Your Business Isn't Ready for AI (Until You Do This)

0:00 Fidelity, a company managing 11 trillion dollars, is currently refusing to sell the most valuable thing it owns. It's not simply money, it's what the business knows. AI has made knowledge the most

0:13 valuable asset your business can own right now. And we all need to both organize it and protect it. In this video, I'll show you the four steps you need to take to start the process and

0:23 how a Cassa framework can help you repeat it across everything you do, making your most valuable knowledge truly useful to AI systems. Do this well and you will not only remove your

0:34 repetitive work, but you'll be able to replicate step four, how I built an informed assistant that briefs me every morning on everything that's happening everywhere. It's wild. If all that

0:45 sounds useful, as it should, make sure you're subscribed here. Stick around for this quick demo and check out the Aentic Business Accelerator that we've just opened this week to help you all do this

0:56 as part of our community and give you all the templated systems you'll see here to get you started. First, proof of why we all need to do this. Okay, why are Fidelity refusing to sell any of

1:07 their information? Because everybody can access the same public AI models. The difference is what you combine them with. And Fidelity has decades of transactions, customer calls, notes from

1:19 actual client meetings, and a huge amount of industry knowledge around 8 pabytes of it. Their head of data, Mahia Sha, put it rather neatly. Everybody gets the public model. The advantage

1:31 comes from combining it with your own context. Now, my business is not fidelity. I do not have 8 pabytes of anything. I have about 400 notion pages and a worrying number of halffinish

1:43 voice notes, but the principles are exactly the same. AI can already do quite a lot of my work. The harder bit is giving it the right context, supervising it, and knowing whether the

1:53 answers are any good. And connecting information therefore isn't enough. The value appears when knowledge is connected, current, and usable by someone other than you. So, let me show

2:04 you how I would start. four steps in order without migrating your entire digital life. So step one of the Cassa framework is to consolidate. It's gathering your important knowledge and

2:17 context into one place and atomizing the right parts of it so that it can be used and read. I'm going to show you this. This is my uh business OS system. I'm demoing it in this and in Claude where

2:31 you'll see some of the results later on. Focus less on how I'm using this platform in this video, more on the central principles. The first things I want you to focus on are your brand,

2:41 business voice, and business identity. This is essentially to personalize with what I'm going to call primitives, primitive guides you want your AI to use. I would start with doing something

2:53 like this. Start with one or two pages. one page that explains who you are, who you help, what you sell, and then a second on how you talk, and how you want to communicate. If you wanted to work

3:06 with AI in something like Notion, you can just go and find documents, upload those as context, and you can just dictate what you want to do. My favorite tool to do this is Whisper Flow, and I'

3:16 I'd say something like this. I want to create a brand uh identity for Simon Pitman and Better Creating. And you can see that it just drops it in. And then every time you speak to your AI, you can

3:29 make sure that it reads this page. The second example is something like this, a personal voice profile. So I have one of these for both my brand and for me as an individual. And it means that when you

3:41 get AI to write for you, it sounds like you. Essentially, I've written this down as a primitive. My core identity, who I am. I've got stuff on my belief and stance. uh three ne non-negotiables

3:53 about every bit of content I write should be useful, actionable, and non-obvious. And then content filters. This is stuff around how I create content because I'm a content creator.

4:02 But you might put this around how you speak to clients or the kind of ethos you like to follow. Now, if all of this looks a little tricky to do, here's my hack of how I did it. I actually first

4:13 created this. I got my system to write me a 60 question interview. So you could do something like question me on the questions in this page one by one. You can see it asks you the questions.

4:26 What's something you believe about content that would make most people uncomfortable? So you build all of your answers. You can have that as a resource if it's needed in my guides and

4:36 primitives view here. But then you can ask your AI to then build a more atomized stripped down version. So this process of building these two documents will give you your identity and it will

4:49 give you your voice of you as a brand and as an individual. If you want to see what that looks like in a different system, let's say Claude when I'm working in co-work which works within

4:58 folders on your computer. You can see here that it's working in my specific folder. And if we go to that folder, you'll see in my copy of my co-work OS. Check the video below if you haven't

5:08 seen it. I have an about me folder. And what happens is when you speak to Claude, it will read your Claude MD file, the global instructions for your system, and tell it what to read. And in

