that off and it will tell you exactly where

1:44:13 to paste in your key. You don't want to paste it right in here into the chat thread. You want to paste it into thev file. So you can see I already have that in there. So it says, "Hey, that already

1:44:21 exists." But if it doesn't, it'll give you a placeholder. And then all you have to do is ask for the file path that you can actually open this up. So I could rightclick this, go to open with, and I

1:44:28 can just open this with whatever I want. I'll choose default app. That opens up my notes where I can go ahead and see right here it added Higsfield API key. And then you will just go ahead and

1:44:36 paste in your API key right there. And then make sure you save that before you close out of it. and then just go, "Okay, cool. I put in that API key. Can you see what models are available for me

1:44:45 through that Higsfield API?" And then just go ahead and shoot off a test video generation where there is a man sitting at a laptop using Higsfield and using codecs to generate videos. But you guys

1:44:57 won't have to worry about that at all because I have this skill right here called Higsfield API generations. And it has all these defaults in here. It has all of the settings. It has all of the

1:45:05 different things that you might need or you would have your agent have to normally go out and do research on. Everything's right here. So basically all you have to do is you've put in your

1:45:12 Higsfield API key into your EMV. And now you can see it's basically going to use that whenever you need an image or a video and it knows exactly how to work with Higsfield API. It knows about all

1:45:20 of the different functionality and options that it has in there. And now it's just generating me a bunch of different images and videos that you can see right here on screen with just using

1:45:29 natural language. So pretty cool stuff. If you want to grab this skill, it's completely free in my free school community. The link for that is down in the description. Just head into the free

1:45:35 school community. You're going to go to classroom and click on all YouTube resources and there will be a huge database with a bunch of resources. You should be able to find it right in

1:45:42 there. And that skill is also going to tell you exactly how much each of your generations cost you so that you don't have to be constantly going back into your Higsfield dashboard to look at the

1:45:49 analytics. You will just get that right there in your chat thread. So, while that's running, I'll just go over here and show you real quick what actually this stuff can look like. So, we have

1:45:57 Higsfield API benchmark out of all of these videos that I generated. So, I'll show you each one like how much it costed or sorry, cost. My goodness, how dare I. You can see that the total spend

1:46:08 for these was $7.40 and we were using models like Cance 2.5. And once again, this would have been only available to us on a 60 bucks a month plan. And then you're just sitting there paying 60

1:46:19 bucks a month even if you don't generate anything that month. So, anyways, we have some video outputs here. You can see this is Cance 2.5. This one has no audio, but you can generate these with

1:46:27 audio as well. So, that's a super super clean shot. Same exact prompt went off to cling 3.0. And this one has I don't know those letters look a little bit different compared to that image, but

1:46:37 they have similar sort of like background. The letters in this one are also off. So, it's really good to get in here and shoot off the same prompt to a bunch of different AI models and see

1:46:46 which one comes back best. In this case, Nova came through only as Nova in Cedance 2.5. So, in this case, it was the best. But, of course, with Cedance 2.5, you are going to be paying a little

1:46:55 bit more, but once again, it's only pay per usage. So here, this one was 1.6 bucks. This one was only about 20 cents, and this one was only about 50. But they also can generate sound. So let me play

1:47:06 you guys a few examples of some UGC content. And we'll take a look at the cost as well. >> I used to skip breakfast. Now I grab Nova protein coffee. 20 g of protein,

1:47:18 real coffee, and I'm out the door. >> So this was Cedance 2.5, and it was $3.23. Here is Cling 3.0. I used to skip breakfast. Now I grab

1:47:30 Nova protein coffee. 20 gram of protein, real coffee, and I'm out the door. >> And as far as realism, I think Cling 3.0 did a really, really good job here with this character, you know, selfie style

1:47:43 UGC for only 60s, whereas this one you wouldn't say. I don't know. It's like it's not noticeably better as far as like makes a huge difference. But in general, I mean, Cance 2.5 is in my

1:47:53 opinion the leading video generation model right here. But we also have Mini Max and this one was a $110. >> I used to skip breakfast. Now I grab Nova protein coffee. 20 grams of

1:48:04 protein. Real coffee. >> Anyways, I think that this one was clearly the worst out of just these three very, very simple videos. But it's important, like I said, shoot off the

1:48:13 same prompts at different models. See how the cost differ. See how the quality differs. See what they're good at. because Cance 2.5 might be the best at like people and product shots, but maybe

1:48:22 Miniaax or Cling is better at different sort of like cartoon animations or something like that. And then of course you can use Codex image generation. GBT image 2.5 is really really good. It's

1:48:30 probably my favorite, but there are also other image models you can try out like Higfield Marketing Studio here or Grom Imagine 2.0, Soul 2.0, and these are obviously much cheaper than generating

1:48:41 videos. But I did take this exact same prompt and put it into codecs. And this is what we got here. So if you like this better then maybe you want to just stick with GBD2.5 and then you also have the

1:48:50 benefit of this just counting towards your codec subscription or your chatbt subscription rather than paying separately per usage for that. Anyways, hopping back here into codeex. What's

1:49:00 cool about this is Codex will go ahead and do all the research. So you don't have to think about the API endpoints. You don't have to give it anything else. You just tell it to figure it all out

1:49:07 and it will. And so that's why back in here you can copy this prompt if you want or you can just say like figure it out because you can see this prompt is literally pointing it to the actual

1:49:15 Higsfield API documentation. You can see it says your key authenticated successfully. The live catalog returned 76 endpoints. 18 are text to video, 44 are imagetovideo, 11 are text to image

1:49:26 and plus there's a few other routes. And it's going to use Cance 2.0 for this first run. The request was accepted with this ID and now it's basically just going to keep pulling on that request

1:49:35 until the generation is done. And polling just means it's going to check in and then if it's not done, it's going to wait and it's going to check in again. If it's not done, it's going to

1:49:41 wait and it's going to keep pulling or checking until that video is actually done. And then it will download the MP4 locally. And so, while this is finishing up, let me just show you what else is

1:49:49 available inside of the Higfield API. If you go to pricing right now, it says that you can get up to 50% off. And this is only for the next 7 days. So, if you're watching this early, go here and

1:49:59 you can choose two video and one image models to get 50% off on. So, you'll literally just come in here and it will say, "Hey, what are your models you want to choose?" I personally chose Cance 2.5

1:50:09 and Clink 3.0. And then for the image model, I chose the marketing studio image. But then you can just go ahead and look at all of the models in here. And you can see the different modes that

1:50:16 they run in like standard or turbo or, you know, 4K, whatever it is. And you can see all of the information about all these different types of models and their pricing. And then the analytics

1:50:25 will basically just show you how much you've spent. Um what is your request looking like? What are the models you like to use and things like that. So just a pretty little analytics dashboard

1:50:34 for you. But you can see we have this video back. Now, this is already looking a little bit weird with Cance 2.0 because the laptop is facing the wrong way, but let's give this a quick play.

1:50:48 Now, I'm really not quite sure if anybody uses a laptop like that, but either way, this was Cance 2.0 and there is a better model out there now. But what's important to realize is that the

1:50:55 model isn't just going to run perfectly for you. You have to be able to prompt it in the right way and you will start to build skills around how you like to prompt image generations and video

1:51:03 generations. And that's where this gets really powerful is when you work in all of these different skills and you can add things too like your brand guidelines or a consistent character

1:51:11 that you want to have and then over time your ability to just inside of codecs generate these different videos and image assets is going to get better and better. Now I did want to end off this

1:51:21 video talking about is this actually cheaper? And the answer is yes and no. It all depends on the break even point because now that you're being charged per usage rather than based on your

1:51:29 credits that reset every month, there's a break even point. And if you're paying per usage to the point where you're going over that break even point, then this is probably where you'd want to be

1:51:38 on a subscription. But if you're only infrequently generating videos and images on a monthly basis and you're below that break even point, then it's much cheaper obviously to be on the pay

1:51:48 per usage billing, especially when you're trying to figure out like what is your cadence, then starting with pay per usage is a lot safer than just committing to a monthly subscription.

1:51:57 And so I've run the math here and on the plus and ultra plans within Hicksfield, this is what it would kind of look like as far as those break even points. So if you're on the plus monthly, it's about

1:52:07 fewer than 19 clips. And if you're on the plus annual, then it's roughly 16 clips. And if you're below that, then the API makes more sense. If you're on ultra, it looks like 50 on monthly or 39

1:52:17 on annual. So ultimately, these are the three kind of questions I wanted to answer here. If you're maxing out your subscription plan every single month consistently, then you probably just

1:52:25 want to stay on a subscription. If you're only generating things occasionally, like maybe some months you'll run, you know, 15 videos, some months you won't do any at all, then API

1:52:33 will win. And if you need programmatic workflows or you want to bake this API into your own, I don't know, your own website or SAS product or your own internal tools or whatever it is, then

1:52:42 the API is going to win there as well. So, it's always hard in the AI space or in the tech space to just come out of the gate and say, "Yes, this is objectively cheaper or yes, this is

1:52:49 objectively better." It's always very situational. So hopefully this all kind of makes sense and you can kind of use this to figure out in your specific situation what is probably the right

1:52:58 path for you. And what's really cool is when you get all that dialed in, you can do all this in bulk. So we've actually done things where we've generated assets for like our events or for you know ads

1:53:06 or whatever it is and we basically just use codeex or claw to help us brainstorm all the briefs and all the angles and then we just shoot off a bunch of workers to generate all those assets and

1:53:14 then you wake up and everything is done. Now, what's really cool is what you can actually do with Codeex as far as building websites when you actually have learned now how to generate other videos

1:53:23 and other images to put on those websites. So, we're going to hop into this next section where I kind of show you guys what's possible with Astra and with Codeex when it comes to designing

1:53:30 things like websites. We have a new AI design king and it's not even close. Take a look at some of these outputs I've been getting with Astra. Look how cool this scroll is. How about this

1:53:39 example with GLO? I think that this is just absolutely incredible. I think Asha does a really good job at making things 3D, dynamic, immersive without making it overwhelming because this is still a

1:53:48 very clean hero section. Now, this one for perk form is a little bit more out there, but it's very impressive what it was able to do, but I still think it's pretty clean. And we have other types of

1:53:56 animations here. Look at these cans come in. I love how it was able to create these in 3D. Now, I just put out a video that I said Fable 5.1 was the design king because it helped me create

1:54:04 something like this, which I loved. But Astra is so much better with it so far. Now, all of these examples that we just looked at and that we're going to keep looking at were oneshot prompts. Think

1:54:13 about that. If I took the time to actually sit down and iterate on these, they would be so much better. This was Astra's first pass at an AI Automation Society website. You can see it tried to

1:54:21 keep the same vibe. So, right now, the one that Fable built, I think, is the winner because I've iterated on this like 15 times, but as a first shot, I think that this is really, really good.

1:54:31 It's on brand and like I said, it has these animations and these interesting, you know, like immersive parallax things, but it's just it's not too much. I've been seeing you guys talk about in

1:54:40 the comments how, oh, I don't think these sites convert. And I think you're right, but it depends on the three Ps that I talk about, the pain, the person, the promise. But I love how subtle this

1:54:49 is, but instantly I think that this feels premium. You feel the depth and it's not too overwhelming, and it still fits the brand that you know that you have here. It still fits the messaging

1:54:57 that you're trying to get across. even something like a luxury watch brand. I think that even just this little element right here that lets us sort of scroll through really makes it, you know, it

1:55:06 sets it apart from other types of websites in this industry because it's not just about the hero section. I think that the way that all of these sites are designed are very on brand. It's not

1:55:15 like one super specific template that we're following every single time. It's different each time. So, I know that I've just been jumping all around with examples. I just wanted to show you guys

1:55:23 some of the different things that it's been able to do and some of the different ways that it uses its creative freedom and taste because I basically just told it to make me a bunch of

1:55:31 websites. Now, the one thing that all of these that I just showed you have in common is that they used my scroll craft skill, which if you guys don't have that, join my free school community

1:55:39 linked in the description. That's where you can get the skill, and I'm constantly updating that thing as I find out more about AI design. Not only does that skill understand things like

1:55:46 layering, as you can see here, and scrolling, but it also understands things like typography and spacing and all of those other elements that you need to make the website actually feel

1:55:55 good as you continue to scroll through. But I think that's been one of the biggest things that I've realized lately is the layering adds so much depth because you can clearly see how the text

1:56:03 is on a different layer than the background image, which is on a different layer than the bike, which is on a different layer than the rocks right here. And that's what makes it

1:56:10 feel super immersive without being over the top, at least in my opinion. because at the end of the day, all of this stuff is subjective and it's all about taste. But like I said, my scrollcraft skill

1:56:18 has all that knowledge in there. And what you've noticed about all of these different websites I've shown you so far is that they're all different types of businesses. We had a content agency, we

1:56:25 had physical products, this one is a consulting firm, and all of these have different vibes. They have different ways that the sections are actually structured. This one has like a sidebar

1:56:34 over here. And we haven't seen that yet on the other ones. Now, on top of website design, I'm also talking about things like motion graphics and design using something like hyperframes. So,

1:56:42 let me play this example for you and pay attention to the 3D elements and also pay attention to the music and how it's synced to what's going on. and I'm not going to play this whole

1:57:21 thing, but it also created me a vertical version. So, I'll play a few seconds. Now, those obviously aren't anything crazy. they're not going to go viral. But if you consider that that was just

1:57:37 two prompts and took me, you know, 20 to 30 minutes, how much time would that have taken and how much would that have cost you if you paid a human editor to do that? So much more time and money.

1:57:47 And here's an example that I think is even crazier because what I had to do here is for AIS Live, we have this folder where we have like over 100 gigabytes. You can see 152 GB of footage

1:57:57 to look through. And all I did was give it the link and I said, "Hey, create me like a sizzle reel recap that's a minute long from this event." And look what I got. logged off yesterday just buzzing.

1:58:05 I had goosebumps. I was so excited and I had so much fun. Cost doesn't justify price. Price justifies cost. What is the value of what you know and

1:58:38 what you know how to do? This culture that we have here at AIS is is just really special. So, the only thing wrong here is this domain. That's not the real domain. But

1:59:05 I am just so impressed by this. And I know some of you guys might look at that and be like, uh, I saw some editing mistakes or eh, whatever. But the music, what it had to think through all of

1:59:14 that. It had to look through all that footage and basically figure out how to tell a story out of it. And I thought it did a really good job at that. It also had to look through all of our logos and

1:59:22 our designs and our branding and our images and our assets for that. It found everything and it pulled it all in. And I was just like honestly for 35 minutes is how long it took. So so impressed

1:59:32 because I've tried to do something like that with Soul. I've tried to do something like that with Fable 5.1 and it did not go nearly as well. this thing just has the understanding and it's so

1:59:41 pragmatic and logical about it that I am just like seriously it's it's it's another huge leap forward to me now when we talk about actually killing AI design slop or having things that really go up

1:59:50 to the next tier of course it's using things like skills that you might find on GitHub like mine or like other people's but what I think is the most important thing is inspiration it's

1:59:59 really hard to just give an AI model a task like make me a website here are the brand guidelines and have it just actually create you something that doesn't look very AI generated so what

2:00:06 you have to do is you have to have inspiration and you can find that in the form of something like godly.design where you can find actual websites to take inspiration from or you can find

2:00:13 individual components like the hero sections or the CTAs or the footers and you can find these components. You can do the exact same thing on something like 21st.dev. There are so many amazing

2:00:23 things here. Like I said, hero sections, different components for the UI, you can find forms, AI chats, you can find buttons. Using all of these things and giving these code blocks or giving these

2:00:33 website URLs to Astra makes them so much better because now I could come into any one of these 10 sites or the other ones that we had generated and just be like, you know what, I really like this, you

2:00:43 know, um, button for example right here, but I would like this to be changed out with, you know, this button that I found on 21st. Because I feel like it's more on brand or I like the, you know,

2:00:51 dynamic element around it. You can also come to something like awards with three W's, awards.com that has a ton of award-winning sites. And you can also come in here and organize by category,

2:01:01 which I think is really important because an e-commerce site probably needs to look different than, you know, a restaurant and hotel site, which probably needs to look different than a

2:01:07 fashion site. Because yes, there's one element of AI slop where it's like, you know, you've got those certain types of buttons and the certain type of fonts and the certain gradient in the

2:01:15 background. But the other element of AI slop is when it feels like you're looking at something that is meant for someone else. If you're a 70-year-old man and you're looking for medication on

2:01:24 your website and you're scrolling through a 3D world, you're going to be like, "This is the worst thing I've ever seen." But if you're a teenage dude and you're trying to buy new headphones and

2:01:31 the headphones on the site are spinning and you can see inside of it and it's like transforming. You're going to think that's sick. And that's just the truth. So think about the pain, think about the

2:01:38 person, think about the promise, and think about inspiration sites. And then you just get out of Astra's way. You give it your brand guidelines, you give it the copy, you give it scrollcraft

2:01:47 skills so you can create all of this sorts of cool stuff. And then you watch the site or you scroll through it. You say, "Uh, I don't like this. Fix that. I don't like this. Fix that." And then you

2:01:56 just keep iterating. And that's how it's done. So CEX is probably the best browser use agent I've ever played with ever. And browser use is super powerful for many things, which is what I'm about

2:02:06 to get into the next video about. But basically, it just can control your computer. Not only can it control your computer, but it can also control browser tabs and it can save sessions

2:02:15 and things like that. So, if you can't use an API for some reason, maybe the website you want to use or the service you want to use doesn't have one, you can use browser use. And like I said,

2:02:24 Codeex is just the best at it, at least right now. So, I'm going to hop into the next video with you guys about how you actually can take advantage of Codex's browser use. Codex's browser use is

2:02:32 probably the best that I've ever tried before. It can literally do anything on a browser or on your actual local computer. And if you need to be signed in for something, you just sign in once

2:02:41 and Codex will save it. So, the possibilities are endless. You can actually automate anything. Now, let me show you guys how to set it up, how easy it is, and a few good use cases. All

2:02:48 right, so I'm using the Codeex desktop app, which is just honestly the main way that I like to drive Codeex. It's super super slick, and we get the browser to pop up on the right hand side, which

2:02:56 I'll show you guys in a sec. So, luckily, we live in a world now where most tools that we need to use have an API. So, it's much easier to just connect codecs through the API, or if we

2:03:05 just go to the plugins, there's hundreds of plugins in here. But every once in a while, we don't have an API, and you can do that with browser use. But before I show you an automation like that, let's

2:03:12 just take a look at this example where you might be building a website. You might be building some sort of app and you need a browser, an agent to actually test through things. If you guys watched

2:03:21 a recent video I did on like building an AI SAS in one day, then this was actually from that demo where I had a browser agent get spun up and test the heck out of our app. And it did 85

2:03:31 focused automation checks with real browser workflow testing, which means it was clicking around, it was typing things, it was hitting buttons, and it was trying to break the app. And it

2:03:38 found a ton of bugs. and then we were able to just fix them. So, in this example, I have a super simple website spun up here on the local host. It's a form submission as you can see, and

2:03:46 there's a few issues before we actually go ahead and start testing this thing. You can see like this spacing is weird. This thing is overlapping. And so, what you can do is you can click on this

2:03:54 annotate button, and that lets you choose very specific elements to actually change. So, I can click right here on this phone number box, and I could describe, hey, this is kind of

2:04:03 like overlapping. It's out of bounds. It's covering up the company box. So, I need you to fix this. And then when you shoot that off, it actually goes over here into your chat. And then you can

2:04:11 make another annotation like, okay, so this whole box isn't really aligned on the same y-axis as number one and number three. So you need to fix that as well. And then those annotations are saved

2:04:19 over here and we can just go ahead and shoot them off. So that's kind of like a lovable style or like a claw design style thing that you can do with the inapp browser in Codeex. What else is

2:04:28 really cool about this is that you can actually just drive this thing. So you can go to Google, you can search for Grand Canyon images, you can go to all, you can go to different links if you

2:04:35 want. You can use the browser right here while you have Codex open. But of course, Codex can also just take control of it. Now, we can see right here is my mouse, but also this is what Codeex's

2:04:46 mouse looks like. And that's how you know it's now sort of getting ready to take control of the browser. Okay, so now that those changes have been made, let's actually start testing out our UI.

2:04:54 I want to see things that maybe me as a human, I would maybe go through here and test like five or six edge cases, but we can see how Codeex can do it way quicker and way more. Okay, so you have the UI

2:05:04 open. I want to make sure that there are no bugs. I want to make sure that this thing can't break. So, what I want you to do is use your browser use and just test the heck out of this thing. Try

2:05:12 different things. Try your best to break it and let me know what bugs you find and what things we need to fix. And so, by the way, while this is running, there's a difference between browser use

2:05:20 being headless or headed. Headless means it's running in the background. So, you can keep working in the foreground and you won't actually see it going on, but it can run a lot of things in the

2:05:28 background. But, if you want it to be headed, that's basically where you're able to see it. So, we're probably going to be able to see some things right here. But if you wanted this to just

2:05:34 work for a few hours for you headless in the background, you could ask it to do so and then it wouldn't disturb whatever work you're doing. So if you ever hear that terminology, that's what it means.

2:05:41 So there you go. You can see that the mouse is moving around right now. It tried to click continue and then it says first name is required. Enter a valid email address and enter a 10-digit phone

2:05:50 number. And now you can see it just entered in some information and it was able to hit continue. And it's just going to keep going through this QA check. We're

2:05:57 validating that some of these fields are required options and that the continue button won't actually let it move on without filling those in. And by the way, the reason you're seeing some of

2:06:05 these UI bugs is because I asked this chat to like make there be a few UI things to fix because I wanted to show you guys the annotations. So just ignore that. Mainly what I'm trying to show you

2:06:14 guys now is the actual browser use QA function. So here you can see as it's clicking around, as it's trying to navigate this UI, it already said the core click path mostly works, but I've

2:06:24 confirmed two real data integrity problems already, which is that obviously invalid contact data does get through and changing the country code to UK resets to US when you return to edit.

2:06:33 I'm now testing keyboard submission and restart behavior, which often exposes bypasses the click path misses. So think about it like this. No matter what you get, you're going to want to test it.

2:06:42 But would you rather have something that has been not tested at all, or would you rather have something that has been tested 100 times already by an AI agent? Especially if you tell the agent to try

2:06:51 to break it, it's going to find some things that need to be fixed. No matter how good you are at trying to predict edge cases, when you push an app into production and users start using it,

2:07:00 people are weird. People are unpredictable, so they're going to do things. And so this agent is able to hopefully simulate weird humans and just find things that you might not have

2:07:08 thought of. You can see here, it even switched this to mobile view to still make sure that everything's working. And look at this. In mobile, it found a few failures in responsiveness. So that

2:07:15 isn't something that I yet had tested. Okay, so this thing finished up now. And you can see that we have a bunch of things that passed, but a bunch of things that also failed. So it was able

2:07:24 to find all of these bugs because it was testing the UI. And it didn't even do it that long. If we wanted to make this even harder, we could do a /go goal and we could say, "Hey, don't stop until

2:07:32 you've tried a 100 different unique scenarios of, you know, UI edge cases and then come back to me." But it still found a lot of things that were broken. And now all we'd have to do is say,

2:07:39 "Okay, cool. go implement those changes, go fix all those bugs. Okay, so let me show you guys another example. This is when you maybe have some sort of platform where you want to be able to

2:07:47 like download statements or something, but they don't have an API. So you can actually interact with it with Codex's browser use because it basically uses vision and then AI to analyze like what

2:07:57 it's looking at and where to click. And I would I just want to show you how easy it actually is to build these. All I said was, hey, you know, I want to build this skill blah blah blah, what would be

2:08:04 helpful? And you could record a quick loom. You could send screenshots with arrows, but you actually don't have to. you can basically just describe pretty clearly what you need to do. So, what

2:08:13 happened here is it opened up Relay. I signed in and it saved that session. Now, I said, "Okay, in the relay dashboard, you're signed in. What I need you to do is go to the dashboard, click

2:08:20 on accounts, find the statements, um, click on these two accounts, download them as CSVs, and then put them into this folder." And it was able to do that on one shot. I didn't have to give it

2:08:29 any feedback and it downloaded them and saved that whole flow as a skill. Now, this is super helpful because now I could at the end of the month just set this up as a scheduled routine because I

2:08:38 could set a scheduled routine here and say, "Hey, use this skill on the most recent, you know, month statement in Relay." And for the most part, Codex is going to save your sessions when you're

2:08:48 logged in. However, there are some limitations like the fact that Relay is going to pretty much boot you out because it's, you know, it's a banking platform. It's going to be secure and

2:08:57 boot you out. So, look at this here. I asked it to do this again and it said, "Hey, relay is open at the login screen, but you have to sign in again." Now, what's really interesting that you can

2:09:04 do is you can actually open up your settings. You can have a password manager inside of Codeex and you can see this goes to your browser and now you can go to import them so that Codeex can

2:09:14 actually fill that out for you. And that way, you're not putting it into the chat history or anything like that. You're saving it more securely. So, all you have to do for that is create a CSV file

2:09:21 that looks like this. I've got the name, the URL, the username, the password, and any notes, which in this case are empty. I imported that here right here as a file, a CSV file. And now we have

2:09:30 relay.com password. Go ahead and try this again in the password manager for the inapp codeex browser. You should be able to log in now and then run the skill on June. All right. So over here

2:09:40 now you can see the browser has taken control and it is going to put in my email and my password. Now if this asks me for a two-factor authentication code for my phone or something, then I'm

2:09:49 obviously not going to be able to do that. I'd have to do that manually. But hey, if I was on a Mac and my text went there, and I used computer use to do that, technically, you would be able to

2:09:57 automate that whole thing. But I would say you probably want to be careful about stuff like that. But as you can see, it was able to pull the passwords from my password manager, paste those

2:10:04 in, and now it should be able to log in to relay for me. You can see here that the mouse has moved over to accounts. It's then going over to statements, and it's going to be able to download these

2:10:13 two statements for us. So, you can see it's pretty cool that it was able to figure all this out. Remember, I've only ran this skill once, and all I did was instruct it with natural language to do

2:10:21 this. Now, realistically, is that the way you should do it, especially with something like a bank account? No. What you should do is have it do the skill. You sit here and watch it and have it do

2:10:31 that like at least 10 times where you're very closely watching it to make sure nothing goes wrong and you're making sure the script in the skill is super strict because at the end of the day,

2:10:40 it's using vision, it's using AI, and that's how it's deciding what to click and what to do. So really the way that I think about it is when you want to automate something, your first call

2:10:48 should always be to go for an API. Like that's what you want to do. It's the fastest. It's usually the cheapest. It's usually just the most consistent. Now, if there's no API, that's where you

2:10:58 would maybe say, "Okay, cool. Well, now let me see if I could use like a deterministic, you know, like macros script, right? And that's where you'd basically be able to have Yes, you know,

2:11:07 it'd be more of a browser use because you have a mouse that's clicking around, but it's not like your mouse needs reasoning. Your mouse doesn't need to look at the screen, figure out where to

2:11:15 click, and based on the next UI that pops up, where do I click again? This is basically more like, okay, I have to click on this pixel, then this pixel, then this pixel every single time.

