Cursor just made something incredible...

0:00 A new AI app is here and it feels very different from anything else that you've probably tried. It's called Grockbot and it's an agentic system for everyone and it feels like you're just having a

0:13 conversation with your agent. The first thing you're going to notice is it looks and feels just like any native chat application. Telegram, WhatsApp, all of them. We have threads on the left, we

0:25 have chat on the right, and it's very simplistic. You do not see things like the actual code. You don't see a directory structure with files in it. You don't see even that shimmering

0:39 thinking text. It's a actual big departure from what you usually see from cursor or codecs or claw code. And it's actually welcome. I'm excited to see a new design. And yes, it's not this

0:51 massive departure, but I still think it's quite different. And so, let me just show you what it looks like when it's thinking. So, you hit go. And there it is. My little guy is spinning. It

1:00 looks like he's thinking. You don't see any tool calls. You don't see any code being written. It is all very, very simple. And again, this is for a mass audience. I really think that's what

1:11 they're going for here. But it is still extremely powerful under the hood. You still get all of that power that you're used to with a codeex or a cursor. Okay, so here's the thing. This took me a

1:22 little while to get used to, but every single thread is an individual agent. They are not threaded by topic, which is kind of what I've been used to. So, if you look at cursor and you look at the

1:35 left sidebar, kind of looks the same, but when you dive more into it, you can actually see every single one of these threads is basically the same agent, but they are separated by topic. So each

1:47 thing I'm working on, whether it's a PR or a bug fix, it's all individual threads. And it doesn't feel like these are all individually created agents. And that is very different in Grockbot. For

1:58 each of these, they really do make you feel like they are agents. So let's create a new one. Click this plus button right here. And right away, you select a new thread. You can choose one of your

2:11 existing agents. So this is my computer cleanup, my chief of staff agent, my email agent, calendar agent or you can create a new bot. So let's create a new bot. And immediately upon creating a new

2:22 bot, I don't do anything except for start typing. It is thinking what I want to do. It's going to give me a few suggestions. So hey Matthew, good to meet you. And for people who are not on

2:33 the cutting edge of this stuff, it really tries to handhold you through figuring out what you want to work on. So what do you want me to be focused on? research and writing, inbox and email,

2:43 projects and code, day-to-day ops, and something else. So, I can simply type Amazon shopping and hit enter. Okay. And it's going to make a bunch of assumptions about what I mean by that.

2:54 And let's see what it does. So, what should I handle for you? Find and compare products. What are you shopping for? Drop a product, a use case, or even a rough vibe, budget, must haves, and

3:04 I'll dig in. So, I can say cameras. Cameras are a big category. Let's go for phone mirrorless DSLR and let's do 500 to a,000. So again, it's just asking me all these questions, but you're going to

3:17 see something cool eventually happen with this. And so my one critique so far is I actually like having it feel like it's just one agent that I'm talking to rather than a bunch of different agents

3:29 working on my behalf. And I'm going to show you how I solved for that in Grockbot, but that is the way I like to work. I want it threaded by topic, not by agent. And honestly, there's probably

3:40 not that big of a difference between those two. It's just more how they present the information to me. Okay, so it's saying digging through Amazon now for kit. It is still working. And I can

3:50 send another message. Show me live on Amazon. Now, that's where this gets really cool. This is what sets Grockbot apart. If you look up in this top right corner right here, there's this little

4:00 computer icon. And if I click it, you can see it actually spun up its own environment, its own operating system, and it was looking on Amazon.com for me. And so this is fully usable. Like I can

4:15 move this around. This is Linux, I believe, Thunar File Manager. And it is a full computer. And every single bot that you spin up has its own full computer. And the cool thing is even

4:27 though every single agent spins up its own fresh environment, they share authentication. So you only have to log in once. So if I log into Amazon now and then I spin up another agent and open up

4:39 Amazon, it will also already have been logged in. So I can continue. I can do work myself. I can use this just like I would use any computer. And it's really, really easy. And then there's this

4:50 button up here, teach a task. And I'm going to come back to that in a moment, but this is really cool. All right. So, here it actually sent me a screenshot. Here's the Canon EOSR50 kit at 8.49.

5:01 Great. But again, if I want to actually see what it's doing and interact with the live website, I simply click this right here. This is the operating system for this agent. Now, I talked about

5:12 really only wanting to interact with a single agent. And I like that because I figure it learns my personality. It learns what I like. I have this long context with it rather than spinning up

5:23 new context every time. So, the way that I solved it is actually by a suggestion that the cursor team had, which is have a chief of staff agent. This is the main agent that I'm working with day in and

5:35 day out. And one thing that's super useful is allowing your agents to publish things to the internet. So, let's say you want to share a document, you want to share a presentation, and

5:44 the easiest way to do that is with the sponsor of today's video, Here. Now, let me actually just show you how to do it. So, I'm going to spin up a new bot, install the here. now skill and then

5:53 this agent and all the other agents will now have access to publish to the internet very easily. Literally, you just say publish this and it'll give you a here.now URL with your published

6:05 document in seconds without having to log in and it is completely free out of the box. All right, installed here. Now is ready. I can publish it to a live site and then it's asking me if I want

6:16 to sign in. So, when you publish something without signing in on here now, you get it for 24 hours and then it disappears. If you want it permanently, all you have to do is sign in. All

6:26 right, so let me show you how easy publishing is now. So, I have this picture of this 3D printed boat that my son printed and I want to publish it to the web. I simply give it to Grockbot. I

6:35 say publish it and there it is. Within a few seconds, it probably took about 8 seconds, I have a public link that I can share with anybody and anybody can now access this photo. Thanks again to here.