5:20 my about me file, I have exactly that. I have an about me MD file, and then I have a voice profile. The whole thing is exactly the same. And the point of this is that I think with any good system, it

5:33 should remain platform agnostic. And in my business OS, you can then scale this to be used for a team. We have a button that creates a filtered hub of all the views of our workspace filtered by the

5:47 individual. And in that hub, they have their own personal voice profile template, interview template, who I am template. Every time they use their AI instructions, it will ensure that it

5:59 reads those files. But not only that, it will also ensure that it connects into the shared guides and primitives where there can be a brand voice and a business identity. We've scaffolded this

6:11 so that the system on boards people to use it. But just take the principles and you'll be finding the fastest way to feel a real difference that context can make when you talk to your AI. So if I

6:23 speak to Claude in notion, you'll see it load all of this context immediately. Let's plan a newsletter in my voice. It's loaded my personal voice. It's loaded a primitive guide to not sounding

6:34 like AI. It's adopted an agent mode, which is my content strategist. And it's loaded my brand identity. And there you go. It begins the process and works with me from an informed place. Now, the

6:46 principle applies to the wider knowledge your business owns. You would be ingesting key information and systems, processes into a single location. But let's talk about that in a moment

6:58 because next you need to ensure that your system helps both your team members and your AI reliably find and store information. Think of this as giving important answers a home. I think when

7:11 you're trying to consolidate and clean up your information, you don't just want to dump everything in. And to be honest, that feels overwhelming and too much to do. So think about it as that. Important

7:22 answers with a proper home. You might simply start this by writing down what are the key questions I have to answer every week for you. That might be pricing. It might be on boarding

7:32 refunds. You might think of this in a couple of ways. First of all, create yourself a little database in notion with a wiki which allows you to put key examples of work or SOPs. We actually

7:44 have an SOP for how to write a standard SOP and you can feed this to your agent just by mentioning it and get it to help. But you could start even more simply. One answer on one page. Give it

7:57 an owner and use that link whenever the question comes up. We've done it like this. We have a customer service knowledge management database. All we do is every time a question comes up in our

8:09 customer service email that we don't know the answer to, I work with my VA to give them the answer and we document it in this database. The reason this is useful, now that AI makes finding that

8:19 information quick and easy for a user, we can just query it. And most importantly, you can do things like validate, list when it was last shared, and know if it's out of date, and indeed

8:32 approve answers by clicking them. How this works for us is in our agent hub, we have a set of instructions that essentially curates our customer playbook. Your mission is to keep the

8:44 knowledge base well-maintained so that an AI agent could one day triage the entire email using nothing but this playbook. I won't go into all the details here, but essentially it has

8:54 these tasks. It does a weekly inbox sweep. It drafts a new SAP entry, fills gaps in the database, and audits any stale entries. As an example of this, it means that a user in my team can do

9:07 this. How should I respond to someone asking about whether agent OS, life OS, or business OS is right for them? It is using the customer support curator. It's looking through the entries in the

9:21 database. It's found those and we're getting an answer. It means that my AI is speaking from specific curated knowledge, not just searching the internet. It drafts me a drafted reply.

9:34 It follows through the information of what each thing is and it gives me what I need. This is the power of connected information. That's all well and good, but how does it know where to find

9:45 things and to not burn through tokens doing it? That's the next step. So, step two is architect. You need a map to find things. I think for many the cost of using AI truly meaningfully can

10:00 end up being really quite high in terms of token usage therefore cost and the amount of time an AI may take to find information it needs and deliver it where it is asked to deliver it. Now if

10:14 you've got a big business that can scale into cost really quickly. So what I am doing here in this templated system is really three things. One, it's a set of databases where all the information

10:27 lives, internal knowledge, tasks, projects, clients. Two, it's a set of dashboards where humans can view the information they need to view and know where to go to find it. And three, it's

10:40 a set of agent instructions for AI because we need a way for both AI, that's instructions, and humans, that's dashboards, to be able to understand, search, and edit the information quickly

10:53 and easily. And in this instance, it's a map. This is probably the most valuable page in my entire system. And it will look through the shape of the system. Here it explains a homepage, dashboards,

11:06 databases, guides. It has a list of where everything is. So it not only says this is the database but it explains you root here for these things and these things are related to these things. This

11:19 is not very readable for us. But it doesn't need to be. It needs to be a accessible map. But what it does mean is that if I go and ask AI a question, what's the top three most high impact