2:11:26 That's something that you could do for much cheaper and much safer by just programming a macro. And that's not that hard either. You just ask Claude or Codex, hey, help me figure out how to

2:11:34 make like a deterministic macro script to just click on these pixels in that order every single time. But if you can't do that every single time because there is an element of vision or

2:11:43 reasoning, then you need to come in here and you need to do something like some browser use. And luckily, Codex has your back. I've tested a lot of different apps and different CLIs and different

2:11:52 things for browser use. And so far, Codex is just really solid. So, as you can see now in my downloads, this is the statements that it just downloaded as a zip file and it basically took these and

2:12:00 it put it into my Herk 2 project. If I go in here and I scroll down to statements and I go to um probably June. Yes, June. You can see right here, it put those in here in the correct spot.

2:12:11 Okay, so that was an example where we needed to save a password somewhere, right? But I've done this with things like school. I've done this with things like X. I've even done this with things

2:12:19 like Instagram that don't make you constantly resign in because it saves that session. So, let me show you another example. I'm going to go over here to YouTube and I'm going to grab

2:12:27 the URL from the video that I uploaded earlier today. I'm going to go back into Codeex, paste in that URL, and say, I need you to turn this YouTube video into an X article for me. and then use

2:12:37 browser use to go ahead and um format it and put it into an X article as a draft and then I will go ahead and review it. So, I know some of you guys are probably thinking, why don't you just use X's API

2:12:47 for that? You can, but it's a little bit weird with the formatting. It's also weird with sometimes adding things like media or thumbnails for some reason, at least with my testing that I used to do.

2:12:57 And it's also it charges you, right? Because you have to pay X per usage in order to do something like that. And so because this is just part of my codec subscription, I can just run this skill

2:13:07 and it will go ahead and format everything for me and take care of it. I even have this skill worked out where it can go ahead and open up my YouTube videos and it can take screenshots of

2:13:14 specific moments and put those into the article too where it makes sense. So obviously this is a bit of a loaded process because it has to do a few steps before it actually has the article ready

2:13:22 to put into X. So while that's happening, let's just open up a new chat and let's see if our X session is there. Can you open up a browser? I want to see it right here in the app and just go to

2:13:32 X and scroll around a little bit and show me if there's anything new that has just come out um like in the AI space. Just scroll on my X feed. And so because this was a brand new session, we can

2:13:41 actually see if X is already signed in and if it's able to interact with things. Cool. So it just opened up X. Let me get my mouse out of here. You can see

2:13:52 clearly that this is already signed in as me and my feed is visible and now it is going ahead and it's starting to scroll on the X page. So just start to think about already how could you create

2:14:01 some sort of skills out of this and then schedule these as tasks. So maybe every morning if you wanted it to scroll through Instagram and things like that or scroll through your own profile or

2:14:10 mark things as red or unread, you could totally go ahead and do that no problem because you can set up the schedule tasks right in here and use browser use skills that you've already set up. So

2:14:19 you can see the two things that I already noticed was Grockbot and Grock 4.6. So this is an interesting case though because there is an API but using the browser use is actually cheaper. So

2:14:28 that's why in this case we decided to do it and test it out. Now what about computer use that you guys have also probably heard. These are basically the same thing. It's still going to use

2:14:36 vision. It's still going to use AI. But the difference here is that the browser is basically things that are on the internet or things that you would type into a browser like Chrome. But the

2:14:45 computer use are things that might be on your actual computer. So changing your computer settings, opening up desktop apps of things. So, let's go back into codeex. While we're waiting for this X

2:14:53 article to finish up. Let's see how this is going. It is still breaking down and turning it into an article. Let's see if we can do some computer use. So, one thing you will have to do though is

2:15:01 you'll have to go to plugins and you'll have to actually install computer use, which is just as simple as typing in computer use and hitting install like literally 2 seconds. And then, hey

2:15:09 Codex, I need you to use computer use to open up my desktop app for Glido. And I just want you to go to the settings and turn on the ability for it to allow and use like Bluetooth devices. So, just go

2:15:20 ahead and get that set up for me. And so, this is a little bit different, right? Because it's not going to open up your inapp browser. It's going to actually kind of show your screen being

2:15:28 taken over. It said, "I'll stop if Windows asks for an admin password or presents a security sensitive permission that it can't safely interpret." It shouldn't run into any of that, but it's

2:15:36 good that it sort of has that guardrail baked in. Interesting. Okay, so it said Bluetooth access is a privacy permission, so the computer use won't let me actually do that, but what it

2:15:45 said it's going to do is open up Glido. As you can see, I didn't open that up. It opened up Glido for me. And it doesn't look like on my recording you guys can actually see this right now,

2:15:52 but my screen kind of has like a big blue hue around the edges. So you will clearly know if Codeex is using your computer. And there you go. You can see that it

2:16:06 actually turned on the include Bluetooth devices button. And I think it got confused because I think at first it thought that I was messing with my actual computer settings rather than

2:16:14 like an inapp settings. So it thought that that was a security thing. But it looks like it figured that out on the way. And I think the computer use thing is pretty interesting because you can

2:16:21 remote into your codecs from your phone. So I can control all of these sessions on my phone. And what I could do is on my phone I could say, "Hey, computer use, go ahead and change my settings or

2:16:31 go ahead and do something on a desktop app that's on my PC that I'm actually not sitting in front of." And then you could control your computer from your phone. So anyways, we're back in the

2:16:40 session where it's creating that X article. You can see that it is actually creating some thumbnails for the actual article. So hopefully we're getting close to the point where it's going to

2:16:47 actually open up X and start to put stuff in there. But I'll check in with you guys when we get there. Okay, so it just decided to open up X. You can see that it's going to the drafts in my

2:16:56 signed X and then it's going to will go ahead and format this new article and then we'll take a look. It's clicking on the plus button. Man, I think it even just like resized the page a little bit.

2:17:05 It was getting a little picky. In goes the title and then it just pasted in the entire content of the article. And now it's uploading this thumbnail which I don't think looks very good. that part

2:17:14 of the skill has not been refined at all, but it's still cool that it was able to make that thumbnail and then go ahead and put it in there. So, I literally just have to make sure that it

2:17:22 looks okay and I like it and I can publish it. Now, I know what a lot of you guys are probably thinking when it comes to this browser stuff. It's like, what's the point, Nate? Like, where's

2:17:29 the value? And honestly, I understand where you're coming from. And that's why I wanted to show you guys this piece where it's like pretty much API will handle 95% of your automation things

2:17:37 that you need. And then a lot of the browser use opportunities might just be a script, but then when you need it, if you have this in your back pocket, it's good. And you also might think, okay,

2:17:45 cool, but like how much time does this really save? This probably saves me like maybe 5 to 10 minutes, like realistically. But what's cool about it is that I can basically have this

2:17:54 pipeline now where I have a bunch of these browser use skills. And so when I need to do something, I just open up codeex, I shoot off that skill, and I can go back to the other task that I was

2:18:01 working on. It kind of just like helps you do more without context switching as much. And as you start to stack more and more of these pipelines and combine them together, you can get some really cool

2:18:10 things to happen when you start to chain together these skills. And look, this one actually did it. It basically did the thing where I said it opened up the YouTube video. It took screenshots of

2:18:18 it, and now it's going ahead and it's planting some screenshots from my video in the article. So, if that doesn't prove the capabilities of this thing being intelligent, take screenshots,

2:18:26 look at the page, know where to click, know what to do, then I don't know what else will. Besides the fact that I've had this thing play chess, I've had it play games. I've had it set a world

2:18:33 record on Tetris, it can do a lot of cool things. Look at this. You can see that it even took the image and drugg it all the way down to where it actually fits in the

2:18:43 article. I haven't actually seen it do that before live. That was pretty cool. All right, I hope you guys are enjoying all of these use cases. This next one is personally one of my favorites. I used

2:18:51 Codex to edit the intro for this video. I use it to edit probably this course that you're watching. I use it a lot and my team also uses it a lot for courses and for ads and for creative content.

2:19:01 So, it is one of the things that truly helps us move really fast because editing is no longer the bottleneck when you can use a tool like Codeex, especially if you've never edited a

2:19:09 video before and you don't have video editing experience. So, let's hop into this next section about how you can use Codeex to help you edit some videos. Stop prompting Claude. Andre Garpathy

2:19:18 thinks there's a much better way to work with AI and his method has three layers. Layer one is the spec. Instead of giving Claude a task and hoping it understands you, work with it to create a detailed

2:19:26 spec first. Have Claude interview you about what you're actually trying to achieve. Then break the work into smaller checkpoints before it starts. So Nvidia just made building with AI

2:19:34 basically free. They're giving developers free API access to over 80 AI models. Here's how you use it. Go to Nvidia, choose the model you want, and generate an API key. You can choose from

2:19:43 models like Kimmy, GLM, Deepseek, and loads more. So, you want to start a real business, but you can't afford people for content, marketing, sales, and all the other types of work yet because you

2:19:51 can now install AI agents for free. One to help create content, one to help follow up with leads, and other types of agents to help handle the jobs that your first employees normally would do for

2:19:59 you. Isn't it crazy that with just my natural language, I can get all of these crazy motion graphics right here, and I can get all of these insane animations over here as well. I can have subtitles

2:20:08 appearing at the bottom, like you see. I can take my face and I can put it in the bottom right corner, or I can bring it up to the top left corner. And I can even play you guys clips of a bunch of

2:20:15 my different YouTube videos right here in this sort of 3D fashion. So by the end of this video, you will know exactly how to actually do this, even if you've never really used codec before or you've

2:20:24 never edited videos before. So I don't want to waste any of your time. Let's just get straight into this video. All right, so I hope you guys are excited to learn how to do this. What you're going

2:20:32 to need is you're going to need the Codex desktop app and you're going to need to install Hyperframes. But I'm going to show you guys all that. Before we go through the setup, I just wanted

2:20:38 to show you a few examples of how I've been using it and the skills that I'm going to give you guys for completely free. So this first example is just having it help me edit YouTube videos

2:20:45 because what it's able to do is it transcribes the audio and then it can cut out mistakes and stutters and things like that and you know dead space. So for example in this one I gave it two

2:20:54 different recordings. I gave it my face cam and I gave it my screen recording. And this was a 14-minute video that got cut down to 9 minutes and now it has like the background and it also has a

2:21:03 few cuts. So let me just show you what I mean by that. So here's the video. It's not live yet but this is how it starts off. Today I'm going over the 18 core Codex concepts that you actually need to

2:21:11 know in order to start using it right away and getting value from it. It doesn't matter if you're not technical at all or you've never used codeex before. By the end of this video you'll

2:21:18 understand exactly how this thing actually works so that you can start using it right away. So as you can see that first animation with the 18 cards coming into the screen that was

2:21:26 hyperframes as well as you know codecs here driving the actual edits, the rounded crops, the background, all of this. And so part one we're going to start off here is foundations. So we're

2:21:37 going to kick off with concept number one which is projects. Now a lot of you guys I'm assuming have used chatbt or claude in the now all of that logic was once again hyperframes. It basically

2:21:47 created the card that said you know number one projects. It then decided to also zoom me in here because it realized that for the next you know 10 to 15 seconds I might be talking about

2:21:56 something but I'm not showing anything on screen. So, I prompted it in this way and I gave it the skills in this way to understand that when there's nothing going on on the screen, maybe you do a

2:22:04 full screen, but then when it's time to do more of a tutorial, you go back into this view. So, anyways, throughout this whole video, that's basically what it does. It's able to jump between full

2:22:12 screen and this basically, you know, template or scene and then it also creates those cards in between each of the concepts. There is like a transition screen. All right, so that's one quick

2:22:21 example. Let's take a look at some short form content which I think is getting much much better. So, here I have it doing two reels. One is about Claude and one is about Nvidia. So, let me just

2:22:30 real quick play these two for you and then we'll break them down. Stop prompting Claude. Andre Karpathy thinks there's a much better way to work with AI. And his method has three layers.

2:22:38 Layer one is the spec. Instead of giving Claude a task and hoping it understands you, work with it to create a detailed spec first. Have Claude interview you about what you're actually trying to

2:22:46 achieve. Then break the work into smaller checkpoints before it starts. Layer two is the verifier. Before Claude does anything, define exactly what a good result looks like. Then give it

2:22:54 ways to actually check its own work. Whether that's another AI model or real data or tests that it can run on its own. And layer three is the environment. Build a workspace where Claude already

2:23:03 has your instructions, your knowledge, your skills, and your rules every single time you use it. So instead of becoming better at writing prompts, you're building an entire system around Claude

2:23:11 that makes it better every time you use it. I made a full YouTube video breaking down Karpathy's method and exactly how to build this yourself. So just comment method and I'll send it to you. So

2:23:19 Nvidia just made building with AI basically free. They're giving developers free API access to over 80 AI models. Here's how you use it. Go to Nvidia, choose the model you want, and

2:23:27 generate an API key. You can choose from models like Kimmy, GLM, Deepseek, and loads more. Then take that key and plug it directly into whatever you're building with, so you can experiment

2:23:35 with AI apps, automations, and other projects without immediately racking up API costs. And because they're OpenAI compatible APIs, you can use them with loads of existing AI workflows. If you

2:23:44 wanted to start building with AI, but didn't want to burn your own money testing different models, this makes it way easier. So just comment key and I'll send you the full setup. So I thought

2:23:51 that those outputs were really, really solid. You can see that what it does is it kind of starts off with like a visual hook. I guess we'll go to the clawed one first. We have the three missing pieces

2:23:59 and what it did here is it's kind of creating that open loop because it's saying, "Hey, yeah, you know, Andre Carpathy uses it in this way and it's creating this open loop of, oh wow, I

2:24:07 need to stay to the end of this short to see what all three of these are." Because as we go through, we can see we have number one. We could talk about little number one and then we go into

2:24:14 number two, the verifier. And then we get into number three finally at the end, which is about the environment. And so about every one or two seconds here, it's switching. You know, we've got some

2:24:23 scenes where it's like this with B-roll up top and me down here. It's got some where it's a full screen animation. It's got some where it's full screen me. And all these animations are sort of like

2:24:31 paper style. There's texture. It's 3D. They're always moving. So, we're never having one element be still. You've also noticed that it went out and it got all of this B-roll. It either created it or

2:24:40 it recorded it itself. And the subtitles, it was made by Hyperframes and Astra. The music, the sound effects, all of this is being synced to the actual reel for engagement. And pretty

2:24:51 much same thing with the Nvidia one. This one was shorter, but it's a similar style. This one has more green accents rather than orange because it's about Nvidia, but it has the same types of

2:24:59 animations. It has the same types of, you know, scenes, and it has the same way that it's fast-paced. It's always switching. We have music. We have real B-roll that it went out and it collected

2:25:07 on its own. And here's even more AI generated B-roll that it created on its own. I think that this is getting really, really good. And I've turned this into a skill. And like I said, all

2:25:16 of these are skills that come inside of one sort of student kit is what I'm calling it. So, if you want to get it all for free, just go into my free school community. The link for that is

2:25:23 down in the description. And then you're going to go to the classroom and you'll go to either AI skills or all YouTube resources. I have it linked in a bunch of spots, so it's really easy to find,

2:25:31 but you'll be able to find it right in here in the AI skills section. It'll be called like hyperframes student kit. Here's another quick example. This one is an AIS reel, and it made this one for

2:25:38 me in both um 16x9 as well as 9x6. But let me just real quick play the 9x6 version for you guys because it's kind of like a reel, but it's a little bit different cuz this one was meant to be

2:25:47 an ad. What if you could build the team before you hired it? So, you want to start a real business, but you can't afford people for content, marketing, sales, and all the other types of work

2:25:55 yet? Because you can now install AI agents for free. One to help create content, one to help follow up with leads, and other types of agents to help handle the jobs that your first

2:26:03 employees normally would do for you. It's like building your first AI powered team before you hire your first employee. And that doesn't eliminate every future hire. It actually gives you

2:26:11 the capabilities to start before you need them. At AIS Live, I'll show you how to build your own AI OS, a personalized system of agents, tools, and workflows that helps you operate a

2:26:20 real business. You'll discover the most important AI agents for starting a company, and how to install them for free. You'll also see how an AIOS can support content, marketing, sales,

2:26:29 operations, and even recruiting as your company needs to keep growing. So, if you want to start a highly profitable business without building a huge team, then click below for more details on AIS

2:26:37 Live. Now, it has similar elements, right? But what I want you to pay attention to here is all of the B-roll that it went and collected from me. It grabbed YouTube videos of mine. It went

2:26:45 ahead and it grabbed past AIS live, you know, segments. So here's one with me and Pat. We can go into here's like a VIP session. Here's a video of me like talking about my iOS and a second brain.

2:26:55 So it was able to use the context of what the ad was about. Here's a session with Hyper Agent and Alex. It used the context of what we were talking about and it actually put it inside so that

2:27:04 this doesn't just feel so much like theory. It feels more real. and even animated this logo here at the end, which I thought was just beautiful because it had to go to my Google Drive

2:27:12 to find the AIS Live logo. And I think that this turned out really, really nice. And I'll play one more for you guys real quick, which was a call to action that I gave Codeex. Real quick,

2:27:21 guys, just wanted to say if you're enjoying this video and you're enjoying the way that I teach, please consider subscribing and joining my free school community. The link for that is down in

2:27:28 the description. We're building a global community of AI builders and entrepreneurs and business owners. We're almost at 450,000 members. We're almost at half a million, which is super super

2:27:36 exciting. and it would just mean a lot to me if you joined. We have a lot of resources in here. We've got agent skills. We've got every YouTube video where I give away free resources. Like I

2:27:45 said, skills, guides, templates, docs, GitHub repos. I give it all away in here for completely free. We also have a 7-day challenge. We have a build your AIOS course. We've got a lot of fun

2:27:54 stuff going on in here. Events, certification programs, tons of cool things that we're building towards. So, if you're enjoying it, if you like the way I teach, please definitely check it

2:28:03 out. Hope to see you guys in there. Let's get back to the video. Now, what I liked about this is this one felt very clean, right? It wasn't super fast-paced, super energetic. There's low

2:28:10 music, but once again, it's proving to us that it can go grab screenshots, it can crop them, and it can use context about me and my business in order to make these feel more relevant. It

2:28:18 grabbed all of these different, you know, skills and docs, and it put them in here in a really, really nice way. It even went and screenshotted some of our courses from the community. It grabbed,

2:28:26 you know, event assets and pictures and videos, and it just made this call to action feel so much more real, so much more human. So, if you guys are getting excited by that, go ahead, like I said,

2:28:34 to my free school community and grab all of those skills to be able to replicate those sorts of outputs. But now, let's actually get started with building this. Like, how do you actually do it? And I'm

2:28:42 going to walk through a real example with you guys. If you guys remember the intro from this video, I'm going to build that out live right now with you guys. So, the first thing that I want

2:28:50 you to do is you're going to open up a new project. So, I would just open up a new folder in your desktop or something. And we'll just call this one for now. I'm going to call this um Hyperframes

2:29:00 demo. And that is the project that we're going to be working inside of. So I've got this folder on my desktop. Now I'm going to come into here on the lefth hand side and open up a new project. I'm

2:29:10 going to choose for this to be a local project. I'm going to give this a name. So Hyperframes demo. And then I'm going to choose that actual folder so that Codex can work inside of it. So that was

2:29:19 on my desktop and it was called Hyperframes demo. Cool. So there's the project. I'm going to select that and open this up. So, when you open up a new chat in this project, there's going to

2:29:28 be nothing in here. So, if I go over here and I go to my files, there's literally nothing here, but we're going to start to fill this up with things like video projects and different

2:29:35 renders and different files and things like that. So, the first thing that I want you to do is I want you to go to this GitHub repo, which I will link in the description. And this is called

2:29:43 Hyperframe. So, you're basically just going to copy this URL for the Hyperframes GitHub repo. You're going to open up Codex, paste that in there, and then basically just say, "Hey, Codex, I

2:29:53 need you to um basically pull in this hyperframes repo so that I can um use it to have you edit videos for me and stuff. I need you to make sure that all the dependencies are installed and

2:30:05 everything like that and, you know, grab in some skills or things that you need inside of this codeex project in order to actually make this stuff useful." So, that's the first step. You're going to

2:30:13 go ahead and shoot that off. Now, while this is getting set up, let me talk about the mindset and the idea of what we're actually doing here when we're editing videos with AI. So, there's a

2:30:21 few steps, right? And in this case, I'm taking the example of we're giving it some sort of footage and having it edit it and animate it and things like that, but it can also create things from

2:30:30 scratch. So, if you gave it um a rough outline and you wanted just motion graphics and like sizzle reels or like SAS demos, it could do that 100%. But that's not what I'm going to be focusing

2:30:39 on right now. You would just remove this first step. So basically in my mind, step one is you have footage and what you need to do is you need to transcribe it. And basically that means it needs to

2:30:48 find out all the text that you're saying in that clip and exactly when it happens, the exact second to the millisecond of when you say certain words so that it can sync it up with

2:30:58 different animations and different text like subtitles for example. So transcribing comes first. You can do this with a free local thing that it can set up like called whisper, but I

2:31:06 actually like to use 11 Labs for this. So I'm going to show you how to set that up as well. After you transcribe, then the next thing that we want to do is we probably want to cut. So, basically

2:31:15 having it cut out um you know, mistakes or five seconds of silences and things like that. So, that would be step two. Step three then in my mind is to plan the beats. That's kind of what it calls

2:31:26 our beats. So, beats are basically different scenes. So, for example, with this reel, like this install capability slide, this is one beat and then like this is one beat and this is one beat.

2:31:35 Every basically scene is a different beat. So now what we like to do is we like to have Astra, the model, read through the text, understand the intent, and start to plan out the beats of

2:31:45 what's actually going to go inside of this reel or inside of this video. And then what it's going to do is it's going to use its, you know, skills, and it's going to use hyperframes, and it's going

2:31:53 to use other tools like this in order to actually do the generation of the HTML. It's basically just creating HTML and it's animating it. And that's what really gives us all these 3D elements

2:32:02 and it gives us this text and it gives us what we are basically looking at and considering the motion graphics or the motion design. And by the way, part of planning the beats is like understanding

2:32:11 that there's going to be like, you know, music that's going to sync to it and sound effects and all of that is basically coming into um this element of planning. And the thing about it is we

2:32:21 have to have transcribed and cut all of that first in order to be able to do this properly. And then the last step here is basically honestly it's kind of a loop. So the last step is basically to

2:32:30 verify, meaning have the agent watch through all of this again. It will watch through it. It will take screenshots. It will look at the transcript again. And basically what we want to do is we're

2:32:41 going to do this loop of verifying and then going back into here and using the skills again and then verifying and then using the skills again. So we get in this loop of just going like this until

2:32:51 the agent is confident that everything is in bounds and everything syncs properly and everything feels how it should. So that's kind of the flow. Let's actually go get set up. So, we are

2:33:00 now inside of this project and it's still getting everything set up. So, what I want to show you what to do next is we want to do um 11 Labs to transcribe. So, you're going to go to a

2:33:10 new page. You're going to open up 11 Labs and create an account if you haven't. You can get started on this for like a five bucks a month plan, which it it'll last you quite a while. So, you're

2:33:20 going to come in here and then what you're going to do is you're going to go on the bottom left to developers and you're going to click on API keys right here and you're going to go ahead and

2:33:26 create a new key. I'm just going to call this um Astra demo. You can have this expire. You could just leave this at never. And then as far as restricting the key, if you want to say, okay, this

2:33:36 key is literally only going to transcribe stuff. That would be speech to text. You could give it access to speech to text. If you also want to give it access to like sound effects and

2:33:45 music generation and image and video generation, you can give it access to other things as well. But right now, I'm just going to start with just speech to text because all we really need for this

2:33:52 endpoint is the um transcription. So, I'm going to create this key. This gives us basically a password. So, don't share this with anybody. And I'm going to delete this as soon as I'm done filming.

2:34:00 So, don't even think about it. But, you're going to copy this to your clipboard. You're going to go into codeex and then we're going to come into your files right over here. And you can

2:34:07 see now it's starting to build out some folders and files and things like that. And what you need is you need a file called aenv which lets you store secrets and passwords like your 11labs API key.

2:34:17 So all you have to do is just you can chuck in a message even while it's working here. And look it actually opened up a local host which is kind of how we can see our video projects. This

2:34:24 is just you know it just lets us test them and look at them and even like move elements around and stuff which is really really cool. But I'm just going to close out of that because that was

2:34:31 just an example. But what I can do is say when you do the transcriptions I want you to use 11 labs. So, can you please create me availabs API key in there? And also make sure

2:34:43 that you include this in the agents.md that says whenever you know Nate wants you to transcribe a video, use 11 Labs for that. So, the agents.mmd is this file. This basically just explains how

2:34:54 the project works. So, right now, because there's not much information in there, it just has some very basic stuff from hyperframes. But as you start to learn more about the way you like to

2:35:02 edit videos and the way you like to store your projects, you'll basically want to come in here and update it with certain information. So right here you can see it says, "Okay, I will add that

2:35:09 rule into the agent.MD so that in the future I'll stop the whisper setup which is that local thing that I talked about." Now nothing's wrong with whisper, it's just slower. And so 11

2:35:18 Labs, yeah, you have to pay for it, but it's much quicker and I like that speed and it's not that expensive. So now ourv has a placeholder for the 11 Labs API key. So, what you're going to want to do

2:35:27 is just paste in your 11 Labs API key right inside of here. And by the way, if it's not letting you paste or type anything in here, um, which it should, but if it doesn't, then you can just go

2:35:36 into here and click on open in default app. And for me, that opens up in my notes. I can paste in the API key. Hit save. And then, as you can see, it's going to refresh live right inside of

2:35:45 here. All right. So, at this point, we have enough information in here to actually be able to use hyperframes. So, I'm not going to use my skills that I mentioned that I showed all of these

2:35:54 examples where I use my skills. I'm not going to use these right now because I just want to show you what this kind of looks like at its core. But don't forget that you can definitely grab those

2:36:01 skills and come in here and make these things better. And so one thing that you can do is you can give it a video file and just say, "Hey, edit this for me." You know, add motion graphics, make it

2:36:09 tell a story, blah blah blah. Or you can be very specific, which is what I'm going to do for this specific example. So let me pull up the video that we're going to be working with. It's right

2:36:18 here. It's called edit intro. So let me actually open this up and we'll give it a quick watch. This is just my natural language. I can get all these crazy motion graphics right here. And I get

2:36:24 all these insane animations over here as well. I can have subtitles up here at the bottom like you see. I can take my face. I can put in the bottom right corner. Okay. Anyway, so now you see

2:36:30 that I'm basically planning out what I want to happen as it's happening. So here's what I'm going to do. I'm going to grab the file path of my um video. I'm going to come back into here. I'm

2:36:42 going to do a /goal prompt for this, which basically just means that I'm setting a goal and it's going to keep pursuing until it hits that goal. I'm going to paste in the file path of that

2:36:50 video and I'm just going to start instructing what I want to happen when I want it to happen. But first, I have to set up the loop of transcribing, cutting, planning the beats, and then

2:37:00 it's going to go ahead and get to work. But instead of planning the beats, I'm kind of directing this a little bit more. So, this is kind of like plan/accept beats. But anyways, let me

2:37:08 show you what this is going to look like as I instruct this thing. So, I've just dropped in a video for you. This is going to be an intro. So, I need you to help me turn this into a

2:37:19 super premium and polished edited video. So, the first thing you need to do is I want you to transcribe the video because we need all of the beats and all the motion graphics to sync exactly when I

2:37:28 say them. And then I need you to cut out anything if there's mistakes or if there's um you know, silence. There shouldn't really be like a second or a half second of silence. I want this to

2:37:37 feel fast-paced. After you've done those two things, it's time to start actually animating this video. So, let me tell you what I'm looking for when it comes to the animations. So, I start off this

2:37:46 intro by saying something like, "Isn't it crazy that I can use just my natural language to get all these crazy motion graphics over here and all these crazy animations over here?" What I want to

2:37:54 happen then is I want this whole vibe of this video to be very, very like Apple style motion design. So, like liquid glass, clean animations, um, and they feel very professional. So, when I say

2:38:08 motion graphics over here, I believe that my hand was on it was pointing to the left side of the screen. So on the left third of the screen, I want them from top down, sort of like my hands are

2:38:20 creating these motion graphics. I want them to come into screen and just make them, you know, against like some sort of liquid glass card where we see three different types of motion graphics.

2:38:28 Maybe like cards and documents moving around or um laptop screens and charts and elements that are being like animated and things like that. Just make things over here that are impressive and

2:38:39 that tell a visual story. And then similarly when I go into the insane animations over here as well that's on the right side of the screen. So same thing I want them to come in from top

2:38:47 down like I'm drawing them in. I want these to be 3D. So I want these to be against some sort of um maybe little overlay in the back so that we can see them against the background just like a

2:38:58 subtle dark overlay. And these should be 3D. So um they should maybe look like they're spinning. They should have shadows. They should have depth. And these can just be 3D like letters or 3D

2:39:09 um animations and I don't know like symbols that are moving. Just make these, you know, subtle and professional, but they should certainly give some sort of like wow factor. And

2:39:17 for those two beats, I want them to stay on the screen while I'm talking about them. And then I want them to leave the screen or sort of like fade away at the same moment. So the motion graphics on

2:39:27 the left come in first, animations on the right come in second, and then they leave the screen at the same time. The next section is pretty simple. When I say I can have subtitles appearing at

2:39:37 the bottom as you see, I just want a super super clean um overlay at the bottom of the screen with subtitles, they shouldn't cover um my face. So, they should be sort of near the lower

2:39:47 third. And if you could just have the words like subtly highlighted with maybe like the words are white, but they're highlighted in blue when I'm actually saying them. So, make sure they're

2:39:55 synced exactly to the transcript when I'm saying them. Now, in the next section, I go on to say, you know, I can take my face and I can move it to the bottom right corner or um the top left.

2:40:05 And what I want you to do for that is I want you to take my face cam and sort of shrink it into like a um a vertical rounded crop. So, you kind of just bring in the left and right sides to keep me

2:40:17 centered. And you make it rounded crop with a drop shadow. And you put it against this background, which is the full screen background. And so here what I'm doing is I just pasted in this

2:40:25 picture which I like to use as uh my background in some YouTube videos. And I want you to make that background 50% opacity so it's a little bit darker. And then for the next part when I say I can

2:40:35 play you guys these um clips from my YouTube videos, I want you to go to my YouTube video directory locally which I'll give you the file path too. So I will just copy in this file path real

2:40:45 quick right there. What I want you to do is just grab a few of my most recent videos, maybe like four or six, and I want you to sort of fan them out, like animate the process of you fanning them

2:40:54 out, and have these be 3D with like a little bit of a drop shadow, and they're sort of like rotating and swiveling back and forth just to create a really cool effect. But I want them to actually be

2:41:03 playing throughout. I don't want them just be static images of the videos. I want you to just play, you know, all of them are sort of actually playing at the same time. And then to end off the

2:41:12 video, what I want you to do is basically against that background I gave you, if you could bring my main face cam, sort of align it on the left side of the screen, but make it bigger so it

2:41:21 kind of touches the top and bottom of the screen, but we should still see the rounded crop and the drop shadow. And then on the right side of the screen, we can just have some other text and other

2:41:28 motion graphics that tell a story based on what I'm saying. So like um you know, you don't have to be technical, you don't have to have used codecs before, you don't have to have a video editing

2:41:37 background. if you could just make little icons and animate them in a fun way to tell that story. And then at the end, you know, it says like, "Hey, I don't want to waste any time. Let's just

2:41:45 get straight straight into the video." Do the exact same thing there. And that's going to pretty much be the end of this intro. After you have then done all of that, I just want you to make

2:41:53 sure it looks good. I want you to sort of like verify it by taking some screenshots and looking through, making sure that the beats actually sync to the text that I'm reading out and make sure

2:42:03 that um everything's in bounds and everything looks professional and it's giving off the quality that you want it to. I'm not looking for a PC or a version one. I'm looking for a finished

2:42:12 product ready to go here. Out of breath a little bit. So, that was a massive prompt, right? That was me being super super specific about what I'm actually looking for here. And that's kind of

2:42:20 what this looks like to start. The way that I got to all of these skills that I now use for YouTube videos, lessons, reals, ads, is because I did that so many times. And now if you get an output

2:42:31 here that you really, really like, you say, "Cool, that was awesome. Turn that into a skill." And then next time, you don't have to give such a big prompt. You basically just say, "Hey, edit this

2:42:38 video. Use this skill." And then when it comes back, if you have feedback, say, "Hey, I didn't like this. Make sure you do this next time." And then update the skill, too. So, what I just did is

2:42:47 really the hardest it'll ever be. And then it's always just about iterating on those skills. So, I'm just going to go ahead and let this cook a little bit. I will check in with you guys when this

2:42:54 one comes back with the first version. And hopefully it's what you saw in the beginning of this video. Okay. So, I have not yet watched this, but you can see that this went from the original

2:43:03 being what was this? A little over a minute, a minute and 5 seconds to now being cut down to 28 seconds. Let's go ahead and give this a watch and see how this looks. Isn't it crazy that with

2:43:13 just my natural language, I can get all of these crazy motion graphics right here? And I can get all of these insane animations over here as well. I can have subtitles appearing at the bottom like

2:43:21 you see. I can take my face and I can put it in the bottom right corner or I can bring it up to the top left corner. I can even play you guys clips of a bunch of my different YouTube videos

2:43:29 right here in this sort of 3D fashion. So, by the end of this video, you will know exactly how to actually do this even if you've never really used codecs before or you've never edited videos

2:43:38 before. So, I don't want to waste any of your time. Let's just get straight into this video. Okay, not bad. Not bad for a first pass. Um, you can see how my very specific prompting obviously benefited

2:43:48 us here. I think that these look good, right? Like we get some 3D animations, we get some motion graphics. I like it. I think that's a good way to start. I do think maybe we need something more in

2:43:56 the first two seconds. Yeah, I think we should add something right there in the first two seconds just to make it more engaging. But the other thing I noticed is right here. It's playing videos,

2:44:03 which is good, but they're like inter overlapping with each other, right? So they it's not realistic. What we need to do is make this feel more legit. And also, I want to make this text bigger. I

2:44:12 don't think that text is big enough. And I think this looks fine. This is all simple. Nice little animation as well. Basically, that's my feedback. So, let us shoot off prompt number two. I'm

2:44:21 going to once again do another slash goal prompt. Awesome. So, I mean, that was 18 minutes. That was a really good first pass, but let's do another version here. What I want is a couple things.

2:44:32 First of all, in the first two seconds, I want you to get creative here. Do something that feels on brand to make the first two seconds more engaging. I'm not sure I want like a

2:44:43 full screen cutaway and I don't want anything that's going to lay on top of my face, but you know, I start off the video by like, isn't it crazy that with just my natural language, some sort of

2:44:54 animation, some sort of motion graphic, make that a little bit more emotional? Just do something right away to sort of like hook the viewer in. Now, besides that, the only other changes I need are

2:45:04 in the part where you show my YouTube videos, you show the four. I like how these look like they're 3D and they have depth, but if you could make this more realistic because they clearly are like

2:45:16 cutting through each other almost like it's not realistic. We need this to be more like the physics of it need to feel better. So, if you could just maybe make them thicker and make sure that they're

2:45:26 they're fanned out behind each other rather than like cutting through each other, that would be great. And then the text on that screen, it says like your videos or something. um if you could

2:45:35 just make that larger because that text was pretty small. I want it to be bigger and easier to read. But besides that, everything else I thought looked really good and it matched the transcript

2:45:43 really well also. So make those changes please and then let me know when you're done. So that is prompt number two going off as another goal. And you can see that this is the actual hyperframe

2:45:52 studio that it created for us. Um this is at 107. Let's just reset that. And what's cool is I can literally move things in here. So I can actually take this element and I could increase the

2:46:00 text size over here. I think font this is where I would do it. I could change this to what, like 80, and see how that looks. So, it would reload and now I can just sort of manipulate this stuff right

2:46:10 from here, which is pretty cool. You know, you can move stuff around and um you can see there's even different elements for the actual like 3D scene and the actual video inside. So, I'm not

2:46:18 going to mess with that right now. I honestly don't use that interface too much. But if you're editing a really long video, let's say it's like a 10-minute video, and you just want to be

2:46:24 able to like tweak one thing really subtly, it's probably easier to come in here, make a little tweak, like just move this over and then export it rather than having to do the prompt layer

2:46:34 again. So, it's really nice that Hyperframes gives you this local host environment as well. But for something as simple as what we're doing, I just wanted to shoot off another natural

2:46:42 language prompt. So, I will check in with you guys once this one has finished up. All right, so that took 10 minutes now and V2 has come back. Let me go ahead and open this up full screen and

2:46:49 we'll give it a watch. Isn't it crazy that with just my natural language, I can get all of these crazy motion graphics right here and I can get all of these insane animations over here as

2:46:57 well. I can have subtitles appearing at the bottom like you see. I can take my face and I can put it in the bottom right corner or I can bring it up to the top left corner and I can even play you

2:47:05 guys clips of a bunch of my different YouTube videos right here in this sort of 3D fashion. Okay, so this was much better. The text got larger and this feels more 3D and it also just feels

2:47:14 like, you know, it just looks better. And so right here, this is pretty much the only section where I didn't tell it exactly what I wanted. And this is what it came up with. I don't honestly love

2:47:22 it. But for now, I'm just going to keep it in because I told it I didn't want a full screen takeaway, which would have meant it created its own animation to be the full screen. But actually, let's

2:47:30 just see what that would look like if I just shoot off that quick prompts real quick and say, "Awesome. Now, can you create me one more version where in the first two seconds before you cut into

2:47:39 like the motion graphics and everything, let's see what this would look like if you did a full screen takeaway." So, it would be the full screen is animated um against some sort of, you know,

2:47:47 background that feels on brand with the rest of the intro and maybe the text is sort of comes in in an animated way that says, "Isn't it crazy?" So, anyways, just going to shoot that off. This

2:47:57 really shouldn't take long at all, but just to show you guys that it can also do things like, you know, a full screen cutaway and animate it in that way, too. So, I'll show you guys that in just a

2:48:05 sec. All right, so that one came back and it's done. And I have a feeling that it's basically just going to be super simple right here. I'll play it from this interface. Isn't it crazy that with

2:48:12 just my natural language, I can get all of these crazy motion graphics right here and I can get all of these Cool. So, that is looking good. We can see basically all this did was just add some

2:48:20 very simple animated text that came in here. It had a little bit of layering as well with these little liquid glass looking cards and it kept the background consistent to what we actually used

2:48:28 later over here. So, I thought that's pretty good. Anyways, guys, at this point, we've set up the project. You've set up something to transcribe if you decided to use 11 Labs and you

2:48:36 understand now the loop that we actually go through in order to do all this. I didn't dive too much into this actual interface and how you control it, but like I said, it's very, very intuitive

2:48:43 once you get in here and it's a nice touch by Hyperframes. So, don't forget if you want to get this student kit that has all of the skills that you need to build these sorts of reels and videos,

2:48:51 you can download this for completely free by going to my free school community. The link is in the description. You'll go to classroom and click on AI skills and it will be in

2:48:58 there. Okay. So, now that you have a really good understanding, at least I assume, of how Codex works when you're in it and you're sitting at your desk and you want to drive automations and

2:49:05 drive, you know, skills and different sessions like that. But what if you want to build automations that run without you being there at all? And I know we've talked a little bit about scheduled

2:49:13 automations, but what if you want to build things that live somewhere else? I don't know if you guys have ever used Nitn before or make, but what was cool about that is you could just deploy

2:49:20 those automations in, you know, make server or NEN server or something like that. But what Codex lets you do is it lets you build automations in code. I know we don't want to look at code, I

2:49:29 understand, but it lets you build automations in code and then you can just deploy those automations somewhere else. And now you don't have to have your Codex desktop app open or you don't

2:49:38 have to have your laptop turned on in order for those automations to keep running all the time. So we're going to talk about how you can use Codex to build automations and then host those

2:49:45 automations in something called trigger.dev. It is a lot simpler than it sounds. So let's just hop into this next video. Codex lets you build automations right inside of it using scheduled

2:49:53 tasks. However, because these scheduled tasks basically just send messages right into an actual chat thread, what that means is this eats at your actual weekly usage limit. So, it doesn't really make

2:50:03 sense to build a bunch of your personal and company automations inside of Codeex. A lot of the most useful automations can run somewhere completely independently of codecs, but we can use

2:50:11 codecs to help us build them. So, that's exactly what I'm going to show you guys today. Even if you have no technical background at all, by the end of this video, you'll see how to build scheduled

2:50:18 automations or ones that are triggered on some sort of event that live outside of codec so they don't eat away at your subscription. So, I don't want to waste any time today. Let's just get straight

2:50:26 into this one. So, like I said, when you have an automation running inside of Codeex all the time, it's eating away at your subscription limit. So, what I'm going to show you guys today is how we

2:50:34 actually build our automations with Codeex and then instead of actually hosting them there, we're going to host them in something called trigger.dev. Because when we actually build the

2:50:42 automation, what happens is we can actually just turn that into code. So instead of having, you know, our schedule task run a skill on a certain cadence, it basically just turns that

2:50:51 skill into code that can actually be executed and then that code is just given from codeex to GitHub and then GitHub basically says, "Okay, trigger.dev, you're supposed to run this

2:51:00 code, you know, every hour or you're supposed to run this code whenever someone submits that form submission." So if that sounds overwhelming or scary at all, don't worry about it. It's going

2:51:08 to be super simple. Just think of it as the idea that you're giving your skills to something else to run it so that it doesn't eat your codeex subscription. So that way instead of your usage limit

2:51:16 looking like this, it will look more like this one down here, which is green and a lot more full. So what you'll need for this video is a CEX subscription. You'll need a trigger.dev account and a

2:51:24 GitHub account. That's what we're going to be using. Okay. So there are three different types of automations that I'm going to show you guys today. And I'm also going to explain the difference. So

2:51:31 it's going to be a scheduled automation. We're going to have a web hook automation. And then we're going to have a codeex SDK automation. So let's start with just number one. This is a

2:51:39 scheduled automation, which basically just means we're going to have trigger.dev dev execute our code that we write on a schedule every 30 minutes 6:00 a.m. on Mondays however we want to

2:51:49 set up that schedule very similar to the way that the schedule tasks inside of codeex actually work. So the first step of planning out the automation is the planning. It means we just want to kind

2:51:58 of brainstorm with um Astra in this case to help us make sure that it fully understands what we want built and to make sure that we're not, you know, missing anything. We're not we're

2:52:08 thinking through all the different scenarios. And yes, this example is going to be very simple, but it's just to show you how this actually works. So, I want to build an automation, and I'm

2:52:16 going to put this on trigger.dev to actually be hosted there. But you need to help me plan this out and then actually build it. So, the first step is just the planning. I want this to be a

2:52:24 morning routine automation. So, this is going to go off, let's say, 6 a.m., and it's going to look at my calendar for the day, and it's going to just give me a brief. So, it will have to check my

2:52:32 Google calendar and it will just look at the events I have and then if there's any additional research or anything like that that it could do to be more helpful. It can just sort of help me

2:52:40 prep for the day. Is there anything else important that I'm not thinking of and do you understand how this will actually be built out? And what's cool is it will actually be able to use some of the

2:52:50 context that's already in your projects if you are working inside of a project because you can see the first thing it said is I'm going to check your existing morning routine workflow so that I can

2:52:58 actually understand a little bit better how you want this built. So, trigger.dev hosts something called TypeScript, which is just basically a coding language. If you used something like modal, then you

2:53:07 would be writing it in Python. So, don't worry about this word. It just is basically the language of the code that we're writing, but don't worry about it. So, it's explaining now how is this

2:53:14 going to work. So, the morning trigger would go off at 6:00 a.m. America Chicago time. And then we would look at Google calendar and we would read today's events. We would read their

2:53:21 descriptions and their times and their attendees and locations. And then it would decide whether the events need prep. So, if it's with someone new, we have this stuff that could be helpful.

2:53:29 if it's a podcast, we could have this stuff. So, it's going to do a little bit of AI reasoning here, and then it's going to use AI once again to write the short brief. And then after that, it

2:53:37 will send it to me, but I didn't actually tell it where. So, it said, "Where do you want this to be sent? Um, what's the timing? What's the calendar scope?" It's asking me some questions

2:53:45 here. So, all of these are good thoughts. I want this to actually be sent to me in ClickUp. I want this to be sent to me as a DM to Nate Herk, so to myself and ClickUp. And you will use the

2:53:55 Upit AI account to actually send this. As far as timing, we want this to be 6:00 a.m. And let's just have it be on weekdays. It doesn't really matter to me how long it takes. If it if I get it at

2:54:05 6:05 or 6:10, that's fine. The calendar scope, we're going to use my main work calendar. And research boundaries, let's just do public web research for now. Later, we could always work in other

2:54:16 things like meeting notes and clickup and Google Drive and things like that. As far as the research, there's really no reason why you should be spending more than, you know, 25 cents for

2:54:23 research at this point, but we can increase that later. Right now, let's just assume that the calendar is locked at 6 a.m. So, don't worry about checking in later. So, now that I've given it

2:54:31 some more information, I will go ahead and send that off as well. Now, what I wanted you guys to notice here is figuring out if the automation is one of these two things, which are it's either

2:54:42 a deterministic automation or it's a non-deterministic automation. And so, what this means is basically is it a predictable process or is it unpredictable? And usually the

2:54:54 unpredictable automations when you get into this territory over here, this is when we really need to go for, you know, a full agentic harness, a full codeex routine because these are very

2:55:03 noneterministic. They're unpredictable. And in this specific example, this was pretty simple. We basically have, you know, a 6 a.m. trigger. So we have 6 a.m. it goes off. Then what happens next

2:55:13 is we will read calendar. And so that is the next step. And so far there's not even any AI involved. where we start to get AI is in this next step when we sort of do like the decision-m on do we need

2:55:25 research and you know what would we do so this is research and this would be an AI step so I would fill in this with green just to indicate that this is in fact AI the other thing that we would

2:55:35 need AI for is drafting the message because you can't really draft a message um with code unless you were just using explicit placeholders and it's not as flexible and then the final step is to

2:55:45 send that message to ClickUp so what I wanted you guys to see here is this is a very deterministic automation because it goes 1 2 3 4 5 oops five and it happens in that order every single time. So that

2:55:58 is a deterministic flow. What's not deterministic in here are these specific AI steps. So step number three and step number four are nondeterministic because the research is going to be different

2:56:08 every time based on the input that goes in. And then the draft message is going to be nondeterministic because it's different every time based on the research that goes in. So these are

2:56:16 important things to think about because the more deterministic your automation gets, the more of a waste it would be to use codeex to host that. There's no reason to use the full Astra agentic

2:56:26 loop inside of codeex if your automation looks like this. 90% of business use cases or personal productivity use cases are more like this. They're more of the deterministic automations that help you

2:56:35 out and really save time. So anyways, just wanted to call that out. Hopefully you're learning some new stuff in this video. All right guys, real quick break for a message from today's sponsor,

2:56:44 Hyper Agent. Now, I know that a lot of you guys are running AI agencies and a very tedious part is teaching the same thing over and over again to every new client's agents because every client

2:56:54 wants it done a certain way. So, Hyper Agent handles that with skills and memories because you can teach an agent a skill once like your due diligence framework or your press release

2:57:03 framework or format or you know some API that you always end up wiring in and it will just keep it in its memory and then it will build up memories as it works on things like the client's preferences,

2:57:14 their data sources and how they want things formatted. And then you can watch the whole fleet from a command center that runs eval rubrics and it can do AB tests so that you can see which agents

2:57:23 are actually getting better and what each client costs you to actually run. They can also sit in your Slack channels and they can jump in when something needs doing instead of just waiting

2:57:32 around for you to ask. So if that sounds like something that could be useful for your agency, then check out Hyper Agent through the link in the description and you'll get some free credits to start.

2:57:40 So thanks again to Hyper Agent for sponsoring this part of the video. But now let's get back to it. So anyways, Codex comes back and says, "Okay, cool. That gives us a clear plan. Here's

2:57:47 exactly what we will do. Here's the budget. Here's the research. Here's all the stuff that's how it's going to work. And now, let's actually go ahead and start building it. It also calls out

2:57:54 this hosting detail, which is that Trigger Dev's current free plan scheduling window can delay a 6 a.m. start by up to an hour. So, if we really need this right at 6 a.m., you might

2:58:02 have to upgrade your Trigger.dev plan. But, as far as getting started, you can start on the free plan. All right. So, sounds like you have a good idea of how this works. What do you need from me?

2:58:10 Are there any different API keys that I need to get, or do you already have access to all of these keys? So that's really the next step to be thinking about because if you have been building

2:58:17 inside of codeex and you're using a bunch of these plugins, so you've connected to all of your accounts inside of these codeex plugins, those will not transfer over to trigger.dev, which is

2:58:27 why I really like to use plugins not very much. I like to use thev file and I like to put all my API keys in there so that if I ever need to move them over to cloud code or if I need to move them

2:58:37 over to trigger.dev, I already have them in one spot and I can just copy and paste them. And I will show you guys how that works later. But yeah, now that we've done sort of the hard part, the

2:58:44 heavy lifting of actually planning out what we want, the implementation is the easy part because now we let Codeex actually go ahead and build this out for us because as you can see, it says,

2:58:52 okay, it has access to Google Calendar, ClickUp, OpenAI, Perplexity, and we will obviously set up trigger.dev, but basically everything we need to move this over once Codex has built it is

2:59:02 already here. If you don't yet have your API keys for Google Calendar or ClickUp, you would just say, "Hey, can you help me set that up with Google Calendar, with ClickUp?" and it will tell you

2:59:10 exactly where to go and what to click on and what to get. So getting API keys and getting that set up is not really technical. It's not really a technical lift anymore. It used to be a little

2:59:17 more intimidating in the old days, but now it's so easy. So if you don't have API keys, just ask Codex to help you find where those are and just go grab them. But now we're ready to start

2:59:26 building. So I'm just going to tell it, yep, go ahead and start building. And I'll check in with you guys when we're ready to move this over to trigger.dev. But the one thing I'll say is before we

2:59:32 move it into trigger.dev, we want to have Codex do as much verification on it as possible. So, if it's able to test it out and basically fix it before we move it over, then that's good. You know,

2:59:41 we've talked a lot about giving agents a way to verify their own work and telling them to verify their own work so that you're not getting their first attempt. So, go ahead and build this out for me.