11:33 tasks in the task list for me to do right now? You'll see a couple of things happen. It loads that map, follows its rules and approvals for what it's allowed to do, and it's gone into

11:43 productivity coach mode because I have specialist agent modes that focus on this. It then queries the specific database it's meant to read. That will be the task database. And it's actually

11:55 following specialist instructions to do that very well, following my favorite frameworks for thinking around productivity. And here we go. It gives us the three key items that it needs to

12:06 look at. What's great though is it's also helping me refine this. It's saying that my current task list can't be trusted right now as I've not kept it up to date. And it reminds me to go and

12:15 clean that up. Fantastic. And if you wanted to do this in Claude Co-work, it would be exactly the same principle. At the top level in about me, there is a context map which does exactly the same

12:27 thing. It's just an MD file that lists all the locations of my information. Now, before we move to step three, if all this is seeming like something you want to get better at, take a moment to

12:38 go and check out agentic businessmethod.com because just yesterday we opened applications for my new accelerator program on going agentic with your business. It can help you take

12:48 scattered context, messy AI instructions, and an overwhelming amount of busy work and turn it into a consolidated system with AI support where you engage aic help in the right

12:59 place and prioritize being a business that is human first. That's what this is about. Simplifying and making work good for your team. Alongside the office hours, live training sessions, community

13:11 support, a full MCP guide library that helps you use uh the systems that we've got, we deliver two systems to you included in it and on board you to them. This exact Agentic Business OS that

13:24 we're looking at and the Claude Co-work equivalent. So go and check it out. You'll leave with a full collaborative AI ready system to adopt AI easily as a non-technical founder, entrepreneur,

13:35 operator in a business. we've got your back. So, the step three is to systematize your processes. Start by turning one repeated task into a skill or an agent.

13:47 Write the process down once as a skill, a clear set of instructions that a person or an agent can actually run. In my system, that might be something like one of these. We've got Andrea, our

13:58 customer service analyst. This is a notion custom agent. It runs on credits and we can set it up in settings and it has triggers that allow it to run a set of instructions which is this is your

14:12 job what to include in the report steps to take how to analyze emails and then connectors what it can read and what it accesses. This is access to the context that we've already built the products

14:24 and services knowledge base I showed you earlier and most importantly my email. This means that when we run Andrea automatically on a Monday, she will create a brief and deposit them into

14:38 here, the customer service weekly report. This means that every weekly meeting I have with my virtual assistant and customer service manager, they and I can both look at what's happened. We've

14:49 seen one product returned, 38 problems solved, and seven bits of positive feedback. It gives me an overview of what's happened. This is incredibly useful, not only for me to have a

15:00 positive relationship with my team members and celebrate their work, but also for us to more quickly get through a check-in. It makes such a difference. If you want to start really simply

15:11 though, I would start with something like this, the second part, which is shared skills. You can set a skill to be used by you. You can see here this page is being used as an AI skill. It's

15:24 connected to my skills library. I won't go into loads of detail, but it will break down a process. This one breaks down the process of adding a new entry to a knowledge base and atomizing the

15:35 information so that an AI can read it. It fills out the properties, talks about how it should structure an entry and has a gate on what quality looks like. It means that once you've documented it

15:46 once, it will repeatedly run and improve over time. And what we've tried to do in in our system is think about how you can scaffold skills. Think about what specific skills can feed other ones so

16:00 that you're not rewriting it every time. So adding an entry to a knowledge base is used by a load of my agents and can work for any knowledge base that I want to feed. This is the modern version of

16:11 an SOP, not a six-page document sitting in a folder marked important. It's a working set of instructions with context examples and checks that need to be hit in order to get a good result. Now, if

16:24 you were doing this in Claude Co-work, you can do this in the customized tab. Go to skills and here I have a skill which describes how to review my system and keep it clean and up to date. And

16:36 hey, if you're not very confident in creating skills, Anthropic make a skill creator and we built one for notion as well. If you wanted to take this further, you

16:48 can even document expertise. We have a set of specialist knowledge bases in my system. In this I've actually got one which is an entire knowledge base of atomized information ingested by the AI

17:04 using that knowledgebased skill to create a set of principles, patterns, concepts, protocols for creating great AI skills, agent instructions and prompts. This was all built around some