2:59:49 Make sure you feel confident in how it works and you've validated that it will work before you actually tell me we're going to move it into trigger.dev and then once you're done, we will move

2:59:57 everything over and host it. Now, you can see what it's doing is it's building it out and it's testing it. You can see it's using test-driven development. It is sending DMs. It is looking at how

3:00:06 much all this costs and it's basically optimizing for what I actually said. It even made sure that the DM was the correct conversation. And the reason why this whole deterministic or

3:00:13 nondeterministic thing is important to understand is because yes, you could easily do this as a routine inside of codeex, but I actually did this before where I had a routine that was very

3:00:22 simple like this and after about a month, it started to just go rogue. It just randomly started to send to different DM channels and it started to send to like the team channel and stuff

3:00:31 like that. And the reason for that is because it was an agent on the back end. the agent was just interpreting the message different every time and it just acted differently. But what we're

3:00:38 actually building here is we're building a script so it doesn't have the option to act differently. We're literally saying, "Hey, this is what runs and it's goes the same way every time." So in

3:00:46 this code, we're basically like hard- coding the direct channel with the channel ID and the message ID and everything that gets sent rather than giving the agent the full connection to

3:00:56 ClickUp and saying, "Hey, just send it to Nate." Anyways, this has just finished up. So you can see, whoops, that it was built and validated locally. And we should have a test brief in

3:01:04 ClickUp. So let me check that real quick. Awesome. So you can see right here that I do have this DM where I see my Tuesday morning briefing. I have my day right here. Every single one of

3:01:12 these things has a link to my calendar as well. If I click on this, for example, it pulls up the actual event in my calendar. And then it also comes back with, hey, here's some useful prep. And

3:01:21 here are some things to notice. So we know that the ClickUp connection works. We know that the AI is working on the background. And we know that the Google calendar connection is working as well.

3:01:30 And it says that everything has been built and validated. And this costed about 1.33 cents. So us giving it the gate of, you know, maxing each run at 25 cents seems pretty realistic. Anyways,

3:01:40 now we have to deploy to trigger.dev and validate one run there before we turn on the schedule. And the cool thing about this is yes, you're going to be able to follow this tutorial so you know exactly

3:01:48 how to connect to trigger.dev, but if you didn't, so how exactly do I get this automation from the code that you built into trigger.dev? And while it's doing this, let's go ahead and make sure we

3:01:57 have a trigger. So, head over to trigger.dev. Go ahead and create an account. Like I said, you can get started for free. Now that I'm in here, you can see I'm on a free plan in this

3:02:05 account and I just created a new project up here called YouTube. And this is what it will look like in order for you to sort of like get everything set up. There's nothing yet in here. So, this is

3:02:13 how it gets really cool because we're able to basically just connect codeex to trigger.dev and it can basically talk to trigger.dev in order to do all of this. Now, you can also have it connect to

3:02:22 GitHub, which is something that I do recommend because that way GitHub connects to trigger.dev and the three of them talk together. And the reason why we like to use GitHub is because then

3:02:30 you can have like different version control and you can have other people more easily contribute to the automation if you need someone else to change it or something like that. And Codex can

3:02:38 connect to trigger.dev and GitHub via the command line tool. So, that's what I would probably say next is let's get connected to GitHub and trigger.dev dev via the command line and then we can

3:02:49 just push our code to GitHub and then trigger.dev will sync with that code. And what this will do is it will use the CLI and it will prompt you to just authenticate in. So it will do one of

3:02:57 those things where it opens up the browser, you sign in with GitHub, you sign in with trigger.dev, and then you come back into Codex and it says boom, got it, I'm connected. So here is the

3:03:06 trigger.dev authorization. I just have to go ahead and authorize right here. And then it says return to your terminal to continue. So Codex should have gotten that now. And you can see that because

3:03:15 my codeex is already authenticated to GitHub, I didn't have to do that. But it would be the same exact flow if you haven't done that yet. So now I want you to create a new private repo for this

3:03:24 automation in my GitHub account. And I'm just going to shoot off this message to steer codeex in the right direction now that we know we're connected to these two tools. And you can see that it says

3:03:32 the API keys, the calendar snapshots, and anything sensitive are excluded from Git. Even though we're creating it as a private repo, it gets completely excluded. And that's the whole point of

3:03:41 the env. So, what we'll have to do in trigger.dev is we will have to manually move those over. And look at this. Because Codeex's browser use is so good, it also pulled up my trigger. account in

3:03:51 this browser. So, if there's anything that you're confused about navigating the trigger interface, you can have it help you out by just clicking around right in there. But, it says that we did

3:03:59 create this GitHub repo for the morning brief. So, basically, this repo, just think of it like a Google Drive. It holds the actual code and the rules for how this automation runs. And

3:04:08 trigger.dev will grab this and actually host it. So, it's super simple. But also if you make different versions you will update it here. So you could roll back to previous versions if you want and

3:04:17 that way if a team member later wants to take over this automation all of the details are here and they can contribute to this project. Okay. So take a look at this. It actually created its own

3:04:25 project in my trigger. So it said in morning brief and settings and git is the GitHub repo connected. So basically what that's saying is okay it got into my trigger.dev. You guys saw me create a

3:04:35 new project called YouTube but it decided okay I'm going to create my own project. So it made one called morning brief. And now what we need to do is we need to see if this trigger.dev project

3:04:43 is connected to our actual GitHub codebase called morning brief so that they can sync and talk to each other. So that's what it's asking me to verify. It said to go to the settings in here, but

3:04:52 you can see that it says GitHub isn't connected. So it's actually not syncing to anything at all. So what I need to do maybe is install the GitHub app. But let's see. I'm just going to say I don't

3:05:01 see any deployments. I don't see any integrations. When I go into trigger.dev, it tells me that GitHub isn't yet connected. So that's what I see. But this is really cool, guys. You

3:05:11 can see how proactive this thing is. It actually just went ahead and started doing all of the connections rather than telling me what to do, which is pretty cool. But every once in a while, it hits

3:05:19 a snag and it will just tell you, hey, can you check this? Can you check this? But hopefully with this context, it's going to be able to help out with that now. And now you can see it's going to

3:05:26 open it up to actually do it itself. So you can see the mouse right here moving. This isn't me. This is Codeex using its browser use to check on all this for us. Okay. So you can see that this says

3:05:35 connected and verified and morning delivery remains disabled until the full hosted delivery test passes. So that's good. And it actually moved it back under the personal account where the

3:05:44 GitHub connection already exists and ClickUp still uses UPAI. Now let's go in there real quick so I can show you guys something. If I go back into the other project and it actually did it in a

3:05:54 different org. So if I go back to my personal org, this is where we should see the morning brief. There we go. So I was just looking at the completely wrong one. And if I go to tasks, this is where

3:06:02 we can actually see the interface of our weekday runs. And what's cool is it did two different tasks. It did the scheduled one. So this one you can see is the one that goes off at 6 a.m.

3:06:12 Monday through Friday. And then what actually happens when this runs is it calls this other I guess runner. So it calls this and this is the one that has the actual process in it. So it does it

3:06:24 in two parts. And that's pretty cool because as you create these longer, more complex workflows, they call individual tasks and pieces that codeex will build out for you and it just helps with some

3:06:33 visibility. So anyways, what we can do here is if I go into the schedule and I hit test schedule, this basically starts running this. So I'll hit run test and now this runs in production. And then

3:06:42 what you can see is that it actually calls this other little task. And now this thing is running. It's showing us how long this is taking. It's going to show us all these little different

3:06:50 details so that if this errors, we could copy all this data, give it to Codex and say, "Hey, this is what happened. Help me fix this." So, you can see that this already finished and we can look through

3:06:59 the payload. We can look at the status which came through as disabled. And you'll notice that I didn't get a new DM. So, we have to figure out why this didn't seem to work. My first thought

3:07:08 would be, okay, well, do we have all of our API keys and all of our secrets in here? And if I go down to deployments, this is where I can click on environment variables. And you can see that we have

3:07:16 a bunch of stuff in here that got added today. Today's September 15th. So Codeex actually took our ENV and via the CLI put everything in here, which is awesome. So we should have our ClickUp

3:07:26 ids. We have our workspace ID, client ID, all of this stuff should be working in here. So what I would do is I would say, okay, so I just ran a test run inside of trigger.dev, but I didn't get

3:07:37 a ClickUp message. I didn't get anything. Can you just check and see what happened there? So you guys remember how it came through and said status equals disabled. It's because it

3:07:46 had morning brief enabled as false. So if I go back into our trigger.dev and I go to our environment variables, you can see right here, morning brief, it says false. So we would basically have to

3:07:56 come in here, edit this, and we would have to change this to true. Hit save. And now if we go back, let's see if this fixes the issue. So I'm going to go into the morning brief weekdays and hit test

3:08:06 and go ahead and run that. And there we go. This looks a little bit different now because you can see it called on our different runner. We can see what's actually going on. And we can see this

3:08:13 thing actually in real time sort of like playing out. We got a little bit of a green message here. We get output equals delivered. Let's open up my ClickUp. See, was this a new run? Okay. Well,

3:08:24 unfortunately, this still doesn't look like it was a new run. So, once again, we have to go figure out why this didn't work yet. Because if you guys remember, it worked when Codex was testing it

3:08:31 locally, but it didn't work yet in production. So, okay, that run seemed to work. It showed that we actually got the output, but I didn't get any DM in ClickUp. So, take a look at that run and

3:08:44 help me figure out how do we fix this. Okay, that's really interesting. What happened was the run found today's earlier test message. So, it skipped sending a duplicate because it would

3:08:52 have been essentially the exact same message. So, it seems like that run didn't actually fail. It was more so done by design. You can see the existing message was sent at 12:01 and we just

3:09:01 tried to run a new one at 12:33, so it didn't get sent. That's why it's always important that when you're having AI build you automations and build you code, you don't have to know what every

3:09:09 single line is doing. But you do have to test the stuff out to figure out how it works and why it does what it does. I mean, look how powerful this is. It's basically able to deploy everything,

3:09:17 debug everything, move around in the interface, and then just test it all for me. And we're just using our natural language, and we're using our intent to drive all of this. This is so much

3:09:27 quicker than the way I used to have to build automations. it worked in this variable that shows that if it was already delivered and it got triggered twice for some reason to not send a

3:09:35 duplicate message. So duplicate prevention remains intact. I think it did a really good job there. But I do want to prove to you guys this works. So I'm telling it to override that just so

3:09:43 we can see that it works as expected. And then we'll move on to the next automation. There we go. We can now see that we got the hosted proof test for the morning brief. And now we can feel

3:09:52 better about turning this thing on inside of trigger.dev and knowing that it will actually have all the connections set up because we've seen it in here. We've tested it. We know that

3:10:00 all of the environment variables are also moved over. So, we are all good to go. All right. So, we took care of the scheduled type of automation. Let's look at one that is triggered by something.

3:10:09 So, typically this is triggered by a web hook. So, we're going to go back into Codex and we're just going to start having it do two things for us. It needs to build sort of like a very simple

3:10:17 front end that will actually be our trigger for this specific example. And then we'll build the back end, too. So, I'm actually going to try to do this in one fell swoop. I'm going to do a slash

3:10:25 goal. And here we go. So now I want to build a web hook triggered automation inside of trigger.dev. So this is kind of a two-parter. The first part is we need to use or sorry I need you to help

3:10:38 me build just a very simple local host that is sort of like a form submission that you might see on a website. Collects information like the name, the email and um team size and what they're

3:10:48 looking for. So, a simple landing page, a simple form that we can fill out, and then when the user hits submit on that, that's what I want to trigger the actual automation on the back end that we're

3:10:58 going to build and put in trigger.dev. So, basically what I want is for um that automation to just send me a ClickUp DM. So, same DM as before, and I want this one to just say, "Hey, you got a new

3:11:09 form submission. It's for this person. Here's what they asked for, and here is how I recommend you reach out to them." So, sort of like just creating a draft. So, I'm going to shoot that off and it

3:11:17 might have some questions for us. But this is a very simple use case just to show you how it works with a web hook. But this is still a very much a deterministic automation. Even though it

3:11:26 is AI, it is still deterministic. So, let me show you real quick what this one looks like. This one is a web hook trigger. So, this is just going to be a form submission. That is basically what

3:11:34 kicks this thing off over here. From the form submission, what happens is it's going to actually read the form and it's basically just going to have to do one thing, which is create the actual um

3:11:44 message that will go to us in ClickUp. and then it will basically send it to us in ClickUp. So, it takes a very similar shape as to what we've already seen, except for instead of going off on a

3:11:53 schedule at 6:00 a.m., it goes off based on the actual action which is triggered by us. And then once again, the only time we actually see AI inside of this process is just here. Now, you could

3:12:02 actually do this with no AI. If you wanted the form submission to just shoot you a message in ClickUp, you could do it with placeholders and you could do that. you'd save yourself some money and

3:12:11 some time because there'd be no AI message or sorry, there'd be no AI step and that would be 100% deterministic every single time. Now, obviously, there are some edge cases you might want to

3:12:19 think of. You know, how do you build the form so that they can't, you know, spam it a thousand times in a second? How do you build the form so that if you're expecting an email field, you're

3:12:26 actually going to get only email fields? There's a lot of other edge cases to think through. But what's cool about that is after you build this, you could say, "Okay, cool. Now, spin up 50

3:12:34 different sub aents and have all of them test this thing, try to break it, and then let me know what you find." and you can really stress test and QA your own automations before you actually have any

3:12:43 sort of human or customer find these bugs or anything like that. So, that's pretty cool. It's going to spin up a local page. It's going to use the existing trigger.dev deployment on the

3:12:52 back end. And now, because we've already set up the connection with like GitHub and trigger.dev, all of the automations we want to build in the future are going to be much much easier. So, I'm

3:13:01 basically just going to check in with you guys when this one is done and show you how that works. Okay, so it says that this is done. You can see that we have this form here to fill out. It says

3:13:08 that it has built and verified this end to end. So submitting it now will start the hosted trigger. Task. It will generate the outreach recommendation and it will send everything to ClickUp. So

3:13:17 let's go ahead and real quick check in trigger.dev. We do have this new one right here. I'm going to refresh just in case. And you can see that we have this new one called form submission. It has

3:13:26 already been triggered a few times. It looks like actually let me just zoom out so that we can see this better. It has been triggered five times. So this was Codeex testing it. And now let us go

3:13:34 ahead and test it ourselves. So, I'm going to go back into Codex. I'm going to fill out this form. So, we're just going to say Alex Morgan, Alex company.com.

3:13:44 Team size, we will just put 20 plus. We are spending so much time on, you know, making hamburgers and it's just becoming a real bottleneck. So, I'm looking to see if we can automate that. And let's

3:13:55 go ahead and shoot that off. What do we get on this front end? Your request is in. Let's go back to trigger.dev. Let's see if we get some sort of run. I'll go back to the tasks. We should see it

3:14:05 looks like there's a new one executing right now. If I click into this run, we should see that all of this is actually going on. So that proves that this web hook has been set up correctly. Now, in

3:14:14 the old days, we would have had to set up the URL and kind of sync them together and like build out the payload. But we don't have to do that at all and it this is the payload basically like

3:14:21 the information that comes through and now we have the output. Let me check and see if we got this in ClickUp. Sweet. Okay, so we actually got a lot of these. These are all of the demo examples that

3:14:29 it tried, you know, the new form submissions. But this is the one that we just submitted. You can see right here, Alex Morgan. Here's the email. Here's the team size. Now, this took no AI.

3:14:38 This is what I meant when I said that this could be just placeholders. So, if you wanted just to just get this notification, no AI needed. But here's where we needed AI, the recommended

3:14:46 outreach and the email draft. And this is what had to be obviously generated with artificial intelligence. So, that's kind of the trade-off there. But, super simple, super easy. That took me no time

3:14:56 to build. And I barely had to do anything as far as like wiring things up by hand. So, that is a web hook automation. That doesn't obviously have to be just a form submission. That can

3:15:04 be a new record in the CRM. That can be a new email entered the inbox. There's so many different event-based triggers that you can have for a web hook style automation. By the way, guys, I've got

3:15:13 this completely free SOP for you about getting your first aation client. It's going to go over the exact steps that has been proven for hundreds of our AIS Plus members to get their first paid

3:15:22 gigs. It goes over the one sentence service pitch that can get you started today. Why your first client should cost you money. The five-minute video that answers, can this person actually

3:15:29 deliver before you've actually received any money, what to do when you have zero case studies. There's so many good things in here that are going to help you out. Even if you already do have

3:15:38 clients, I would recommend grabbing this because, like I said, it's yours completely free. So, if you want to grab this, there's a link for it down in the description. Let's get back to the

3:15:44 video. All right. And for this last use case, now that we've done the web hook as well, this one's a little bit special, and this is when you kind of want that full agentic loop that you're

3:15:53 used to getting inside of codeex, but now you're getting it programmatically, and that is through the codeex SDK. So, a really good opportunity to see if you want to actually use the Codex SDK is

3:16:04 when you need to bring something that might be an already existing routine into something like trigger.dev. Now the thing about this is most of the time if you're running an automation in a

3:16:14 scheduled task inside of codecs, I would recommend you keep these as a codec scheduled task because then it actually eats away at your subscription. But if you need to do this at scale and you

3:16:23 can't use your subscription programmatically, then you probably want to build the automation with the Codex SDK and then host it on something like trigger.dev. So if that ever does become

3:16:31 the case, you kind of get the full agentic loop. So here is what that might look like. We saw these first two examples. Now, what I'm going to do here is this is going to be a scheduled task.

3:16:40 So, we're just going to call this actually let's just call this every 30 minutes. And I actually do this for real inside of a a scheduled task inside of my codeex where 30 minutes it wakes up

3:16:50 and we basically just have this full agent. And what it does is it basically it looks at my I'm just going to call this an agent because it does a lot of things. It looks at my alpaca account

3:17:00 and actually let me just like go down here and make some tools. Just pretend that these are tools. So, it has to like check alpaca, which is where I have like some trading going on and my agent here

3:17:09 helps me trade. It will also, you know, like do research. It can also just like look at like past logs and stuff like that. So, it's leaving logs for itself every time so that the next time the

3:17:22 agent wakes up, it's not completely stateless and just like doesn't know what to do. So the reason why this is the SDK rather than just being more of like one of these deterministic AI

3:17:33 automations is because we don't know how many times it might need to look at the logs and read things or look at the research or check back in in alpaca if it's doing research for 10 minutes and

3:17:41 it's like wait let me just check the current holdings one more time because things may have shifted this it it has the ability to basically just go back and forth you know it can go like this

3:17:48 and then this and then this and then this and then this and then this and then this and it's basically an agent and it has that luxury and then what happens is It essentially just sends an

3:17:57 email with a recommendation. So, I'll just call this email. And it sends that to Grockbot. And then Grockbot actually places the trade because OpenAI models and cloud models don't actually let you

3:18:06 trade anymore. They just won't do it. I think it's cuz all this security stuff is going on. But Grock still lets you trade. So, my Grockbot is the one placing the trades and Astra is doing

3:18:13 all of this heavy lifting research. Now, yes, you could do this exact same sort of thing in a deterministic flow, but that would just look different. It would basically be a predetermined amount of

3:18:23 research. So, for example, it could hit like a research step or maybe it hits, you know, two research steps. So, I'm just going to like sort of demonstrate this like this where maybe it does two

3:18:37 different like research sources. So, maybe this one is firecrawl and this one's perplexity or something. And then it basically will send it to an agent to um look at the research and can create

3:18:47 the draft, create the email and then send it off. But what if it looks through this and it's like, "Wait, I want to do some more research." Because now the AI agent has analyzed it and

3:18:55 it's like, "Wait, let me do some more." We would have to work in somehow the ability for the agent to loop back and do more research. And that's essentially why we would just use something like

3:19:02 this where we're now using the SDK to give us that sort of autonomy. So, let's go ahead and build that out real quick. Now, this would be a very similar process where I would like you to say,

3:19:12 "Hey, I want you to interview me about this process. I want to get very clear on what this does." Especially the more autonomous your systems get, there's more room for error. There's more room

3:19:20 for, you know, you're increasing the risk, you're increasing the cost, you're increasing complexity, you're increasing the maintenance, you're increasing a lot of things. So build the simplest

3:19:27 solution possible and only move up the sort of AI systems pyramid as I call it when you truly need that functionality. So I would go through this whole process of interviewing to get what I want. But

3:19:36 we have this luxury here of all of this is already set up because we have these routines going. You can see that I have this um thread called challenge thread where I'm doing this actual trading and

3:19:46 you can see that every like 30 minutes or so it's basically just starting this routine and it's doing the whole agentic loop. We see that one is actually running like right now and this is the

3:19:54 this is codeex. This is codeex working and we're basically just trying to move this over to trigger.dev. Now I'm not really wanting to I obviously want to keep that using my subscription but for

3:20:02 the sake of the demo. So what I need you to do now is I want you to build me a codeex SDK automation and we're going to host this inside of trigger.dev. dev. Now, what I want to do is I basically

3:20:13 want to transition the Astra trading challenge, that thread, I want to move that routine over to trigger.dev. So, we basically wake up like every 30 minutes during trading hours, one check before,

3:20:25 one check after, and we do this research loop. We check Alpaca, we check the logs. So, familiarize yourself with that actual process, and then just turn this into a Codex SDK automation for me and

3:20:34 host it in trigger.dev. Let me know if there's anything else that you need to build this. So once again, the hardest part is already done here, and now we're just going to get that built out. So

3:20:42 I'll check in with you guys when this is all done. Okay, so this run took about 15 minutes, a little over 15 minutes. And we can see now that this is ready for our review. It did 44 passes or 44

3:20:52 tests and it paused the old desktop routine, which I'm going to say, you know, I don't want it to pause that. This was just a demo. So anyways, um it was being pretty proactive there,

3:21:01 though. But anyways, what it did is it made a new project. So it created a full new one rather than what we were doing earlier. And I'm actually just going to open this up full screen in my desktop

3:21:10 over here. Give this a quick refresh. You can see what it did is it basically made a few different like tasks or runners or whatever you want to call them. And they kind of call on each

3:21:19 other. So you can see here there's the check. This ran 18 times. Um then we have an install plan. We have a pre-flight and we have a proof. So I'm not exactly sure the right order to run

3:21:28 this in, but this is what it did. And if I go to the runs, we can see exactly what happened. So it basically runs, you know, these failed, these completed. We've got these checks. So what we're

3:21:38 going to want to do is we're going to want to figure out what does each one of these do? So I'm going to go ask um Codex that real quick. So I can see that you actually created four different

3:21:46 tasks. We've got a check, an install plan, a pre-flight, and a proof. Can you just explain to me what these are and how they work together and how these use the codeex SDK to give us sort of that

3:21:55 full agentic loop that we're looking for here? Okay, so Astro check is the actual recurring worker. that is the one that should fire on the schedule. The other three help it do things. So the

3:22:05 pre-flight checks the codeex runtime, alpaca connection, SIP market data, policy files, and the checkpoint storage. And it can optionally make a small model test call. So this really

3:22:15 only needs to run during setup or troubleshooting. We have proof, which runs the full research and reporting workflow, including the the test label, email, and clickup delivery. It prevents

3:22:24 actionable trade tickets during the test, and this will run before enabling the routine. currently disabled after verification, the install plan during setup, and then the check at each

3:22:34 scheduled check. So, it's actually interesting. It made all of those to help it build and to check. But now, I'm not sure because each check now follows this process where we have it loads up

3:22:42 the context. It asks codeex what needs investigation. It finds the evidence. It has Codex evaluate all that. And this is with the SDK, of course. It validates and delivers, and then it saves the

3:22:52 handoff. So the agendic part is Astra choosing what to research using web tools, doing deeper evidence, and adjusting its conclusions based on the result, which is pretty interesting.

3:23:00 This is the only one that will actually run now that this is all built. So I'm going to go back into here and we're going to go to Astra check and we're going to hit test and just test this

3:23:08 thing out right here. Now look at this. It just called this attempt right here. And we open this up, we can see everything that's actually going on. Now this is pretty cool. The reason why this

3:23:16 did nothing is it actually sent this payload over which was like the time and the date and then the output was hey this is early because this was set up by Astra in a way where it's supposed to

3:23:26 check in during market hours and right now the market has already closed. So the next scheduled run isn't until tomorrow morning before the market opens. So that's why it basically fired

3:23:35 and it was like oh you know this actually isn't right. I'm not supposed to run yet so I'm outputting the word early. But I do know that this works because if I go into my ClickUp thread

3:23:42 where I'm actually getting these notifications every day from the actual routine, you can see that these started to come in. You can see we have the test migration here, 341, 347. We have

3:23:51 another test migration for Nate's review. It's doing the research. It's giving us assessments. It's linking things. It's looking at our account. It's pulling in all the sources. It did

3:23:58 another one at 352 for the market close. So, this is currently working. And what it's actually doing is it is messaging Grockbot. This is my trading Grockbot that actually places these, you know,

3:24:08 trades based on getting an email. So everything transferred over to the point where it's getting these test migration reports as you can see because what's going on is trigger.dev automation with

3:24:17 the Codex SDK like it said, it's checking, it's doing research, it's looking at the evidence, and then it's making recommendation and it's sending it to ClickUp, but also as an email to

3:24:27 my Grockbot. So you can see we now have already set that up in trigger.dev dev to basically mimic a codeex routine with Astra on the back end with a full codeex agentic loop, but we could now trigger

3:24:37 this programmatically and we could do it by web hooks and on different schedules. And like I said, you only would really do this when you need it to truly be programmatic and at scale because

3:24:45 otherwise you want to throw it on your subscription because paying for Astra via API credits is obviously more expensive than paying for Astra via subscription. But that is how these

3:24:54 three types of automations differ. And now I can come back in here and I can cross out Codex SDK. So Codex also has this really cool feature inside the desktop app called voice mode, which

3:25:04 basically just means that you can actually just talk to it and it will talk right back to you. So it's super super cool. Let's hop into this next section about how to use Codex's voice

3:25:12 mode. Hey there Astra. I've got a few tasks today. The first one that I want you to help me do, I want you to delegate off to a codeex thread where we take a video that I made on YouTube a

3:25:20 few days ago. Or actually no, it was just yesterday. So, I want you to turn that into an X article for me and put it in my X account. >> Got it. Let's line that up. So, that X

3:25:28 article is ready. Let me open this up and we'll take a look. We can see that we have all of this formatted. We have that 5x2 thumbnail that we wanted that matches our YouTube one. And now we can

3:25:36 see that these screenshots are coming through. So, the prompt, it analyzed the entire video, screenshotted it, put this in here. It even cropped it. If you guys notice, this isn't the full screen, but

3:25:44 it's matching these up because it understands what am I seeing in this image and where do I put it in the text based on the actual article that was written. Let's also try something else.

3:25:52 So, inside of my school community, AI Automation Society Plus, we just did a little bit of a revamp to the actual course. So, if you could sort of just like look in there, get familiarized

3:26:02 with that course structure and just help me build a really simple, you know, AIS branded landing page. >> Yep. I'll start on that. >> All right. I just got word that the AIS

3:26:10 Plus curriculum page is done and it's looking pretty impressive. Let's open it up full screen. Holy, this is actually the courses. So, I just built this entire personal OS with just my voice

3:26:20 using GBD6 Astra voice mode. Now, at a glance, it might not seem too impressive, but it had to figure out how to connect my entire Herk Brain. It had to go pull all the data from these

3:26:29 sources. It had to connect my real calendar, and this is actually functional, meaning here's my actual calendar, and if I came here and I wanted to add an event right here, we'll

3:26:35 just call this test, save event. This is actually going to reflect on my real calendar over here. Now, there are still a few things that I'm looking to connect, but I can communicate with my

3:26:43 team through Slack, ClickUp, and email right from this interface. So, I can open up this conversation and I could reply to it right here. And I can also kick off new threads via email or

3:26:51 ClickUp from here as well. And I also have community polls from YouTube, from my school communities, what they're stuck on. And I also have industry polls from X, YouTube, things like that. The

3:27:00 reason voice mode is so powerful is because it just uses codecs on the back end, which means it can look through your AI operating system and it can do everything that Codeex can do, but you

3:27:08 can control it with your voice and multitask, kick off different threads. You can do it from your phone, so you can literally be on a walk and doing everything you need to be doing. And the

3:27:15 browse use capabilities from GBD6 Astra are the best I've ever seen. Right here on screen, you can just see a silly example where I gave it two pictures and I had it paint me in Canva. And this did

3:27:23 take quite a while, but if you think about all of the little strokes and everything that needs to do, both on the vision side and the browser use side, it's actually pretty impressive. Now, on

3:27:31 top of all of the capabilities, you also need it to have the right context and connections to actually be useful. So, when I tell you guys that I've been talking to Astra in voice mode, and it's

3:27:38 just been feeling like it actually knows everything about me and my business, it's on another level. So, we just had an event in July called AIS Live and right here there's over 150 GB of

3:27:47 footage to look through. So, I gave Codex this link and I basically just said, "Hey, we need like a 60-second sizzle reel from this event, like a recap." And in only two prompts, this is

3:27:56 what I get back, which is really impressive. Logged off yesterday just buzzing. I had goosebumps. I was so excited and I had so much fun. Cost doesn't justify price. Price

3:28:23 justifies cost. What is the value of what you know and what you know how to do? This culture that we have here at AIS is is just really special.

3:28:55 If you think about the actual story that it told and how much data it looked through and what it was able to create, that was really, really impressive, like I said. So anyways, let me stop yapping

3:29:03 to you guys and start yapping to GBD6 Astra and show you guys what it can do. So all you're going to do is down here you can see there's a little button called start new voice chat. You're

3:29:10 going to click on that. And now it's time to put my headphones in. Hey there Astra. I've got a few tasks today. The first one that I want you to help me do, I want you to delegate off to a codeex

3:29:18 thread where we take a video that I made on YouTube a few days ago or actually no it was just yesterday. It was called like Astra made this. It made the whole video. So, I want you to turn that into

3:29:27 an X article for me um and put it in my X account. >> Got it. Let's line that up. >> And while you're working on that, can we have a separate thread actually start

3:29:36 making the thumbnail for that? >> Sure. Adding that as well. >> Now, I do want to make sure that you're using all my right skills and my right context. So, make sure you're working

3:29:45 inside of my Herk 2 project for all of these little tasks that we're going to kick off today. >> Got it. I'll make sure that's the context going forward. The original

3:29:56 article task has been told to stop and hand off any verified finds. And the thumbnail task has just been told the same. >> Okay, awesome. Yeah, cuz I see that you

3:30:04 kicked off two regular chats and these wouldn't have had the same context. So, it's important that I told you to work inside my Herk 2 project and now you should be kicking off the right ones up

3:30:13 there. And it looks like you are. >> Yep. Doing that now. And the article task now has the thumbnail tasks details so they can coordinate the source, the angle, and the finished cover. Both

3:30:26 replacements are running inside Herk 2. Awesome. Now, for the actual thumbnail I want for this X article, I actually already made one that I really like. um the one that I put on my actual YouTube

3:30:39 channel, it was like an image of um it was like me with my feet up on the desk and the OpenAI logo was editing the video. So, all I really actually need you to do

3:30:51 there is if you could just turn that into a 5x two version so it fits into the X article. >> Sure. I'll update the thumbnail work with that.

3:31:01 >> Awesome. So, I know that probably will take a little bit of time, so just let me know when those are done. Let's also try something else. So, inside of my school community, a automation Society

3:31:10 Plus, we just did a little bit of a revamp to the actual course. So, if you could sort of just like look in there, get familiarized with that course structure. Um, I'm talking about the AI

3:31:20 Automation Society Plus group inside of that one. if you could just analyze the structure real quick and understand sort of like the transformation from a new joiner to the end state once they've

3:31:30 taken all the courses and just help me build a really simple you know AIS branded landing page that I could basically send people to if they're like hey you know what is AIS plus I could

3:31:39 send them there and they could get a rundown of what that curriculum looks like and I want you to also design that in a way where you use my scrollcraft skill to have layers and to have it be

3:31:49 you know dynamic and visual but don't make it too overwhelming So if you could start designing that website as well based on that knowledge. >> Yep. I'll start on that and I'll keep

3:32:00 the design dynamic but not overwhelming. >> Awesome. So now I see that I'm talking to you in one thread but then I can also see when I look at the all the threads that you're kicking off. I can see how

3:32:09 you're clearly multitasking. And then by the way I see the thumbnail that you made. That looks great. That's exactly what I was looking for. Will my other thread that is making the actual article

3:32:19 it knows that that's there. Oh yeah. It looks like you sent that exact message where I said that looks great. That's exactly what I was looking for. So, you are able to actually coordinate these

3:32:26 different codec threads to talk to each other. >> That's right. The article task has the approved cover and your go ad. >> Awesome. Well, while I wait for me to be

3:32:36 able to verify that that all worked, could you let's make another improvement to the Upit OS. Let's see if we could add I guess let's add another section for meetings. I know we already can see

3:32:47 all of my meetings in the actual Herkbrain visual, but let's add another tab on the left that just shows the meeting. So, you pull that in from Fireflies and it's just kind of like a

3:32:56 nice visual of the date that the meeting was and just like a one sentence summary of what was discussed or if there's any important action items from there just so I can visually see all the meetings

3:33:05 that I had. Um, and maybe if there are any important meetings coming up as well. >> Okay, I'll take that on. >> Cool. So, while you're cooking on that,

3:33:15 I'm going to go ahead and actually pause this voice chat. I'll pick it back up with you in a bit when you have some updates for me. But, I want to kick off some voice threads on my mobile device

3:33:24 and see how that all works as well. So, I'll be back in a bit. Okay. So, anyways, just wanted to show you guys what was actually going on here. So, it made this thumbnail, which was actually

3:33:32 based on if I go to my thumbnails, this one. So, this is the one that I wanted to use because this is the one that did best on YouTube. And I just needed it to make that a 5x two. And what you guys

3:33:42 will notice is I didn't tell it to find this exact thumbnail. It looked through my YouTube probably to see which one was the live one. It looked through the local files to actually build that. Same

3:33:50 thing over here. What happened was I don't know what it delegated around, but all I said was I need an ex article for this. And so what it did is it had to pull in the actual video. It's

3:33:58 transcribing it. And now what it's doing is it's making the expost, but it also took 11 screenshots. And so it's going to place those screenshots in the X article at the right spot. So we'll get

3:34:06 to view that in a sec. But I just wanted to call out what's actually going on on the back end there cuz it's very impressive. But now I need you guys to look at this. So you can also do this on

3:34:15 your phone. So imagine you're on a walk or you're in your car driving and you need to kick off some tasks or delegate some work to your team or whatever it might be. Because it has all the

3:34:24 connections in here. You're going to go to remote. You're going to scroll down and this is where you'll Oh, wait. Right here. This button right here. This is where you'll start a new voice chat. And

3:34:33 now it's going to be the same thing. Hey, so I need you to help me kick off a thread inside of um my Herk 2 project. What I want you to do is I just need a nice rundown of the best use cases that

3:34:48 you've seen, the best viral demos that you've seen of Astra 6 um you know, GPT Astra 6 on X. >> Sure. >> So, the point I'm trying to make here is

3:34:58 that I can keep scrolling on my phone even when it's doing things on the back end. As you can see, it's still keeping this conversation Got it. I'll keep going here. Yep,

3:35:08 that's exactly it. I kicked off the Astro research thread and we can keep talking here meanwhile. >> Awesome. So, while we're while we're waiting on all this, could you just take

3:35:15 a look at my, you know, the look at the YouTube space, look at some of my past videos and look at what else is going on in the space and give me a quick analysis of maybe some trends and maybe

3:35:26 some things to start preparing for. I'll scan the current landscape. Now, big trends heading toward 2027 are more AI assisted production and personalization, deeper community play with memberships

3:35:37 and owned audiences, short and long form blending into one funnel, and creators rewarded for building recognizable shows and perspectives, not just one-off uploads.

3:35:48 >> So, the point I was trying to make there was just that you can do this on your phone as well, and it's a super clean, crisp syncing between your desktop and your phone. So, we've got the Up OS

3:35:57 working on adding a Fireflies task or a Fireflies meeting section. So, I can see that stuff. We've got the X article being finalized and we'll pull that up. So, that X article is ready. Let me open

3:36:05 this up and we'll take a look. We can see that we have all of this formatted. We have that 5x2 thumbnail that we wanted that matches our YouTube one. We have all of this formatted in a good

3:36:14 way. And now we can see that these screenshots are coming through. So, the prompt, it analyzed the entire video, screenshotted it, put this in here. It even cropped it. If you guys notice,

3:36:22 this isn't the full screen. We keep going down. This is the full screen. This is a full screen. Full screen. But it's matching these up because it understands what am I seeing in this

3:36:30 image and where do I put it in the text based on the actual article that was written. And there is nothing sensitive in this video. But what it will actually do too is if I have like an API key in

3:36:38 there or if I have like a personal email, it actually will like blur it out. So it does that automatically. And now that is just worked into the skill. So this is now an article that I would

3:36:47 just be able to read through. All of these screenshots are taken. They're cropped if they needed to be. They're blurred out if they needed to be. And then I can just go ahead and publish it

3:36:54 because it did all of that with just my voice. We can also see that this one that I kicked off from my phone is done. It searched through X and it found um multiple posts. One from Ben Davis, one

3:37:04 from Ethan, one from Thomas. You can see that it gives us the actual link to all of these threads. So if I want to click on, you know, this 3D graphics build, I get taken straight to that actual post.

3:37:12 1.3 million views. I did see this one. This one was pretty impressive. We can see that now we have this little rundown. And if we would have stayed on the voice chat, it would have, you know,

3:37:20 told us about all this. It looks like the upp one is also done. Let's see what happened over here. Meetings is live. So, let me open this up and we'll see. Actually, I'm going to open this up full

3:37:28 screen. We now have a section called meetings. You can see on the calendar, I've got one tomorrow. I've got one on Monday. And we can actually pull up the actual calendar details. So, if I open

3:37:37 this one up, does this take us straight to the calendar? Yeah, it takes us straight to the actual calendar. And we see the event details there. And now we can also see past meetings. So this all

3:37:44 hands we can see the team reviewed learning community growth product updates operational improvements AI tools and a shared mission centered on customer happiness. We have an action

3:37:52 highlight Nate Herk share bookmark links to the most used Google Drive folders. Yep. Yep. Yep. And we can see other action items here and we can also open up this recording which takes us

3:38:01 straight to Fireflies. So that quickly we were able to develop a new section into our actual um OS here which is a combination of Fireflies and Google Calendar. And I don't even know if it

3:38:12 uses any AI in the background because Fireflies already processes these AI action items. Maybe it hit, you know, the GBT endpoint or the Enthropic endpoint to actually put this stuff in

3:38:21 here, but really impressive. All right. And I just got word that the AIS Plus curriculum page is done. And it's looking pretty impressive. Let's open it up full screen. Build the skills shape.

3:38:28 What's next? And we've got Holy, this is actually the courses. I wonder if it Okay, it takes you down to here. There's like a curriculum section. Now, I think this looks really good. It pulled a

3:38:39 picture of me. You also will realize that the background, right? The background, you can see these layers. The mountains are in different spots. They move in different, you know, paces.

3:38:48 But what I love about this that it did is if I actually go to our AI Automation Society website, it kept a very similar vibe, right? Like the blue feeling, the backgrounds um of the mountains and

3:38:59 stuff, even on the certification tab, it kept the consistency. Now, honestly, I think these mountains look a little bit cooler and there's like fog, but it kept everything on brand because it was able

3:39:08 to crawl through my workspace, look up what it needed to look up to make sure that it actually understood what I was asking for. And that's what I think is so cool about Astra is when you give it

3:39:16 a task, it feels like it just will contextualize and ask you questions and crawl through things first because it wants to make it more specific to you. And that's what is just really, really

3:39:25 cool. I mean, this is this is the actual curriculum. If I go to AS+ and I go to classroom, that's exactly what happens. you know these are the actual courses and then over here uh wrong one these

3:39:34 are the actual courses that we have in there so it was able to look through all of that and build out the actual curriculum I think this is really really slick all right so voice mode is one of

3:39:43 the big differentiators between cloud code and codec but there are a lot of other things and I will be honest in this space you are jumping back and forth a lot like I use both of them

3:39:51 every day and sometimes you know codeex will come out with a big new release and then the next week claude code will come out with a new big release or maybe there's a new tool like Hermes agent or

3:40:00 Grockbot or whatever you know is going on in the AI world because it moves very fast. Now the first thing I want to say is that you should not be worried because this whole course we've been

3:40:10 talking a lot about building your own AIOS and when you're building your own AIOS you guys know that it's just a bunch of folders and files and those live on your machine. So whether you are

3:40:18 using codeex or cloud code or Hermes agent or whatever the next tool is you just say hey Mr. Mrs. AI agent, look at my AIOS, get familiarized with it, and now you can in a matter of 5 seconds,

3:40:30 all your skills transfer over, all your instruction files transfer over. You don't have to be worried about being like vendor locked. Everything that you're building, not only your local

3:40:37 files and folders, but also your brain, your knowledge on how to work with AI and how to talk to it and the mindset around it, that is all your IP. And it's very tool agnostic. It's very model

3:40:47 agnostic. So, you can be flexible. Don't worry about it. But I do want to show you guys this next video which is about me running a bunch of different use cases on these two coding agents and

3:40:58 just showing you the difference what that might look like sometimes with cost and speed and just like quality. So hope you guys enjoy this next section which is kind of just like an experiment where

3:41:06 I show off the differences between these tools in these different use cases. I just put GPT6 Astra headto-head against Fable 5.1 across 15 different use cases that actually apply to my day-to-day.

3:41:15 I'm talking things like web design, personal organization, taxes, browser use, vision, and building actual automations and software. For every single one of these use cases, I break

3:41:25 down which model did better, how long it took, and what it cost. So, if you've been wondering what model to use for which use case, then this is the video for you. I don't want to waste any time.

3:41:33 Let's just get straight into it today. All right, so these are the 15 different use cases that we have. And like I said, I chose these because these are things that I actually see myself doing with

3:41:40 these models. So, there are timestamps down below if you want to jump around. And at the very end, I'm going to show the final sort of like takeaways and final cost and time breakdown. And by

3:41:47 the way, guys, I've eaten through like three CEX subscriptions, four cloud subscriptions, and thousands of dollars in usage credits. So, I really hope that this one's helpful. All right, so for

3:41:54 number one, I'm not going to read off every single prompt, but here's basically what I asked. I wanted a McKenzie level presentation on the state of SMB report for my Hercules Consulting

3:42:03 brand. So, I needed this to be branded and I needed it to apply to our target audience with this fake consulting business. So, that was Claude. You can see that I gave Codex the exact same

3:42:10 prompt. Now, immediately what I've noticed is that Codex asks more questions or GBT6 Astra specifically asks more questions. You can see right here, it already asked me questions,

3:42:19 whereas Fable just took this and pretty much just ran. And that is something I noticed overall. So, I just wanted to call that out to start. So, let's take a look at these two decks. Here is deck

3:42:27 number one. I won't tell you which model it was yet, but right away, I love the way it feels. It is, you know, capitalized. It has the branding. This feels pretty professional so far. Now, I

3:42:36 already did check both of these for the sources and the verification, and they are good from like the fact perspective, right? The research they both did was good. Now, I'm looking more so at the

3:42:45 way they actually formatted this deliverable. I think we're at a place now where GBT6 Astra and Fable 5.1, they're both so so intelligent. You don't have to worry as much about

3:42:52 fanning out agents and doing research. It's more to me now about how do they come back and present this in a way that tells a story and in something that I can actually like present over, right?

3:43:00 because that's what we want is we want to turn that messy data into something that's readable and something that's high quality. And right now what I'm feeling is that this is very very

3:43:08 branded. It matches our color scheme. We have a footer down here which I love. It's consistent. We have like structure. There's columns. Like it did a good job creating this. The thing that I think

3:43:17 right now though is it's very very wordy. As far as being able to present over something like this, it would be tough because there's so many words and it's tough to control like where

3:43:25 people's eyes will be looking and what they'll be speaking about. So, as far as a presentation deck, not the best, but as far as like a deliverable you might turn in, you might send to someone and

3:43:33 not present, this is very good. And there's a lot of sources down here. This came out to 35 slides. Now, let's look at the other version, which right away I don't like how this isn't capitalized. I

3:43:42 don't know. I just feel like that feels very professional. We have some images here, and we have the color scheme as well. But let's flick through. We immediately notice that this one is not

3:43:49 as wordy. I immediately feel like this one's less professional though because of the structure. The other one had like I don't know, you could tell in each slide there was like columns. there was

3:43:57 a little bit of like a I don't know there was a structure to it whereas this one feels a little bit more abstract. It feels like one big element. You've got one big, you know, title thing. And I

3:44:05 think that this one would be a much better deliverable if I was going to present in, you know, in a room full of people. But overall, as we flick through, this one just feels less

3:44:13 professional. The vibe I get is less McKenzie and it's more Google Slides or like Canva. I don't know. There's also not like a consistent footer or our logo isn't on every slide. So, you know, this

3:44:23 sort of slide, it just doesn't feel very structured and as professional. Even the sources section is way smaller, although that does link the actual sources. So anyways, this one was GBT6 Astra and

3:44:33 this one was Fable 5.1. And I will say in this specific example, I liked Fable better. So what I'm going to do here is I'm going to give Fable the win. So I'm just going to put this down here so we

3:44:44 know Fable won this experiment as far as just the output. But let's look at the time and cost. Fable was 37 minutes and Astra was 23 minutes and Fable was 26 bucks and Astra was 12 bucks. So Astra

3:44:56 was a little less than half the cost as well as you know 14 minutes faster. So the question would be if you used Astra and spent another you know 14 bucks using Astra to improve V1 would you get

3:45:09 somewhere that you feel better about than this first version of Fable. So anyways right now we'll look at the time and cost at the very end but as far as the deliverable I liked Fable's output

3:45:20 better here. All right moving on to experiment number two. We did a sales letter. So basically I said based on what you know about our certification program I didn't give it information. I

3:45:28 told it to you know use my AIOS use my second brain use any research to put together a sales landing page. So writing the copy that's optimized to convert. So think about our avatar think

3:45:38 about their pain. Think about all this and I don't want it to sound AI generated. So I'm not going to read the entire sales letters. That would be super boring. Just going to do a quick

3:45:45 skim through and talk about which one I actually thought was better and why. So here was the one that Fable 5.1 wrote out. You can see it is pretty long. This one was longer. It was about 2,800

3:45:57 words. And if we go over to GBT's version, if I come down here, it just output it right in the chat to me. This was the one that Astra wrote up. And this one was more like 1,700 words. So,

3:46:07 it was definitely shorter. Now, quick disclaimer here. I am not an expert copywriter. I would want, you know, like John or someone on my team who is really good at writing copy to tell me which

3:46:17 one they thought is better. I will say so far what I've been hearing from my team is that they are liking the GBT models lately better for writing whereas in the past Claude always dominated when

3:46:26 it came to writing. But I will say if I was a prospective you know certification student and I was reading these I honestly here would have picked Fable and the reason is because this is not a

3:46:35 cheap program and Fable had things like questions people ask. Will this get me hired? Does everyone who joins get certified? Am I technical enough? I already run an agency. What if I miss

3:46:44 live classes? The way it formatted this makes me feel a little bit more confident or makes me feel more aware of what I'm getting myself into. It also had a section about tuition. It also had

3:46:51 a section about who's teaching, who it's for, who it's not for, how this actual program works. Whereas the sales letter over here with Astra just felt a little bit more, I don't know, it just felt a

3:47:00 little bit more high level. And I think with the sales letter, it's supposed to be detailed enough to answer all the questions and pain that someone's thinking about when they're considering

3:47:07 paying for a program. So, I will say in this specific example, I do think that I liked Fable 5.1's copy more. So here is actually what this looked like. Fable was about four minutes. Astra was about

3:47:17 three minutes. Fable wrote Okay, I was wrong about Astra. It wasn't 1,700. It was actually more like 1300. So about half of the words that Fable wrote. But Astra came in at $143 whereas Fable came

3:47:27 in at almost four bucks. So here I will have to say again that I think that Fable won this challenge as far as just like a oneshot prompt and the initial output. All right, so moving on to use

3:47:38 case number three. Now, unfortunately, guys, I'm going to have pretty much everything blurred out here because this is about my taxes, right? And this is one of those things where you want

3:47:47 something that is very smart to be able to do this and in a way that you trust. So, I'm just going to go through high level the way I feel when I look at these outputs. This first one was Fable

3:47:55 5.1. And before I get into it, I do want to say in this example, GBT Astra asked me like seven questions before it started, whereas Fable didn't. So, that alone makes me feel more confident in

3:48:05 Astra's output. But let's look through here. We go through basically from January of this year to June of this year. So Q1 and Q2 of this year. It pulls all of the numbers. It looks at

3:48:14 federal and state. It looks at things that I need to read. It goes to the input and it shows me all of these different things as far as the parameters, the notes, the values, the

3:48:22 sources. And Fable gave me like a decision logic. It said, "Hey, if you fit into this bucket, do this. If you fit into this bucket, do this." Whereas Astra just asked me those questions and

3:48:30 then gave me that recommendation. Anyways, we have a um P&L. We have some projections. So for the rest of the year, what I might be estimating to pay. We have tax calculations here for the

3:48:39 individual return and all of the entities that are passed through. We've got payments. We've got things that are flagged. And then we have the actual sources it pulled from. So it's not bad.

3:48:47 It's a really good deliverable like light years ahead of what used to happen. But now we have the one from Astra. We have the actual information as far as start here. This is what it did.

3:48:56 We have monthly results. We have the tax forecast. We have Okay, that looks very good. We have all these payments. We have inputs and open items. So, these are other things that I might need to

3:49:05 answer, but this is just way more tailored. It's easier to read as well. We have the source checks. We have the transaction ledger. And this one is basically every single transaction here.