17:21 free resources that I found and turned into a database. And the joy of a specialist database like this is that when you then create a set of agent instructions such as a meta prompt

17:36 engineering expert that uses this database to answer questions, you can get much much better results. I can actually now go and say let's create an agent and build it for me and it will do

17:50 it. So let me give you an example. What's the best practice for building a sub agent instruction according to your knowledge base? The same thing will apply in claude. You have a a folder of

18:01 knowledge and a instruction set that points at it. It's loaded the correct mode using my personal agent instructions to know where to find it and that context map. The best practice

18:11 in short, build the sub agent in a six-part shape and make every always on rule name an artifact that appears in the output. Great. And it's kind of breaking down the process of doing it.

18:22 Let's do another example. Talk me through the best way to build a landing page for my accelerator based on your online business knowledge. So there you go. It's loaded the online business and

18:35 marketing strategist. Now it's looking through the online business and marketing knowledge base. Read all of these key pages that it's identified from the extensive library. Now this

18:45 looks like a huge amount of token usage, right? But think about the quality I'm getting here as opposed to a general answer I would get if I just asked AI generically. It's understanding the

18:57 concept of what we're building, suggesting a page architecture, and suggesting an order to work on the system. Incredible. Now, for our accelerator and templates, we actually

19:10 took this one step further. We've built a dedicated MCP guides server which we give to members of our community and they can ask questions about the template using the template agent

19:24 instructions. Explain to me the best way to start seeding tasks and projects into the Aentic Business OS. Now, I've actually mentioned the onboarding agent in the template cuz my personal agent

19:37 isn't connected to it right now. In any LLM model, a user can connect to the MCP server that we've created, query it, and get answers from our knowledge back to them. This is a great example of we're

19:52 keeping our knowledge gated and owned because that's where our value is, but we can deliver it to our customers in a private way to help them fasttrack to what they need to know. And the answers

20:02 I will get back come from this guide library. This is a kind of a view of everything in the system. So if we went to maintaining a healthy business OS, it lists everything that is listed within

20:14 there. And there are tons of entries about using the system, getting started, bringing your business context into business OS and so on and so forth. So check it out and check it out at the

20:24 Aentic Business Accelerator, too. So the final step is to activate. And this is where I want to show you how I've built a truly useful personal assistant. Now, I've shown you skills

20:36 which can be triggered from your AI simply or custom agents in notion that can run on a schedule. But this is probably my favorite use case for it. Working directly in Claude Co-work. I've

20:47 built myself this a personal morning briefing agent. Every day it runs. And if we click on today's and click on the report, we get this a fully coded report on what's happening. I have what to act

21:02 on, what needs my attention, schedule for the day, active action items that I need to work upon today, what's coming up, and we even have uh a Slack connector right here. So, it's giving me

21:15 a full overview. This is the power of building your context and then being able to activate. And it was really quite simple. I went and created a scheduled task. We go into my scheduled

21:26 task. We can click edit and you'll see this says who it is. Tells it where to find context. Gather information from your task list in notion and all the locations. Render it. I've given it a

21:38 design language using claw design and deliver it to me in these two ways. It's kind of as simple as that. Set the scheduled task to its frequency and permissions and save it and it runs it

21:48 and you get those results. Now, if you wanted to in scheduled now, you can also just create a new task with Claude's help and it'll help you do it and talk you through it. I also have a weekly

22:01 briefer that does two things. It gives me an overview of the week and what's happening, but it also offers me this an overview of trends on YouTube. If you want to learn how to do this, get

22:13 subscribed. I'll be doing a full video demo of how to build this in Claude very soon. So the question remains, can your business hold what it knows when you're not in the room? A year ago, mine could

22:24 not. Now the knowledge and the map and the skills mean that both my team and my AI can find the right context and do useful work without starting from scratch. That is Cassa. Consolidate,

22:35 architect, systematize, activate. Start with one part of your business and then run it on a loop. Do not try and do it all at once. So, if you want the templates or my help building it, the

22:45 business OS, the Aentic Business Accelerator are linked below. Applications close on Thursday the 17th of September for cohort 1, or you can join the mailing list to access the next

22:55 one. And Business OS will be available once we fill the cohort to everyone. And if you want to see the final step built properly, watch this video next where I'll take you through how Claude can

23:07 turn all of this into a really truly useful assistant. See you soon.