3:49:14 So, this Oh my gosh, this is probably thousands and thousands of rows. This is 3739 rows of transactions here. And then we've got all the sources that it pulled from. Okay, cool. So, overall, this one

3:49:25 I think that Astro won, like no question. It was more tailored. It asked questions. I think with taxes alone, it gave me that that feeling of confidence and the structure of the deliverable was

3:49:34 just better to me. Now, Astra did cost more here and it took longer to run. So, 40 minutes compared to 22 minutes and 22 bucks compared to 13 bucks. So, even though it's more expensive, we're kind

3:49:44 of putting that aside right now. As far as the deliverable, I liked Astra here more. So, right now, we're sitting at Astra 1 Fable 2. All right. So, this next one's interesting. I asked it to go

3:49:54 through two of my email accounts and just look through, you know, potentially years of emails to help me find information about subscriptions that I'm paying for, how much total I've paid

3:50:02 them, you know, um if prices have increased and things that I need to cancel, things like that. So, let me pull up the two outputs. All right, so here is output number one from Fable

3:50:10 5.1. Now, as we get into the next tabs on the sheet, I'm sorry, guys, but I am going to have to blur some of the stuff out. It's it's financial information. So, anyways, we have the rundown right

3:50:19 here of the audit. We have the different emails it looked at. We have the different types of services, memberships, programs, retainers, contractors, insurance, all this type of

3:50:26 stuff. It's looking at the top 10. It's looking at our different categories, and it's showing us how to read this sheet. So, we have all of our active subscriptions. It's showing me what

3:50:35 email they're associated with. It's telling me which day of the month I'm getting build on. It's showing the current price. It's showing the monthly equivalent. It's showing me all of the

3:50:42 stats across all of the different payments that we've been making on some sort of recurring basis. It's flagging things and it's ranking them as far as priority. We can see um the different

3:50:50 actual flag notes. So some of these it might be unpaid. This one's getting a price jump. This one's having a seat creep. This one's having I have two subscriptions at once. So maybe we need

3:50:57 to figure out why do I have two different subscriptions on both emails there. So the point being it's able to look through so much information and give you all of these things that are

3:51:04 high priority and you can flag them. You can check them off. You can, you know, delegate that out to your team. We have all charges which actually this sheet came back completely empty. I'm not sure

3:51:13 why, but this sheet is completely empty. And then we have contractors and big payments. So, these are things it looks like. Anything above um either a few thousand or if it was a

3:51:23 contractor that was like a onetime invoice, then it flagged it here as well. And now, let's go over to the Astra 6 version. You can see that they consistently just format Google Sheets a

3:51:31 little bit different. Like, you can kind of tell Astra likes to make the cells larger and and wider and taller, whereas Fable usually keeps them the same. And, you know, they still do a good job

3:51:39 formatting, but either way, same sort of thing. At the audit up front, we have the current recurring baseline of subscriptions. We have the annualized baseline tool payments, tool invoices

3:51:48 without confirmation, claude usage, other bills, um price changes. It's flagging everything right here. Priority review right here. You know, there's a discount ending, there is contract tier

3:51:58 or a contact tier, there is a price increase, there's a term change, there's seats and tool overlap. So, it's flagging all that right away. We then have all the subscriptions. So, this

3:52:06 shows me a list of every single one. It gives me the price. It gives me the um last known bill. Now, this isn't showing me what date. Oh, there it is. Yeah, it's showing me every single day of the

3:52:16 month that we have the next one. We have other bills as well. So, these are kind of the bigger like bigger contracts um contractors or or bigger bills. We have payments. So, this is also like every

3:52:25 single one. And I think this is what Fable was trying to do on this tab with all charges, but for some reason just didn't come through. So, we have this here. And then we have also other notes

3:52:34 as far as different topics and different things to sort of be aware of. So, this is kind of like what it was flagging. So, overall here, I will say that Astra gave me a better output. It just, you

3:52:43 know, Fable messed up one sheet completely and Astro was cheaper here. So 12 bucks, 12 and a half bucks compared to about $50, which is insane. $50 for that output from Fable. Crazy.

3:52:54 It did take 17 minutes whereas Astra took 30 minutes. But anyways, right now we are sitting at 2 to2 Astra and Fable. All right, moving on to number five. We have meeting analysis. So I asked it to

3:53:04 look through all of my leadership meetings and my syncs with John and looking at other meetings that are relevant to the team. And I wanted it to understand or I wanted it to tell me

3:53:12 what are the biggest pain points, three biggest pain points and one of the highest value automations that we could build to remove some constraints and help us scale faster. So Fable gave me

3:53:21 the biggest pain points of me being the single production and review engine which is something that I've talked about 25 of 58 meetings. Certification delivery does not scale past one cohort.

3:53:29 So we're trying to figure out what that will look like. The critical constraint of the business is also a large cost center. 52 students blah blah blah. And then we also have demand collapsed and

3:53:36 every fix is blocked by unknown work and untrusted data. Weekly leads fell blah blah blah. And yeah, all this makes sense as far as our three biggest pain points. And then the one automation it

3:53:46 came up with was an AI customer success layer for theert one live record per student built from circle activity mastery check outputs and attendance. Students upload case readouts to an

3:53:54 endpoint instead of DMing pat weekly due dates and reminders so non-submitters go out through Gmail or beehive. I think this is something that I've wanted to do and that we talked about a lot and that

3:54:03 makes a lot of sense based on all these conversations. And and I also wanted to note here that Fable 5.1 analyzed 58 meetings and if we go over to Astra, Astra analyzed 79 meetings. So it did

3:54:13 get the context of 21 more meetings in here. But anyways, let's take a look. The first one was that growth depends too heavily on a few acquisition channels. Yep, same pain point I think

3:54:21 as Fable pointed out. The second one is that the customer journey requires too much personal intervention. This appeared in undefined AI plus onboarding call reser during certification and

3:54:29 leadership still has to reconstruct status and clarify who owns what. Okay, interesting. So these are different pain points than Fable 5.1 found. Now as far as the automation I would prioritize it

3:54:38 came up with the same thing. A customer success workflow which would check each student's onboarding status, unanswered questions, available progress, identify missteps, prepare follow-up, all makes

3:54:47 sense. Now in this case I think Astra won again like before I even thought about the cost which is absolutely absurd. Astra still won because I liked its recommendation about the leadership

3:54:58 stuff and ownership. And I think that's something that you really want to make sure you have figured out within a team. But anyways, if you look here, Astra was 5 1/2 bucks and Fable 5.1 here was 46

3:55:08 bucks, which just makes no sense to me. I'm I'm very very confused on how it cost that much. I had it look at the session logs. I also did a SL usage and this one cost 46 bucks even though the

3:55:17 active agent time was just 6 minutes. So I'm very confused on where those tokens went. Like I don't understand why. It also analyzed less meetings in Astra. But anyways, that's pretty interesting.

3:55:27 This winner is definitely Astra. So now we are sitting at 3 to2 Astra to Fable. All right, so moving on to number six. This one's super interesting to me. I gave it this folder which has over 150

3:55:37 GB of resources and recordings from our AIS live event in July. And I asked them to make us basically a recap, like an energetic reel, 60 seconds that shows like a recap of the event that we can

3:55:49 use to promote the next ones and things like that. Now, I'm really impressed with both of these outputs. Let me just say because with how much information it had to look through and how it had to

3:55:57 create a story and figure out how to put everything together with music, I think they both did really really well. So, here is Fable 5.1. >> Hello. Hello, AIS Live. My Let's Let's

3:56:07 get some energy going in here. >> Oh my god, I'm so excited for this. You guys are throwing out a fantastic event. >> Literally like logged off yesterday just buzzing. I had goosebumps. Let's Let's

3:56:18 get you guys paid. This was an experience that didn't feel like a webinar. It's been a great few days. Okay, so that was Fable 5.1. Let's take

3:57:06 a look at Astra's version. This has been so energizing. Yesterday I did a workshop where I went all in on every little tool that I use. Cloud code is my primary driver and then

3:57:32 I built like my entire second brain/personal agent on top of it. >> I could go to bed, come back and it'll still be working, you know, 10 hours later to try to achieve that goal.

3:57:52 It's great to be here. You guys have done a a fantastic event. Everything's been super smooth. You guys have been such an amazing audience today.

3:58:07 Now, this one, like I said, is really tough. I think that they both did different things well. I think that Asher did a better job of putting like live clips in there. I think that Fable

3:58:15 did a better job of kind of like gearing it up for the next event in October. There were a few moments where there were some things that were off. Like Fable had a few moments where the the

3:58:23 timing and the voices were felt like not very synced. Astra had one weird moment where it was showing an a video of Russ, but it said Devon Karns and it was just like that was clearly off. But I am

3:58:33 going to give the slight edge here to Astra. But I will say I'm very very impressed by both of those outputs. I thought it was given the vague goal, they both did a really good job of

3:58:42 exploring and creating a story. But anyways, this one is going to go to Astra because you can see this stuff here. Astra took 30 minutes whereas Fable took 50 and Astra cost 16 bucks

3:58:51 and Fable cost 26 bucks here. Now, next we did sizzle reels. I basically gave it an example of a scissor reel that I really liked or sort of like a sass I don't know a motion graphic a motion

3:59:00 design video that I liked and I told it to analyze what was good about it and I told it to use that as inspiration and create us one for AIS. So let me show you guys these outputs. First one is

3:59:09 Fable 5.1. Okay, now let me play you Astro's output. Heat. Heat. N. Okay, so Astro wins this one. I mean, I will say that music is a little bit

4:00:36 annoying. The but besides that, it took screenshots from AIS. Like, it went in here and it grabs pictures of the actual courses. Where was the other one? It took a picture here, which I thought was

4:00:45 just like I love that idea that it's thinking, okay, why don't I actually go get real proof and screenshots from the thing I'm actually building a reel about and put it in there. I also love this

4:00:54 like these 3D designs. I don't know, it just felt like it had depth. You can see the shadows here as it kind of, you know, zooms in. I thought that this as far as the energy that it had was much

4:01:04 better and much higher energy than the first one from Fable 5.1. Now, let's take a look at the cost and time. This one was about similar time, but Fable 5.1 was, you know, a little over half.

4:01:14 So, 16 bucks compared to 26 bucks. But once again, I still think that in this case, I'm going to take that output from Astra as the winner. So, moving on to number eight. This one's a little more

4:01:23 out there as far as I probably wouldn't really be doing this every day for work, but I had them build me these games. So, we have two different games where you're a pistachio and you're trying to escape

4:01:31 the kitchen. So, I'm just going to go through these real quick and you're going to tell me which one you thought was, you know, sort of like which model. So, you can see here it's it's uh it's

4:01:39 pretty smooth. I can double jump. I'm collecting these like serial bits and I have to hit these checkpoints to get to different spots in order to get out of the kitchen. The physics are real. Like

4:01:47 I'm bumping into these things and I can like, you know, hit the spoon and everything like that. It's pretty easy, but you know, I I I missed a cereal, but I'm able to escape and win the game in

4:01:57 about 43 seconds. Okay, now here is the other version. This one This one definitely looks a little bit more I don't know. It looks a little bit more premium, doesn't it? Um, this obviously

4:02:05 this wall isn't there, but it's just because we want to be able to see like this. But this one feels a bit more premium, I'd say. It also feels smoother when I'm actually like controlling this.

4:02:13 It is guiding me very specifically, though. I don't know. I can double jump, I think. Oh, no. I can't double jump. The physics seem to be real still. I can bump into stuff. If I can move things, I

4:02:22 can fall into there. Um, we've got a hot stove. Oops. It's interesting though, they both decided to have like the same flow as far as you jump up on the drawer and then you you know, you jump up on

4:02:32 books to start, you're now on the stove, you're going over to the sink. I mean, that's how most kitchens are, but still the the actual flow is pretty similar. And now we escaped in this one in 46

4:02:42 seconds. So, similar amount of time. I did get all of the crumbs in this one. They were designed very similarly, but I will say I think that as far as like feeling realistic and you know physics,

4:02:51 I liked this version better and this version was Astra. Anyways, look at these stats. They were very, very similar. So Astra was eight bucks cheaper, N bucks cheaper, and it also

4:03:02 took about 40 bucks longer. So very similar stats. Nothing blew anything out of the water here, but I would say that I'm giving this one to Astra. Once again, that one's subjective. Maybe some

4:03:11 of you guys liked Fable 5.1 more, but what I think is really important is that you're applying your own taste and you're finding your own use cases because some of these use cases, you

4:03:18 guys might be like, "Well, I actually liked Fable 5.1 here." Cool. Then you just figured out maybe you use Fable 5.1 for that use case, even though some other people might use Astra for that

4:03:25 use case. It's all about your own use cases and your own experiments. Okay, so moving on to number nine, we have an actual SAS, like kind of a mini SAS, and it's an evaluation app. So you should be

4:03:35 able to put in like automations and then run them against some golden data set and see the score and understand like what you need to fix and then you can rerun them and you can generate reports

4:03:43 and things like that. So you can see here this is the first version. This is Fable 5.1. I think that right away this is a little bit more of an intimidating interface. It's very very wordy but we

4:03:52 can see all of these tests. We can see our different automations here on the left as well. We can add a new one. So if we wanted to put the name, a description, and then actually you know

4:04:01 say how how do you actually trigger this automation? Is it Python or do you hit an API? And then from there, let's just put in some information. You would actually put in the URL or you would put

4:04:09 in the Python script. You would put in the test cases and then you would write, you know, how to grade it. So that is going to be very valuable. And it also has all of this stored on the back end

4:04:17 because it has to store all of these actual um runs and it has to store the code and it has to, you know, show the improvements and things like that. So here we can see this support ticket one.

4:04:26 Um I think they ran two tests. You can see that there's also like a visual breakdown of what this actually does. So this is a very deterministic script. We can see the code. We can see the setup.

4:04:34 We can see the test cases. So, this is essentially our golden data set. There's only 12 examples. You probably need more than that. But this was all generated to actually test the system, right? And if

4:04:43 I actually go in here and hit run, what does this do? You put the name, you put what changed, you put any tags, and you compare it with a different version if you want. And then I just go ahead and

4:04:51 start the test. And so, this came back. You can see, I mean, there were no changes. So, we still got two out of the 10, or sorry, two out of the 12 didn't hit, right? It also went super super

4:04:59 fast because I, like I said, this is super simple. But then you can generate a report. So if I come in here, we can see all of the different tests that we've ran and it would show us like

4:05:06 what's still broke and it would show us the improvements we make. So this is something that, you know, I actually do want to keep building out so we can do this for, you know, our our students and

4:05:14 basically give them a better way to evaluate their agents and things like that when they're building them out because eval obviously we all know are super super important and it's also good

4:05:22 to be able to see them visually. So this was Fable 5.1's version. Let's look at Astra's version. So I do like the interface of this one more as from a design perspective. So, I think that

4:05:30 this one feels I don't know, it just feels a little bit more premium. It's also less wordy, so it's less intimidating right away. I think we have a test library, so we can see all of

4:05:38 these different versions. We have run history. We also have reports over here that we can look at. So, we can see customer support agent. Um, I like how this report is structured. Let's see if

4:05:47 we look at this next one. We can see how it changes over time. This one feels more branded. Like, this feels like something I'd rather get in front of some eyes and get some feedback on

4:05:56 compared to Oops, I just closed out of the other one. compared to this version. This one just feels way more like a PC compared to this I feel like is, you know, something we can almost get out

4:06:05 there and start testing against. So, if I come into an automation real quick, let's say we go to the support ticket router and I want to go ahead and see it visually. I like that a lot. We can see

4:06:13 all the previous runs and I can go ahead and run this eval again and we can see it running a bit better. We're seeing it actually go through the test cases. I like this experience way, way more. And

4:06:22 look at this. Astra came back and cost half the price of what Fable spent in order to do this. It did take a little longer, 2 hours 25 minutes compared to an hour and 20 minutes, but it just so

4:06:32 much better for less cost. So, in this case, I'm giving this to Astra once again. So, Astra is now up, I believe, 7 to2 on Fable. All right, moving on to number 10. We now have this Herk brain

4:06:44 visual. So, this is a local host where I can see basically my brain. We see different clusters over here. Um, this seems more like YouTube videos. This seems like things are actually going on

4:06:52 in the business. Oh, wait. We can see down here claude memory video knowledge and we can remove things or put them back in business wiki skills and agents projects and this is dynamic. It's

4:07:02 really cool because we can see only what needs attention. We can see show all labels. We can see 612 notes 2500 connections. We can analyze things. We can talk to it. We can search for

4:07:12 things. Let's see if I search for um personal OS personal assistant openclaw. We can open up the specific things. And that is a pretty cool visual. So we can actually start to see in here inside of

4:07:23 the brain. And as we click on different things, we see all of the connections that are linking into it. So all the relationships. So that is a pretty cool experience. It does feel pretty smooth.

4:07:31 And I can control this. Uh not too well. Like I feel like I have to click on something to make that the center, but I I think that that's pretty cool. So that was Fable 5.1's version. Let me now open

4:07:42 up Astra's version. Okay. So this one's a little bit different. Um I can still click on all of these things. These nodes are smaller. I don't like how that feels. I don't like this interface

4:07:56 at all, like nearly as much. I have the same sort of vibe. I can click on business strategy or AI and ideas, videos and content, um people, projects, book and writing, all knowledge. Um this

4:08:07 interface is more overwhelming for sure. Like I I definitely like this one more. I wanted just something simple where I can see the brain a bit more visually. And I was honestly expecting Astra to

4:08:16 come in here and kick Fable's butt here, but I think that I like Fables more. And now let's look at the cost. Astra took 40 minutes. Fable took 24. Astra and Fable basically cost the same. Fable was

4:08:27 a little bit cheaper, but I am going to give this one to Fable. I think that Fable gave me more of what I was looking for. You know, once again, kind of a subjective grading criteria, but I think

4:08:34 Fable won here. So, what does that put us at now? We're at 7 to3 Astra to Fable, and we have five more use cases left. So, Fable could still take the lead. All right, guys. So, this one is

4:08:44 an HTML explainer doc. What I did is I gave it my YouTube video, the 25 Grockbot concepts explained. And I told it to explain it to me simply simply simply and visually. And I told it to

4:08:54 take screenshots. So, I said basically, give me a doc that I can read through and it feels like I actually watched the entire tutorial. So, we have all these different parts. We can see the

4:09:02 screenshot to start. This one obviously is is kind of wordy up front, but that's okay. We see how the team fits together. So me, these executives, these operators, we can see all these

4:09:11 different plugins and computers and skills. So that's a good little explanation at first. Now we move into the concepts. So we're taking screenshots and it is being analyzed.

4:09:19 This is Fable 5.1 by the way. We can see the executive bots. We can see the name and job label. These so far are coming through well as far as these screenshots being cropped and annotated, which is

4:09:29 really nice. It's taking a lot of different screenshots. We have the buttons here. We have all this. I think that this is coming out pretty well and it's showing us what we need to see from

4:09:37 the actual video. Now, I do think this is pretty wordy. Um, I think that but it is getting very specific. Like, I do think I could give this doc to some people and they would be able to feel

4:09:47 like they actually watched this video. It's it's cropping things. It's zooming in. It's annotating. I think that this is doing a really good job. It's even showing all of this. Yeah, I mean, this

4:09:55 is doing a really good job here. So, let's go over to Ash's output. This one already, it looks a bit better as far as the HTML. And now let's see if the screenshots are good as well. So we have

4:10:05 one here. It's pointing at a bot. It is pointing at the description, the executives. It's pointing at these different buttons. I will say Fables was more detailed for sure. Like this one

4:10:18 was designed better, but Fable's screenshots were better. Like Fable took more screenshots and made more annotations. And so far, I think that that was a better learning experience.

4:10:28 Even though this one, like I said, looks a little bit more I don't know. it it feels prettier or better. But if you were really thinking about could you hand this doc to a complete beginner and

4:10:37 have them actually go through and feel like they're learning, I think that Fable's going to take this one. But let's look at the cost. Like I still am going to give the output to Fable, but

4:10:45 when we look at the cost, Astra was much cheaper here. So the question is like, you know, if you used Astra and had it spend another, you know, 20 bucks or so, would it be better than Fable's output

4:10:55 right here? And I would argue that yes, it would be better. So it was way more efficient here, but it's, you know, its V1 wasn't as good as Fable's V1 in my opinion. So Fable's going to take this

4:11:05 one. So now we are at 7 to4 Fable or 7 to4 Astra to Fable. Okay, so moving on to number 12, we have this sort of like painting example in Canva, which is a bit silly, but it does highlight the

4:11:16 browser use and the vision really well cuz GBT6 Astra had to look at this picture, open up Canva, understand how to navigate with the different tools and things, and build this out, which it

4:11:24 did. And I think that this does look pretty good, especially when we go over to Fable real quick and I open up Fable's output. If I go to the painting and we open up the browser here, this is

4:11:33 literally what it gave me. And this was the reference image. I mean, this is just this is laughable. This is laughable. I I was expecting Fable 5.1 to do something much better, but it like

4:11:44 I don't know what exactly it tried to do, but this is obviously just very bad. So, no question Astro wins this one. And Astra was also cheaper. So, it's pretty ridiculous here how much Fable spent and

4:11:56 how long it worked just to give me that crazy output. So, Astra wins. Okay, so moving on to the next example number 13. What I did is I gave Fable and Astra this doc which said, "Hey, these are

4:12:08 this is a new course. I want you to put it into the free community as a draft and I want you to use browser use in order to do this." So, they basically opened up school. They went to the

4:12:16 classroom and they put in these draft courses, Astra AIOS and Fable AIOS and they put in their different courses. So they were basically instructed to have the description, the key points, the key

4:12:25 quotes, and any resources. And they had to put the video in here. And so this is Astra's output. I told them to do just the first five. And this looks really good. It had to navigate the interface.

4:12:33 It had to basically come in here, make a new page, it had to put in the video, it had to put in the text. And it did it really well and saved everything here as a draft. Now, if we go to Fable's

4:12:41 output, there are no videos here. This one arguably might have been structured better as far as like this looks cool and these are numbered whereas the other one with Astra, it wasn't like numbered

4:12:51 as well. Like it just it's longer. But the big big issue here is that Fable ran into an issue and for some reason couldn't upload the video. There was something issue with having it download

4:13:00 locally and whatnot. You know, I gave them everything it needed. I gave them the videos. I gave it the links. This is something that Ashstra did way better and I will say I've done this multiple

4:13:07 times in my school and I've used 5.6 six soul with browser use and just in general codecs and the GBT models are significantly better with browser use. So this one I'm going to give to Astra

4:13:18 and honestly the cost and the time wasn't too different here you know um very similar and very similar Astra was a little bit cheaper and Astra's results were much much better so this one's

4:13:27 going to go to Astra as well. So we are now at the place where the score is Astra has nine and Fable has four. So Fable can no longer catch up. But let's look at the last two outputs as well.

4:13:37 Okay, so for number 14, we did a website and basically what I wanted to happen is I gave it this URL and I said, "Hey, I want you to basically help me clone this website. I like the structure. I like

4:13:46 the feel and I want you to make this for Perkform, which is the fake, you know, um, can company." So, this is a really cool scroll dynamic. I know you guys, some of you guys hate this sort of site.

4:13:56 That's not the point. The point is, could Astra and Fable analyze this site and make one that has the same feel and the same vibe for perk? So, here is Fable 5.1's version. And it loads up. We

4:14:07 have this, you know, 3D dynamic element. Coffee reformed. We have like some, you know, DNA looking stuff here. We keep scrolling down. We've got these coffee beans falling. We've got the can come

4:14:16 back. We've got all of this going on. I would say that it's doing a decent job here. It's even got the huge like zoom out. It's doing a decent job of making this feel like the other site. You know,

4:14:26 obviously there are some issues here, but I think it's doing a decent job. Like for oneshot prompt, although sometimes I get into these issues with the scroll. I mean, that's a bit of a

4:14:34 bug, right? when you hit the bottom of the site, something weird goes on there. So, there's obviously some bugs, but it's not terrible for a oneshot. And now, let's take a look at Astro's

4:14:41 version. We have the big can here, which is a little bit dynamic. There's something going on with this text, right? Like the coffee is out of bounds for some reason. Keep scrolling down. I

4:14:51 do like this animation. I mean, that's pretty cool. It's a It's a coffee bean. There's definitely layers in the back. And by the way, this one, I didn't tell it to use any skills. I just said, "Hey,

4:14:59 I want the same vibe. I want you to analyze the, you know, the original site, this one right here." and make me one like this. Right. So, anyways, we're coming through the little sort of like

4:15:08 DNA thing. Coffee and protein together, two drinks in one can. I like that animation. This one overall does feel smoother than Fables. I do think that as far as a first pass. I like the way this

4:15:21 one feels. We get stuck at the bottom and we'd have to scroll all the way back up to the top. Now, I do think that Fable did a better job as of actually just recreating what we saw in the

4:15:31 original site, which was the prompt. Now, I do get stuck. There's a little bit of a there's bugs like this. Like, I get stuck. But as far as recreating this, Fables felt more similar to this,

4:15:40 I think. So, we will give this one to Fable. Um, it kind of needs it as well, but I still think in general, when it comes to design, you guys have probably seen throughout this video, and you've

4:15:48 seen my other videos. In general, I do think I like Astram more for design, but that was like, hey, show me what you can do. Take this URL, build me something crazy in one shot. It was kind of a

4:15:57 weird prompt, right? But in this case, I still think that Fable won. But here's a look at the cost and the time for this example. All right. And this is the last one. Number 15. We have YouTube 12-month

4:16:07 review and strategy. So, it basically had to look through the past year of my YouTube stuff and give me some stats. So, this is interesting. I mean, these are big docs. The one on the left is

4:16:15 Fable. It's 25 pages. The one on the right is Astra. It's 15 pages. So, let's just slowly scroll through and see what's happening here. I like how Astra right away has, you know, this chart on

4:16:24 new viewers and regular viewers. Whereas over here we have, you know, the year and numbers. We have a monthly review. We have total views across the past um ooh interesting. So actually what they

4:16:34 show me first on the left side is from January to now. Whereas over here I'm getting sort of like the actual past 12 months. So those stats are going to be a little bit different. It's showing me

4:16:42 content type. It's showing me the monthly scorecard. It's showing what my best content actually does. It's showing me some packaging stuff. It's showing me what to worry about. So it says news

4:16:50 brings the views but does not bring the subscribers. It's saying um the news packaging is drifting towards hype. The audience is saying so it's saying agency mechanics content has stopped working on

4:16:59 YouTube. That definitely feels to be a bit true for this audience. Now it's deep diving into some packaging. So it's analyzing my my titles over here. Whereas over here when we have the

4:17:08 packaging um what I would keep and improve the evidence retention first 30 seconds who you are reaching and how. So it's looking at the age it's looking at the gender it's looking at the

4:17:17 geography. It analyzes what the comments are asking me to do. So then at the end they kind of like put all this stuff together and they give me the next 90 days and they give me some things to

4:17:25 look at. Overall, they're very similar. I do like the way that this was packaged more by Astra. It just feels a little bit easier to read and it also feels more visual. So I'm going to give this

4:17:34 one to Astra as well. And you can see here though, Astra did spend more and take longer. So 42 minutes compared to 17 and 27 bucksish compared to 9 bucks. So Astra wins, but it was longer or

4:17:45 slower and more expensive here. But anyways guys, that was 15. That was 15 different use cases here that we scored and looked at the cost and time. Astra won 10 of them and Fable won five of

4:17:55 them. But let's take a look at the total numbers here. I had both Astra and Fable give me these stats to make sure they were consistent. And you can see the numbers are exactly the same. Fable was

4:18:05 9 hours 35 minutes and 45 seconds of total run time here, costing us 513 bucks.36. And Astro was 11 hours 19 minutes 24 seconds and cost us 326 bucks.98. So in

4:18:19 total Astro was $186 cheaper but took an hour and 43 minutes longer to run across these 15 different use cases. Now I will say this was 15 use cases. In general my experience has

4:18:31 been that Astra feels quicker and feels more efficient. And also it doesn't give us that 5 hour window in codec and it doesn't give us that fable limit. You know we can just use Astra the whole

4:18:41 week. So, I do like that we also get more inference with Astra in general because of the codec subscription. So, right now, I'm heavily leaning towards Astra for my my day-to-day and my use

4:18:50 cases. But the other thing I want you guys to keep in mind is that GBT 5.6 Soul is so so good still. It's so capable and for the majority of my knowledge work, I could use that just

4:18:58 fine. And that even honestly might be overkill for the majority of stuff that I'm doing. So, it's not like you need to use Astra only for all your use cases. And it's not like you need to go grab

4:19:06 Fable 5.1 for all your use cases. I'm still a very firm believer in they both have different strengths and weaknesses. They both have value and that's why I love doing things like this to find out

4:19:15 what are the different use cases and how do they, you know, behave in different scenarios. And so yes, I am going to be using codecs more right now because that's where it currently is. But this

4:19:23 stuff shifts so fast and I will always keep both subscriptions or multiple of both subscriptions so I can keep testing them out as things change, as my use cases change. Who knows what might

4:19:32 happen. All right, so that was Codex versus Cloud Code and it was also Astra versus Fable and Astra and Fable are the models there. So we talked about how the harness cloud code or codeex is

4:19:41 different than the actual model inside. So that's one thing to be aware of as far as like the difference between enthropic and open AAI and whatnot. The model is different than the harness. But

4:19:49 besides that, I also think it's important to understand that these tools are all going to start to sort of merge into the same standards. At least I think so. Here's one really good

4:19:57 example. Thoric from cloud code came out and tweeted and said hey we're adding support for agents.mmd inside cloud code which as we know that is what codeex looks at for sort of like the system

4:20:06 prompts or the instructions and the point I'm trying to make here is just that is cloud code changing the harness a little bit so now everything is being even more tool agnostic so don't worry

4:20:14 stay flexible and speaking of staying flexible I want to tell you guys about some of the mistakes and some of the lessons that I had learned when I was trying to be as flexible as possible and

4:20:22 selling automations to tons of different businesses tons of different clients because if that's something that you are motivated to do after watching this course or maybe you're trying to start

4:20:30 an agency already. Then let me talk about some of those lessons and the mindset shift of how you approach actually helping a business grow with AI rather than just trying to sell them

4:20:38 some template or some automation that you tried to copy off a YouTube video. This is how you can actually provide way more way more value to clients and that means that you're going to get bigger

4:20:48 retainers and you're going to get more referrals and you're going to get better case studies. It's just going to help your business in the long run. So excited to share with you guys what I

4:20:56 learned here. So right now I'm at this event in Montenegro called Workless AI and there are a bunch of AI content creators, AI experts and a bunch of community members. And on the main day I

4:21:05 closed off the day with this presentation which I'm about to share with you guys. So hope you guys enjoy. Hope you find it insightful. I'll see you guys in there. This is called Still

4:21:13 Relevant Next Year. And it's actually kind of funny because last night I kind of switched up what I was going to talk about a little bit because yesterday in the VIP sections and I talked to so many

4:21:21 of you guys and I just kind of had a feeling that this would resonate a bit more because whether you are I mean I think a lot of you guys are building kind of like an AI consultancy and AI

4:21:29 agency or working with clients. Um but even if you're not I think that this is the way that you should be thinking about using AI within your own business. And um yeah so let me just although also

4:21:39 um I hope this picture that's Montenegro, right? Okay. That'd be awkward if it wasn't. I I wasn't sure. It was Codeex. Threw it in there. So, good to know that that's Montenegger.

4:21:47 Okay, cool. So, let me start with a little story. So, I'm I'm 23 years old at this point. I just quit my full-time job. I just closed this deal with a client and I deliver something and and

4:22:00 he messages me on Slack and he's like, "Hey, can we get on a call?" And this is what he says to me as soon as I hop on the Zoom. He goes, "We are not getting the value that we paid for." And this

4:22:09 really made me think because, you know, I started to question like, did I just make a huge mistake? Am I not actually able to build these solutions? Am I not going to turn this into a

4:22:17 business? And so I realized that the reason that this happened was because, well, first of all, I made this video which went viral. It was like a personal assistant. And so he saw that, lots of

4:22:26 people saw that and they wanted this. But I made three big mistakes here. And that's pretty much what I want to talk about today are these three big mistakes. So the first one is that what

4:22:36 I built for him, the personal assistant was not the constraint in his business. The second one, I guess you guys saw the third one. The second one is that I didn't choose a KPI.

4:22:47 And the third one is that I guessed on the price. So like I said, that's what I want to talk about today. I had a big realization here before I started to scale my business and I was doing

4:22:55 everything in the agency myself all the way from the lead genen all the way to the, you know, invoicing, all of that kind of stuff. I realized that our job is to be the ones to diagnose and prove.

4:23:07 You know, it's one thing to just be kind of like you take the ticket, you develop whatever they're asking for. You kind of evolve from that and then you start to be the doctor and you start to diagnose

4:23:15 the pain that they're feeling. But then you're still not done yet because you have to prove the value. Because if you don't prove the value, then how are you going to win more business? And I've got

4:23:24 another quick story to tell you guys about in just a sec. But first, I wanted to start off with if you guys don't know me, a little bit about my background. My name is Nate. graduated from the

4:23:31 University of Iowa in 2024 and I was studying business analytics and marketing. So don't come from a technical background although I did work with data quite a bit. So I've got a

4:23:39 good understanding of data which was very helpful in automation. So can't lie about that. Um from there I graduated and I was working at Goldman Sachs as an analyst and I was automating a lot of

4:23:49 stuff and what that helped me realize was that automation is not new at all. It feels new because the word AI has been put in front of it and that's what a lot of business owners are kind of

4:23:58 waking up to. And it's funny because when I first started running my agency, I used to joke with my friends and you know some of my my partners when I started when I brought on some partners

4:24:06 that we're not even an AI agency. We are a data infrastructure agency and then after that we're just a boring automation agency. And I think that's a really good realization to have because

4:24:15 that the whole reframe of diagnosing the problem is that people might come to you and they want an AI agent because they saw a YouTube video that went viral or

4:24:22 because they saw something on LinkedIn, but your job is to figure out what do they actually need? Um, and remember how I said the thing about you have to prove the value. I was

4:24:32 building a lot of automations at Goldman and actually you know that's on the next slide. I'll talk about that in a sec. Um, sold my agency for so my agency once we scaled it past $100,000 a month. It

4:24:42 was purely because I was way more passionate about education. Um, but the reason we were able to get to $100,000 a month is because the minimum to work with us was a $20,000 a month retainer.

4:24:50 And the only reason we were able to do those retainers is because of this exact kind of like three-step framework of working with clients. Um, I've got a YouTube channel. It's pretty much the

4:24:58 top of funnel. That's where I am doing a lot of the education. And then my free community called a automation society is where we try to get people into and try to build community. So, real quick, I

4:25:08 wanted to start off with like why did I bring up the Goldman thing? What did that teach me? Number one, like I said, boring is beautiful. Deterministic automations. They're easier to build.

4:25:17 They are easier to evaluate and they are less risky ultimately because they're deterministic. You know the inputs and you know the outputs. Two is that big companies I was you know Hostinger is

4:25:27 doing a lot of stuff with AI which is which is awesome but typically when I've worked with companies and when I was at Goldman companies just kind of move slower. They have a lot of change

4:25:34 management issues and obviously in a or an institution like Goldman in that type of vertical there's a lot of regulation. And then the third thing is that you need to spotlight your value. So quick

4:25:46 story here. I remember I was building a lot of automations. Like I said, I built a few like weekly report things. I built a few dashboards and I was on a call with my entire team. So probably 20 25

4:25:57 people. And I was showing them this automation, you know, on a Zoom call. So your name is right there under your face, right? And I get off the call. It was also my job to send out the meeting

4:26:04 minutes, which obviously we know AI can do now. But I sent out the meeting minutes and I got a response from one of the managers that said, "Good job, Arthur." I'm like, "Who is Arthur?" my

4:26:13 name is right there and I sent you the meeting minutes. And that just made me realize that if you're not explaining to the business owner, to the stakeholder what your automation actually did, then

4:26:23 it's hard to get credit for it. So, you could potentially be in a situation where you're making the business tens of thousands of dollars, but they don't actually know that or acknowledge that

4:26:30 because you didn't prove it. And it's not bragging, it's just proving what your systems actually did. So, those are three big lessons that I had that I tried to carry with me through my AI

4:26:39 automation agency journey. Now think about it like this. So truly the only way for a business to scale is to remove constraints. And it's very simple at a high level. Obviously

4:26:51 there's a lot of things that you know a lot of nuance once you get into the weeds of that. But that's at the end of the day if you want a business to scale. You have to remove constraints.

4:27:00 You might get a lot of people coming to you like I did that says, "Hey, we need an AI agent that does X, Y, and Z because they saw a viral video because their board is putting pressure on them

4:27:09 to have an AI strategy because their competitors are posting on LinkedIn saying, "Look at this AI agent that we built that's doing all this for us." And so that makes me feel like what they're

4:27:18 doing is they're actually just buying relief. They're buying the ability to say, "Yeah, we work with an AI agency. Yeah, we have an AI strategy. We have this agent going to production. We have

4:27:25 blah blah blah." But that's not really what you're trying to provide. You're not trying to provide relief. if you're trying to provide real business value. So, the whole idea is to reframe. I've

4:27:35 done probably 300 or more discovery calls and what I've realized is that they come to you and they say something, but that's almost never what they want. And so,

4:27:45 your job is to figure out why did they book in the call because they booked in for some reason, right? They had they're feeling pain or they're feeling pressure. And so, you're have you have

4:27:52 to figure out what is that actual pain. You kind of have to take a step back there. There's pain somewhere and that is your job. You are paid to solve the pain. And the more pain you can solve,

4:28:01 the more money that your business will make. I think this might have might have died on me. Okay, so the reason I put this up there is because think about if you were

4:28:12 to work with an agency that was going to help you run ads. It's super super clear. Like you know what is your budget for ads? $10,000. And you in a direct correlation because of you ran those

4:28:23 ads, you made $50,000 in sales. That like the agency doesn't even have to prove that to you. It's very clear, but it's not super clear when it comes to automation. So, you have to make it

4:28:32 clear. So, those are the three things I want to talk about, right? Constraints, KPIs stands for key performance indicator, and then how do you actually price this

4:28:42 stuff so that you can clearly show this is how much you paid and this is how much you're going to make your business back. Okay, so starting off with the

4:28:50 constraint, the only way for a business to scale is to remove constraints, to directly attack them. And at a very high level, your business

4:28:59 is either going to be supply constrained or demand constrained. Meaning supply constrainted, too many customers coming in and you cannot service them. Demand constrained, you need business. You have

4:29:08 no people coming in. So I really like to think of this as a pipe. The water coming into your pipe is revenue. It's cash flow. And you obviously want that pipe to be clean, no

4:29:21 leaks, and you want lots of water to go through. So the top image we see is a pipe that has a clog in the middle. And this is probably a supply constrained business because they're having a lot of

4:29:31 water come through, but the water isn't making it all the way to the end. Maybe it's leaking out. It's not coming out the other end of the pipe. The bottom one is one where there's no water in the

4:29:38 pipe. And right now there's no clogs. There probably are. There there always more clogs. But right now this this business on the bottom is demand constrained and they need more water

4:29:46 inside their pipe. So let me tell you another quick story. I had a med spa owner that came to me and she wanted a lead generation AI agent. She wanted to make more money and

4:29:57 I saw the stat the other day. Um, couldn't be outdated by now given how fast the space moves. But the stat was basically that over 50% of all AI automations are like lead genen or sales

4:30:07 or marketing automations. And I think that makes sense because what is the pain a lot of us are feeling is that we want to scale our monthly recurring revenue. And usually the first thing you

4:30:16 think is like oh more leads, more clients. But in this case, the med spa owner wanted more leads. And when I really tried to dig into the pain there and figure out where the constraint was,

4:30:26 the constraint was she ended up telling me that they had lots of leads coming in consistently, but what happened was they weren't showing up to the appointments. And then all of the people that had

4:30:37 showed up to an appointment, there was no follow-up. So the LTV was extremely low, meaning there was a decent amount of water coming in the pipe, but it was just leaking out before it ever made it

4:30:44 to the end. And so we realized together on this call that it would be much more valuable to the business to build reactivation systems, to build follow-up reminders on their appointments so that

4:30:55 people actually started to show up more. And that's just a perfect example of the reframe that you have to do in order to really diagnose what the true pain is. And something interesting here is I

4:31:05 think about a slide that says this later too. There's just a lot of power in silence. So when I asked her the question of like why did you book this call? Like what is truly the pain that

4:31:13 you're feeling here? I stopped talking. She thought for a couple seconds, responded, and then I didn't respond yet because I I had a feeling that that wasn't true. And when

4:31:22 you have that silent pause, they think and they feel like they need to talk more. And I'm not big into like the sales tactics, right? Like I I don't have the sales background, but that was

4:31:31 super powerful. I would pause and they would just keep talking. And your job on the discovery calls is to talk as much as least as possible. Your job is to sit there and listen. And that's how you're

4:31:39 able to diagnose. So that's one quick example, right? But that happened all the time. Now, if you guys were here yesterday on the VIP VIP sessions and I came around to the table, some of us dug

4:31:49 into this, some of us didn't, but there are two questions that I love to ask business owners or even your own business, right? Like if you're trying to help automate your own business and

4:31:55 remove constraints, that works super super well, especially when you pair it with silence. So, the first one is if a business is supply constraint, if you had 10x the

4:32:07 business tomorrow, what would break first? You ask them that and then you just be quiet because they'll think about the flow in chronological order from start to finish. And the key word

4:32:17 there that I put in all caps is first because whatever breaks first is the first clog. Now, whenever you diagnose and you fix that first clog, there's going to be another one. There always

4:32:26 is. And that's the beauty of it. And that's how you get on a retainer because you just keep solving constraints. You keep scaling the business. So that's really good. If there's supply

4:32:33 constraint, if they're demand constraint, you basically just flip it. You say, "What would you do to 10x the business coming in tomorrow?" How do you get 10x the amount of water into your

4:32:41 pipe tomorrow? So, whatever they're already doing to get leads or maybe they're not doing anything to get leads, building systems to support that initiative. And

4:32:50 that doesn't sometimes what you'll realize is as an AI consultant is what like a lot of us are calling it right now. Your job is just as much to tell them, hey, this is the problem that I'm

4:33:00 seeing and we can solve this with no AI. So, sometimes the problem could be it's a very service-based business. they're supply constrained and so the key is to just hire more people. Now, yeah, maybe

4:33:10 you could come in here and say, you know what, let's build you some recruiting agents and some screening agents. That helps the hiring process. That removes that constraint. But that's a really

4:33:17 great way to think about it as well is that an AI consultant or someone that's AI native, it doesn't mean that you use AI for everything. It means that you know just as well when not to use AI

4:33:26 because that perspective is even more valuable in a world where AI is being shoved down everyone's throats and then you just be silent. Okay. So, moving on to the second one

4:33:36 here. We've got the KPI. So, the whole idea here is that you before you accept any payment, you want to choose one number that you're trying to move and you're not choosing this

4:33:47 number, you're choosing it with the stakeholder. If you think back to that example earlier where the business owner said to me, "Hey, we're not getting the value we

4:33:55 paid for." It's because we didn't do well, all of these things, but this one very specifically because I had no way to prove that a personal assistant was helping his business. He came to me, his

4:34:04 pain was, I feel busy. What is the solution in my mind? A personal assistant. When we really dug into it, the reason he was so busy is because he was handwriting all his proposals. So,

4:34:12 the better automation there would have been for me to help him build an AI agent that would automate the proposals or at least get them 90% of the way there. He checks it, he approves it. If

4:34:20 we would have aligned on a number beforehand, then I know what I'm working towards. I know the result that I'm trying to drive for the business. Because productivity is really

4:34:30 interesting. Productivity is actually moving the needle. And you can't move the needle if you don't know what you're trying to move it towards. Because a lot of us

4:34:39 feel like you might have had a day where you sat down on your at your desk all day, eight hours straight, let's say. But maybe you watch a lot of YouTube tutorials. Maybe you played around with

4:34:49 image generation for an hour and you felt productive because, you know, you you clock off and you you sat there for eight hours, but how many of those hours were you actually doing what you need to

4:34:57 do to move that number forward? And for us, you know, a lot of us, you know, it's a business, so there's a lot of different north stars, but knowing what you're building towards that quarter or

4:35:05 that day or whatever it is helps you actually be a lot more productive. And if your agents know your goals, then they can help you plan your days better to be more productive. So, something

4:35:15 that I would like to ask is, okay, let's say we've scoped it down. We've we understand the constraint, and I forget the example I use. Um, anyways, let's say we we've scoped it

4:35:26 down to a certain automation. Then I could ask in a perfect world, what does this look like for your business? If we push this into production and it's working perfectly, what do you want to

4:35:36 see? And from there, you try to drill into one certain metric. So maybe it is an AI agent that sets appointments. Maybe they say, "Okay, yeah, I mean, right now we're getting like five

4:35:46 appointments a week from our manual setters." In a perfect world, our manual setters have those 10 hours a week back to do other things, and we're booking in like 10 appointments a week. And so now

4:35:57 I know exactly what is a success to them. So there's that's completely objective. I can go work towards that and I can prove that. So if someone says to me, "Hey, we're not getting the value

4:36:05 we paid for. Look at the scope. This is what we agreed on. This you said that you wanted 10 appointments a week from this automation and I objectively delivered that to you." I probably would

4:36:14 say a little bit nicer though. Okay. So that's a trackable and objective metric because a lot of this stuff is very subjective. I want a personal AI assistant so I can feel more productive,

4:36:23 so I feel less busy. There's no way that I can actually prove that. There's no way I can prove to the business owner that he is now less busy because of what I built. So, they can't be subjective.

4:36:31 So, like I said, you have a baseline like five weekly appointments. You've got a target like 12 weekly appointments. And now you know what you're working towards. Of course,

4:36:39 you're going to have to leverage some of their expertise because if you know not much about the process and you try to set a baseline target of like, you know, 35 appointments a week, that's very

4:36:47 tough. And now you're not going to be able to prove that. But the idea is that each automation is having something to work towards. And like I said, if this is not you working

4:36:54 with a client, before you go out and decide to dedicate some some of your own internal resources or budget to a project, you should probably figure out what is that one metric you want to

4:37:03 move. A lot of times there are, you know, it it drips into other metrics too, but if you can choose one, it makes the decision a lot easier. And then you can figure out how is that actually

4:37:11 impacting the bottom line of the business. And that's how you prove, like I said, that your automation helped. Okay, the third piece now is on the

4:37:20 pricing side. Now, this is really interesting. I really don't think there's like a standard universal right way to price. Everyone's doing it a bit differently.

4:37:30 Even the big firms are changing up their pricing. I've tried almost everything that you could think of. I've tried performance-based. I've tried hourly. I've tried retainers on hourly. I've

4:37:38 tried value based. I've tried just lots of stuff. Um, the way I feel is that you should not be charging hourly. I think that a lot of people when they're getting

4:37:50 started, I hear, you know, I keep getting nos to my price. I hear that there's imposter syndrome. And so, I think hourly to start is a really good way to get your foot in the door and win

4:37:59 some trust or even start free. I think it's a really good way to start, but I don't think you should be building a business around your hourly pricing. And it's it's really simple. And especially

4:38:08 if you start to think about the way that AI is supposed to make us do more in less time. So, it's all about the incentives. All right, real quick, guys. I mentioned this at the end of the

4:38:18 presentation, but I attach this completely free PDF, which is like a 30-page resource on pricing your AI solutions, which you can access for completely free inside of my free school

4:38:26 community. The link for that is down in the description. You just hop in there, you'll click on all YouTube resources, and you'll find that pricing masterclass guide in there. So, if you want it, it's

4:38:36 there. Let's get back to the video. Let's say you've got two developers on your team and you're providing services to businesses. You've got one developer named Sarah and she's a killer. She gets

4:38:46 something done in 1 hour. And you've got another developer named Dan. He's good, but he's a little slower, three hours. And let's say for the sake of the example, they produce the exact same

4:38:55 quality of solution. You would then bill for Sarah's work, your better developer, you would bill $150 to the client or I don't know, however you want to, you get the point, right? $150. And now Dan,

4:39:06 because he took three hours, you get more money. And those incentives are completely backwards. Why would you make more money for moving slower and on a worse developer? Right? It's all about

4:39:19 incentives. It's all about value based pricing. That this is where we got to and this is what I believe at the moment is the best way to be doing this. And what that means is yes,

4:39:30 the same automation could be different amounts of valuable to different businesses. So you have to kind of take that into account at least when you're going very

4:39:39 custom and you're consulting. This is the way we did it, right? Oops. A client should understand how they will receive 10x their investment. That's kind of the golden rule that I like to

4:39:49 to give, right? Because if you can walk the stakeholder through the math, whether that is a business you're trying to help or whether that's you internally trying to win a budget because you're

4:40:00 trying to, you know, build AI projects inhouse, you need to prove that this investment of their money will get them a 10x on it because that makes it a really hard

4:40:10 offer to say no to. So the way that we would do this is basically take 10% of the projected annualized value of the price of the the automation and that's a gear price. So an example here, let's

4:40:23 say you're building a customer support agent and the manual rep, the human rep spends about 10 hours a week responding to these tickets and this person is paid 40 bucks an hour. So the cost of the

4:40:34 business 40 bucks an hour for customer support. 40 bucks an hour and the rep is does this about 10 hours a week. That's 400 bucks a week. And if you annualize that, that's about $2,100 over the

4:40:46 course of the year, the first year projected annualized revenue. And you take 10% of that. Now, 10% 20%, I play with it a little bit, but let's just say 10% for the sake of example,

4:40:56 that's $2,80 here. Because what you have to assume, they don't always, but you have to assume that they're going to ask you,

4:41:08 can you please tell me how you got to that number? Because a lot of people will come in and they're kind of like maybe they're shopping around. They're trying to get a quote. They don't even

4:41:14 know how much to allocate a budget towards. So if you assume that they're going to ask you, "How did you get to that number?" And you can confidently walk them through the math and then you

4:41:22 just be silent and you don't devalue your work. I'm kind of jumping ahead. I know I got some slides here that cover this, but anyways, you have to figure out what does the manual process cost

4:41:30 the business. And so in this example, right, I I was able to understand, okay, this rep cost the business 40 bucks an hour. They're not always going to tell you salary. They're not always going to

4:41:38 tell you the hourly weight rate, but there's other ways to figure out what does the manual process cost the business. It could be opportunity cost. Maybe they've been having a different

4:41:49 funnel or trip wire that they've wanted to push out for two quarters, but they haven't because they are too busy, right? There's there's other types of cost besides just like the hourly cost.

4:41:57 Maybe you're building a voice appointment, uh, a voice agent because the business misses calls all the time and every missed call could potentially be um, a driveway, right? A driveway and

4:42:09 that driveway could win the business $10,000. So, there's other ways to figure out what is the cost of the manual process. You just have to ask good questions and then be silent and

4:42:19 listen, like I said. So, the key thing to remember here is that cost doesn't justify price. value justifies price. And

4:42:30 this was kind of tough for me to wrap my head around at first because the first thing I thought of as a very new business owner was I need to protect my margins here. I need to make sure that

4:42:38 I'm making this much money or this percent of the deal and I'm keeping that much to account for taxes to account for, you know, what I have to pay my employees to account for our API cost or

4:42:47 our service, whatever it is. I was thinking about the margins, but I think that obviously you still have to think about the margins so you're not running your business into the ground, but the

4:42:55 cost doesn't justify the price. And a really good example that I heard from Jonathan Stark who spoke at our event a couple weeks ago, he was saying like let's say you're paying someone to do

4:43:03 your lawn and you pay them a hundred bucks for the lawn and he's been come, you know, he's been coming to you for for years. All of a sudden he buys a new pickup truck and that's an expense to

4:43:13 his business. So he has higher costs on his end and then he comes to you and wants to charge you 200 for the exact same job for the exact same lawn. That's not something that people will typically

4:43:20 agree to because the cost doesn't justify the price. The value justifies the price. So, like I said, you have to pretend and you have to assume that as soon as you

4:43:28 say the price, they're going to come back to you and ask, "How did you get that number?" And then you just have to walk them through the math. Now, what's really important there is you don't want

4:43:36 to devalue your work. So, let's say someone comes to you, you scope out the right constraint, you figure out the cost of the business, and you give them a 10% number. Let's say it just comes

4:43:44 out to 15,000. And they were basically planning 7,000 to be the maximum they were going to spend on AI this quarter. You don't want to get into this negotiation of devaluing your work and

4:43:54 saying, "Ah, you know, this would be 15K, but for you, we'll do it for 7K." You know, if you're really trying to get your first client and you can get that, then I would say do it reps over cash

4:44:02 upfront when you're getting started. But you don't want to devalue your work. What I would do in this situation is I would say, "Okay, this build, this solution is $15,000

4:44:12 worth of value to the business and you have a $7,000 budget. Let's just drill down the functionality a little bit and treat this as a V1. we'll give you the $7,000 worth of this automation and then

4:44:22 you know next quarter when you've got the budget or once this automation's in production and we're actively earning you back time and money then we'll look at the next one because I think if you

4:44:30 if you set the impression right away that you're willing to bump down your price as soon as they push back then they're just going to keep doing that and you don't want to devalue your work.

4:44:41 So that's pretty much what I wanted to run through today. Um, obviously those kind of three main things, there's a lot more nuance to get into, which if you guys want to chat more, I'd love to to

4:44:50 chat more. >> But think about this with your own business. Um, actually, let's just take let's just take one volunteer real quick. Someone that Yeah. Um, is it

4:45:00 Allejo? >> There we go. Tell us in your business, you know what you're driving towards, right? >> Yeah.

4:45:09 >> At the highest level, are you supply constrainted or demand constrained? >> Demand constraint. need more clients. >> You need more clients. And how are you currently getting those clients?

4:45:16 >> Uh content on YouTube. Uh it's not quite converting. I'm doing webinars and out of 20 people that have assisted webinars over two months, >> uh we've gotten zero appointments set.

4:45:30 >> Okay. So, if you wanted to have 10x the amount of business tomorrow, what would you have to do? >> Uh probably 10x the content.

4:45:41 >> Mhm. 10x the content. uh seems crazy but the methods that I see today would be very helpful. >> Yeah, volume negates luck, right? But you want to 10x the content without

4:45:53 decreasing the the quality of course. So in this case, obviously there's a lot more to dive into here, but you know the way that you're kind of you know your top of

4:46:05 funnel. What does it seem like you should be focusing on automating when you get home in a couple days or next week? Um the problems that I have heard on

4:46:14 sales calls uh turning those into a video. >> Okay. >> Uh that one is huge is I like forget what we talked about

4:46:24 >> and maybe focus on something in a sales call but I'm going to miss something else that they said that is an actual problem. >> Totally. And do you know how you could

4:46:33 automate that? >> Now I do. >> There we go. Um now something I saw I remember talking to you about was the grill me skill, right? Yeah,

4:46:41 >> if you guys don't know the grill me skill, I mean, it's a very simple prompt. Essentially, just at the end of of questions, ask the AI to interview you relentlessly about this topic

4:46:51 because what that does is it gets all of the subject matter expertise in your head into the AIOS that you're building because once you give it all of that nuance, it's able to build an automation

4:47:00 out of that so much better because the way that you would kind of create these AI agents to to uh, you know, operationalize that process is different than someone else might. So obviously

4:47:10 that was a really quick example. Thank you Allejo. I appreciate the theuh participation there. But just constantly be asking yourself like what is the actual constraint? You know how do we

4:47:18 10x the business? How do we actually look through the flow of the pipe and the water and figure out where the clog is or where the leak is and then just directly attack that? And so like I said

4:47:27 earlier when we were working with businesses we would always start with our foot in the door with one project on a value based uh project fee. And from there, we

4:47:37 started to explain to them, we are seeing all of these other clogs because once we've removed this constraint, there's another one. And that's a good thing because still you're

4:47:46 getting more water through. And that's how you're able to get on the retainer because not only did you find the right constraint, you also chose a number beforehand so you can

4:47:55 objectively prove to them the automation helped their business. And then you price it in a way where it was a no-brainer for them to say yes. And then guess what? after you push it into

4:48:03 production, you come back 1 month, 2 months, 3 months later and you show them that number and now you're able to justify asking for that retainer and to just say, "Hey, you know, we can move a

4:48:13 lot quicker if we're not always doing the scoping and we're not always figuring out pricing, if we can just work more and just keep attacking those constraints." And that's the only way

4:48:20 that the business is truly going to grow. So, that's pretty much what I want to talk to you guys about today. I have a full pricing guide. This is like a 29 30 page doc. Um, kind of just like about

4:48:30 what I talked about today on the valuebased pricing and some stuff like that. Um, so go ahead and scan that and grab that doc. I'm sure we can include it in the the resources later as well if

4:48:39 you miss it. But yeah, thank you guys so much. It's been an awesome day and yeah, let's let's have some dinner. So, I hope you guys enjoyed the speech that I gave as well as that free pricing

4:48:50 guide resource. I also wanted to say that I have a second channel right here where you can see this was actually the workless event that I just showed you guys that speech from. And I have the

4:48:57 second YouTube channel where I kind of show some more behind the scenes stuff when I go to AI events or when I do other cool things in the space and just showing you like the other side. So I

4:49:05 will link this YouTube channel down in the description if you want to check it out. But now before we wrap up this whole course, let me dive into even more about how you actually think about

4:49:12 pricing these solutions with valuebased pricing. So let's hop into the last video of the course. All right, so I've sold over a 100 AI automation systems and I've priced a ton of those wrong.

4:49:22 I've undercharged, I've underscoped, and I've just thrown out random numbers that were kind of a guess and I couldn't explain when someone actually asked me like, "Hey, where'd you get that number

4:49:30 from?" So, in this video, I'm going to talk about everything that I know about pricing AI solutions. I'm going to walk you through one real build that I sold, every single number, and by the end of

4:49:39 this video, you'll be able to take any project, turn the client's own numbers into a price that you can actually defend, and then get paid in stages, so you're never carrying much more than,

4:49:47 you know, 30 days of unpaid work. And if you don't know who I am, my name is Nate. I've been teaching hundreds of thousands of people how to build AI agents and how to implement them into

4:49:54 businesses as well. And I scaled my agency to over $100,000 a month, and then I sold it. With our business, we got to the point where the minimum to work with us was a $20,000 a month

4:50:04 retainer. And I'm assuming that sort of engagement is where a lot of you guys want to end up getting to. So, let's not waste any time and let's just jump straight into today's video. Okay. So,

4:50:13 this example was an appointment setting agent. The business had employees manually setting these meetings and this was about 20 leads a week, each one taking roughly an hour of a human's

4:50:23 time. And those employees were costing the business about 40 bucks an hour all in. So 20 hours a week at $40 an hour is about $800 bucks a week. And if you multiply that by 52, which is 52 weeks

4:50:34 in the year, that's $41,600 annualized. And before I said anything to the client about price, I walked them through the entire solution. You know, how the agent would work, what testing

4:50:44 looked like, how it would change their speed to lead, and their lead quality. But I also had to ask them a ton of questions through what we call the discovery phase. Because some clients

4:50:52 want to talk about price right away, but the honest answer to them is, you know, in order for me to give you an accurate ballpark estimate of what this is going to cost, I really need to understand

4:51:00 more in depth the complexity of the system and how it will actually look once it's fully integrated into your actual business processes. Then I priced the build right around 13% of what that

4:51:10 annual number was, which came out to 5,500 bucks. So, the business would basically be paying $5,500 for a system that over the course of the year was going to give them back $41,600,

4:51:22 which comes out to about a 7.5 multiple on their initial investment. Now, typically, I like to say the golden rule is that you need to be able to show the client that their investment is going to

4:51:32 10x over the course of the year because that math makes it really hard to say no to. So, in this specific example, where I accounted for that extra 2.5x that we were missing that would put us at 10 was

4:51:43 the baseline. So right now the baseline was about 20 leads a week. But as the business earned more time back because of the system and this whole process gets more streamlined that baseline of

4:51:53 leads per week would probably start to increase right it would go to 21 and then 23 and then 27 and that's where we are earning them even more money back. Now obviously something like that is a

4:52:03 projection and you can't just go in there and say that you can guarantee that. So be careful about making any guarantees or trying to tie revenue to those specific results or anything like

4:52:11 that. But the general idea is that as the system gets used more, the business will grow, which in turn utilizes the system even more. So you create this really cool like flywheel. And then on

4:52:22 top of the build, we set up a standard maintenance plan. So this was 400 bucks a month. And I want to be clear what that actually is because a lot of people kind of misinterpret this word

4:52:32 maintenance. So that $400 a month isn't for me to bolt on new features every month. It's just me guaranteeing that the build keeps doing what we agreed that it would do. Meaning, if something

4:52:43 breaks or an API changes or, you know, maybe a new model releases or some weird edge cases show up that need a little tweak in order to keep the system living up to the functionality of the scope,

4:52:52 that would be on me and that's covered by the client's retainer fee. But new functionality is a completely separate conversation. So, with maintenance retainers, I always keep the number

4:53:02 super simple and standard across all projects. And at this point in my career, our maintenance package was 400 bucks a month. You just do want to be careful though because if you're

4:53:10 delivering a system that's, you know, let's say $30,000. If maintenance on that solution is going to cost you and your team more than 400 bucks a month, meaning you would be like losing money

4:53:18 on giving away that package, then obviously you can't charge that. But I think that with well-designed automations and well-built automations, it really shouldn't be too timeconuming

4:53:28 for you to maintain them. because once again there's a big difference between maintaining and adding minor feature enhancements and adding you know more functionality. Now a big mistake that I

4:53:37 made in this specific project was that once I was in production what I should have done was gone back and measured how valuable this thing truly was and I didn't do that. So I had no after state.

4:53:48 I had no transformation number and what ended up happening is that cost me on the next conversation when I tried to win more business. So make sure you're capturing the baseline up front which in

4:53:58 this case was you know like 20 leads a week and the speed to lead time because that was taking the humans you know time and then you follow up after a month after 2 months and after 3 months and

4:54:08 you prove to them that those numbers are moving in the direction that the business wants because of your system. And I know that this might kind of feel like bragging, but it's not. You know,

4:54:16 because if you don't put a spotlight on those numbers, even if your automations really really are helping the business, the business owner might not actually feel that and won't acknowledge that.

4:54:26 So, it's really important that you're the one surfacing that stuff. Okay, so the obvious question is why not just bill hourly and be done with it? And I want to give a quick shout out here. A

4:54:35 pricing guy named Jonathan Stark came and spoke at AIS Live and he made some really amazing points on this topic. So, I'm going to talk about some of the things that he brought up, but really

4:54:44 the problem with hourly is that it pays you more to be slow. I mean, picture two people on your team. You're billing them the same 150 bucks an hour. Your best developer and your slowest one. Some new

4:54:55 feature comes in and your best guy knocks that out in a day. So, you bill one day, but your slow guy takes 3 days. So, you bill 3 days. You just made three times the money off of a worse, slower

4:55:07 employee. And if your best guy gets faster, then you're going to make less money off of him. So getting good at the job actually cut your income. It's all about incentives. Would you want to hire

4:55:17 someone who was incentivized to work slower in order to get more money out of you? Probably not. I wouldn't want to. But now think about that in, you know, 2026. You get really good with cloud

4:55:26 code or whatever tool you're using and you can do something in an afternoon that used to take you a week. And if your hourly pricing, then that afternoon because you moved so fast just slashed

4:55:36 your income. So, I'm not saying hourly is never okay. I think for your first two or three projects when you've got, you know, nothing to point at, not a bunch of proof, then billing hourly is a

4:55:45 safer ask. It's a it's a easier ask for the business owner and it's a good place to start. Maybe just like a hundred bucks an hour. But after that, quit billing hourly. By the way guys, if you

4:55:54 want to access this completely free resource, which is like a 27page doc on pricing your AI services, then you can grab that, like I said, for completely free by joining my free school

4:56:03 community. The link for that is down in the description. All you have to do is jump in here, click on classroom, go to all YouTube resources, and then find the resource in there. I promise you guys,

4:56:12 it's in there. But let's get back to the video. So, if you're not pricing off hours, then what are you pricing off of? Well, there's three things to keep straight. So, those three words are

4:56:19 cost, value, and price. So your cost is the floor, the number that you'd walk away if you know it was below that. Their value is the ceiling, the most that this whole thing is even worth to

4:56:30 them in the business. And the price is any number between the two of those that you guys can land on. Now on an AI build, that gap is pretty enormous because whatever it costs you to build

4:56:40 the thing and harden it, their ceiling is a higher they now don't need to make. And just remember this, cost doesn't justify price. Price justifies cost. And Jonathan made a really great example of

4:56:52 a landscaper here. So let's imagine a landscaper mows your lawn every single week for a h 100red bucks. Then one week he shows up. He does the same job he's always done on the same exact lawn he's

4:57:01 always mowed. And now he tells you it's 200 bucks because he bought a fancy new truck and his costs went up. That's not how it works. Nobody accepts that. Cost doesn't justify price. Price justifies

4:57:13 cost. Okay. So the client ceiling is the number you're actually pricing off. and you get one conversation to find it. So, as you're getting to know the business and getting to know about the

4:57:22 automations that they need, you're trying to uncover scope just as much as you're trying to uncover the value to the business. So, you open by letting them dump everything. You know, tell me

4:57:33 everything you know about this, what you've tried in the past, what's worked, where it keeps getting stuck. And your job is to just listen, repeat, and poke. And just run that LRP framework to get

4:57:42 as much information as you can. Take a ton of notes, and then you pivot. Ask them about what happens after we're done with this project. You know, what is the best case scenario? And what actually

4:57:52 changes for the business once this thing is successfully up and running? And then you run three buckets of questions. Why this? Why now? And why me? So why this is checking that what they asked for

4:58:04 actually gets them the outcome that they want. Because a lot of the time it doesn't. And because they, you know, walked in asking for an AI agent, but the real fix might be a really simple

4:58:12 deterministic script and a Slack notification. Then you ask why now? And this is about urgency because if a window is closing or if a competitor is breathing down their neck, then that's

4:58:22 real money to you. And then why me is the kind of uncomfortable one, but you literally ask them, you know, why wouldn't you just do this internally? Couldn't you vibe code this? Couldn't

4:58:30 you hand it to an intern? And the reason you want to do this is because these questions or that question specifically surfaces objections that are already in their head. So, either you can hear it

4:58:39 now while you're on the call with them and you can answer those objections and handle them or it's going to kill the deal later down the line when they're reading your proposal or talking with

4:58:47 their team. And when they give you a real answer like, "Yeah, you know, my nephew probably could build this." You don't argue. You go, "Yeah, he probably could get a version of this up and

4:58:54 running." But what I'd want to know is who's going to watch it at 2 in the morning when the model updates and you know, who's going to help you scale this when you have way more throughput than

4:59:02 you're expecting and when it gets a bit more complex? And then what you do is you stop and you just let them answer because their answer is the actual reason that they're looking to hire

4:59:10 someone. And if you find that you're in a situation where they just keep pushing you for a number, if they keep pushing for a price before you can ask these questions and before you can really dive

4:59:18 in, don't blurt one out. And honestly, if they keep pushing you for a price and all they want is to get a quote out of you so they can compare you across a few different other vendors, then I would

4:59:26 say that's probably a red flag and you just want to get out of that engagement. Because the hard truth is some people are just looking for the cheapest labor. they're not really looking for a

4:59:35 consultant or partner and that's how you want to position yourself and there's unfortunately not much you can do to change that person's mind besides telling them hey you know what I bring

4:59:42 to the table is different I bring a different level of expertise and you know that's where you really want to leverage your proof but ultimately some people know that they could go to Upwork

4:59:50 and get much cheaper labor and that's just what they're going to do so don't take it personally okay so the obvious problem with a lot of this is that a lot of clients won't just hand you the

4:59:59 numbers like they don't want to tell you salaries and exact bottom line numbers and things like that and and they don't have to. So, start with three questions. How long does this take you today? How

5:00:08 many people touch it? And what happens when it goes wrong? These are all sizing questions. They measure the ceiling, not the build. Then you keep stacking proxy questions on top, like you, oh, how many

5:00:17 locations do you have or how many of these come in on your busiest days? That kind of stuff. And I know some of you are on a job marketplace where you have to put a number in your proposal before

5:00:27 you even get a chance to talk to that person. And you can still kind of do a version of this. The job post almost always tells you things like volume or headcount or you can figure out some

5:00:35 complexity there, how often the thing happens. You size it off that and say your assumption out loud. Something like, you know, I've priced this assuming that there's about 200 of these

5:00:44 a month. And you know, if that's off, let's hop on a 15-minute call and I can adjust the price and we can rework the scope a little bit. But that basically just turns the cold price into a reason

5:00:52 for them to hop on a call and talk to you. And here's a quick check. If you can't land on a number, if you feel like you don't understand the value enough to accurately give a number, then that's

5:01:01 probably a signal that you're not ready to write that proposal yet. You need more information. Okay, so now you've got that value number from that total annualized value number, first year

5:01:11 annualized. What I like to do is pick somewhere between 10% and 20% of that number as my starting point. So you say your price, you assume the next sentence out of their mouth is, can you walk me

5:01:22 through how you got to that price? And you basically have to answer that confidently. You have to say, "Okay, yeah, this is a customer support automation. It takes a rep an hour a day

5:01:29 manually, and an hour of that rep's time is about 50 bucks, right? Like that's how much it's costing the business over a year. That equals about $12,000 to the business. So 10% of that $12,000 is

5:01:42 $1,200. So the build is $1,200 bucks. So conservatively, you will be 10xing your investment of that $1,200 just in the first year." And when you go to write this up, don't just send one price. What

5:01:53 I've always done is tiered packages. So you can have like a starter, a growth, and a scale or how whatever you want to call them. And before you get anywhere near the options, the top of that

5:02:02 proposal is three short paragraphs. And none of them are about you or your pricing. They're about where the business is right now in their words with their numbers in it, where they

5:02:10 said they wanted to be and why you're the person that can help them get to where they want to be because of your AI expertise and your systems. So you would then take the first year value. Let's

5:02:20 for now just call it 100 grand to keep numbers even. You could say option one is 10% so $10,000. Option two is 25% so 25,000. And option three is 50% 50 grand. And I'm just kind of throwing out

5:02:32 rough numbers here. But obviously as you move up those tiers you have to have different sort of, you know, value. It still has to make sense. You have to have more functionality or you have to

5:02:42 have different types of results tied to it, right? But the middle one is kind of the one that's designed to win because the bottom one looks thin, the top one maybe makes them wse a little bit and

5:02:51 the middle one looks more correct. And what this does psychologically is really interesting because if you give them one number, their decision is hm should we work with this person? But if you give

5:02:59 them three numbers, their decision is more around how should we work with this person? You know, like which one of these deals should we take? Okay, so one of the most profitable automations that

5:03:07 I've ever seen was one of the simplest because all it did was it took a construction crew's daily phone orders and converted them into the text format that the crew already used. That was it.

5:03:16 It was super simple. It only saved like 45 minutes a day, but it helped them avoid around $12,000 a month in scheduling errors. And that second number isn't freed up hours like the

5:03:26 earlier example was with the appointment setter. It's hard dollars that they were losing every single month because of errors, because of the inconsistency of humans. which is why the saving 45

5:03:36 minutes a day was worth so much more here. And the reason I'm telling you guys this is because typically a good place to start is by thinking about the hours you're saving. But as you get a

5:03:43 little more comfortable and you have a little bit more experience behind you, you start to think about the bottom line impact of the entire business. If you think back to the earlier example about

5:03:51 the appointment setting agent, we basically only attributed the value to the time we were saving to the time we were buying back the business. But what if we would have thought about how much

5:03:59 does a converted appointment actually make the business? So, all of these um appointments that we're helping set with our system, what if each of those closed sales was worth $5,000? We could have

5:04:09 valued that system way higher. And if we would have communicated it that way, we probably could have charged way more for that automation. So, like I said, as you get more experience, start to think more

5:04:18 about that. But once again, it's your job to communicate that value because the business owner isn't just going to see that like that immediately. Okay. Now, I want to talk about underscoping.

5:04:27 This is the number one mistake that I made when I first got started. And then what this caused was, you know, timelines to get pushed and arguing about milestones, right? So, here's how

5:04:38 I structure this kind of stuff so that that never happens. Now, this video specifically isn't about scoping. That's a different topic entirely, but this is kind of more about setting up the

5:04:46 milestones in the correct way. So, what I would do, let's say we have a $9,000 project and we want to split this up into major milestones. So, let's say we have one milestone at the halfway point

5:04:55 and one milestone, you know, once this thing has been pushed into production. And I like to think about those as far as like what do we think we could deliver in 30 days. So each milestone is

5:05:03 30 days apart and that's where you're going to get paid is essentially every 30 days. Then what you do is you could split this into three payments if we have two major milestones. One to get

5:05:11 started, the second one after the first milestone has been hit and the final one after the final milestone has been hit. But this only works if each milestone itself cannot be argued with. It has to

5:05:21 be so objective. It can't be subjective, right? Like there can be no ambiguity there, which we've run into a lot. So, you know, here's what an objective one could sound like. At this milestone,

5:05:31 there is an AI system in the business owner's hands, a PC, a proof of concept that they can actually talk to. And when the owner sends it a question, the agent pulls from the database and responds

5:05:39 within a minute. All of that stuff is provable and it's very simple to write down and the client could go prove it itself. It's not claiming that the database is perfectly optimized yet.

5:05:47 It's not even saying that all the answers are perfect or correct. It's just saying that it works and it pulls there and it responds and that's objective. Now, a subjective milestone

5:05:56 maybe could be something that's obviously easy to argue like, "Oh, the inbox agent is working as expected." Okay, what's expected? You and the client could potentially go back and

5:06:04 forth on that for weeks and you're not going to get paid and it's just going to get frustrating. And then what else could happen is they start to ask for things that weren't on the original list

5:06:11 because it's so ambiguous and they might say, you know, can you add this? Can you add this? That's not, you know, this is a milestone. I need this functionality. And so what you want to do there is if

5:06:20 they do start to try to scope creep, this is actually a great sign because it means that they're excited and it means that they're already starting to imagine working with you more. So the way that I

5:06:28 handle this, I don't say, "Oh, no." Right? Like I say, yeah, it's a great idea and I could definitely see how this would add value to the system. Let's go ahead and throw this on the backlog for

5:06:36 our version two of the project. And as you get more ideas, you can just add more stuff to the backlog because I just want to make sure that we're hitting the milestones that we've already set as

5:06:45 quick as possible for you guys so that we can deliver as much value to the business as possible. Okay. Now, let's talk about what you do if they see the numbers that you've presented your price

5:06:54 and they don't like them and they're saying, "Oh, you know, this is way more expensive than I thought it was going to be. I didn't really have a good gauge. This is not in our budget." So, what you

5:07:02 want to do is say something like, "Okay, it sounds like 20 grand isn't in the budget right now. Why don't we just like reduce the scope a little bit and start with a smaller project? And once this is

5:07:11 working and winning you guys back some time and some more business, then we can move on to the next piece together cuz we've kind of already got it scoped out. And that's how you can avoid teaching

5:07:20 them that your prices move down every time that they frown and you're also not devaluing the work that you would be doing. You're just reducing the scope and that keeps it pretty fair. And the

5:07:29 last thing that I wanted to address here was the other costs that typically go along with these systems which are API costs or cloud subscriptions and you know token usage things like that. Now I

5:07:39 have always said client's account client's card every single time your fee is for design consulting building testing. Now the tokens are a utility bill and utility bills go in the

5:07:50 client's name. I used to start off by running everything under my own billing and invoicing them each month and it just got super messy. I was babysitting the billing. I had to follow up with

5:07:59 clients. Obviously, I built agents to do that, but still, it wasn't fun. And it also left them with no idea what they were paying for and potentially misaligned, you know, some trust. And

5:08:09 what else you should do is probably give them an expected monthly run cost in the proposal with the volume assumption next to it. Now, obviously, not a guarantee, but an estimate of how much this thing

5:08:18 will typically cost per month when it's fully in production and how ideally the system starts to get used more and more over time, like month over month. So the costs are probably going to scale up a

5:08:27 little bit month over month as well. Now, one thing that you do want to think about in your pricing is that there are typically a decent amount of testing costs that go into the system. At least

5:08:35 if you're doing it right, you're spending a lot of money testing before you push anything into production for a client. And these testing costs could genuinely be anywhere from a hundred

5:08:42 bucks to a few thousand based on how rigorous your evals and your QA process is, which I think should be pretty pretty rigorous. So, I would usually just factor that in to our final price

5:08:52 by just bumping it up by like a,000 or 2 or $3,000 depending on the size of the automation and how much testing you think is going to go into it. Obviously, the more AI that's in there, the more

5:09:02 autonomy, the more testing you're going to have to do. And the whole thing, the way I feel about pricing right now is that even the biggest firms, McKenzie, Salesforce, everyone's trying to figure

5:09:11 out this AI pricing thing, and no one has the right golden answer. And I think that a lot of people might disagree with some of the things I'm saying here, and that's okay. But I didn't feel like it

5:09:19 was a great feeling to say, "Hey, you know, Mr. or Mrs. client, can you please go ahead and give us this API key and this one and this one and this one." And then before we're even giving them any

5:09:27 sort of P or showing them any value, we're already spending a few hundred of their dollars just testing the system. I just don't think that's a very good way to kick off a partnership. So basically

5:09:37 what I meant by that is in the testing we paid for everything and then when we moved everything into production we swapped out their API keys for you know compute and whatever else it was that

5:09:48 was costing us. Okay. So if you take one single thing out of this into your next discovery or sales call, make it this one. Before you say any number at all, get the client to tell you what the

5:09:59 problem is costing them in their own words. Just have them say it out loud. And then once that number is on the table in their language, your price is just a fraction of a number that they

5:10:07 have already stated. So what you're selling is the result and the build is just how it gets delivered. Last thing I wanted to call out for you all is that I have a road map, the exact playbook that

5:10:15 I followed to get to $100,000 a month and exit my agency when I started as a solo founder. So if you do want to take that path and sort of like run your own oneperson agency to start and then look

5:10:25 at scaling up, there's a free road map down in the description if you want to check that out. But anyways guys, that is going to do it for this one. I hope you enjoyed the video. video. I hope you

5:10:31 learned something new and if you did, please give it a like. It helps me out a ton. And as always, I appreciate you guys making it to the end of the video and I'll see you on the next one.