6:45 Now, they've been such a great partner and it is so easy to publish anything to the web with your agents. Just tell them to use here. Now, link down below. And the cool thing about Grockbot is that it

6:57 can interact with it can message spin up other agents. So, I can say, "What's the latest email I have in my inbox?" Now, I hope what's going to happen is it will message the email agent to ask for that

7:09 because the email agent is where I've done all my email triage work so far. So, let's see what happens. Okay, so it didn't actually message the email agent yet. And I asked it, did it? And it said

7:20 no. For that one, I just pulled Gmail myself. Email agent is for the bigger inbox sweeps. A single what's the latest I just handle here. So, let me try another one. I'm going to say, how many

7:29 emails are easily safely archable? And this one hopefully will message the email agent. There we go. Having email agent do a dry run count of what's safely archable. No archives yet. So,

7:40 messaged email agent. We can see it right there. And if I click into it, this is actually the conversation between those two agents. So it's really interesting. The conversations between

7:49 agents persist indefinitely also. And that's also building up context for you to use. It's super interesting how they do that. So I can see right here, here's the email agent counting safely archable

8:02 mail on both accounts. Dry run only. Nothing gets archived. If I switch back to the chief of staff, I'll bring the number back as soon as the email agent finishes the scan. Okay. So you can see

8:11 here now the email agent has messaged the chief of staff. I click back to the chief of staff. Only three of 21 inbox threads clear the strict safe bar. And then of course I can take action if I

8:22 want to archive them or not now. So that's a really cool and unique feature how the conversations between agents are persisted and it really does feel like just another part of the conversation.

8:32 I've not seen that in other apps. And so Grockbot very much feels like a cloud first agent, but it can very much control your local computer as well, which is really nice. So I can say

8:44 something like, "Tell me how many folders are on my desktop." Nine folders on your desktop, not counting hidden ones like cursor. So I can control my local computer just as easily as I can

8:54 control this cloud environment right here. And it's just really nice. It kind of is this hybrid local plus cloud. And it can do both and go in between them really easily. And I think that's kind

9:05 of a quite unique feature. The cloud environment feels very native, feels almost default inside of Grockbot, whereas with codecs, it's much more work. Everything defaults to being

9:20 local. And I bet they're heading towards cloud agents. But generally, when you're thinking about cloud code and codeex, it is very much local first. And the next thing I want to talk about is how much

9:30 this feels like it is built for non-technical people. Now, it is still a technicalish product, right? We're we're talking about artificial intelligence. It is not dead simple yet, but this is

9:42 the direction I think a lot of AI apps are going, making it generally easy to understand. And the way that they did that was by stripping out a lot of what we are used to in these chat

9:54 applications or in these coding applications. So, for example, there's no model picker. You cannot select between the Grock model, the GPT 5.6 model, Fable. You can't select it. You

10:06 don't even see it. You don't even know which model it's using. And you know what? I think most people don't care. It might be using Grock only through and through, right? That might be the only

10:17 model it's using. And if so, it's really good. It's extremely capable and it's very fast. So, I've been really impressed with how capable and fast the Grockbot is. It is still thread- based

10:29 like I mentioned, but each thread is its own agent. So that's kind of the still the look and feel that feels familiar from codecs, from cloud code, from cursor, but it's also what most people

10:40 are used to. If they're using even just basic text messaging, this is the interface people have gotten used to and are familiar with. Here's another thing you're going to notice. There is

10:50 absolutely no tool calling shown. It is not showing anything having to do with code and that is intentional. That is what makes me believe this is for a broad audience. So if you look at cursor

11:02 you can see right here explored two files three searches. You can see gpped right here. You can see the actual files being edited and read from. You can see all the tool calls all the things that

11:13 cursor is doing on the way to getting its final answer. And of course there's nothing like this. showing the PR, showing the changes, uh the terminal, the files, none of it is there. This is

11:25 really meant for the broad knowledge work community. Now, another piece of feedback that I have is that it does feel a little bit wonky having a completely separate application for this

11:35 because I have all this work and context built up in cursor and now I have to have cursor for code and I have to have Grockbot for general knowledge work. And I think the line between them, the clear

11:48 definition of what goes where is not going to be so clean. It's going to be quite blurry. Actually, deciding should I do something with Grockbot or should I do something with cursor? And it's going

12:00 to be interesting because you see what OpenAI did with Codeex, merged Codeex and Chai GPT into the same app, one super app. And I actually find having to switch between chat GPT and codeex. I

12:13 almost exclusively use codeex versus chat GPT. I find that entire decision-making process just unnecessary. And so maybe splitting the apps is the right way. But we'll see.

12:25 What I really want is I just put a task in and it knows what to show me. Do I want to see code? Do I want to see tool calls? Or do I want just the most simple version? just show me point A to point

12:39 B. And so we'll see which design pattern sticks. So to make Grockbot incredibly powerful, you want to use plugins. So plugins are right here. It is one of the only features prominently shown. You

12:51 click it and you see all the different apps that you can plug in. You can kind of think of this as like MCP servers or just giving your Grockbot away the knowledge for how to interact with all

13:04 of these different services. So, I've added Gmail. I can add Google Drive just like this. You just click add and it's done. It may ask you to authenticate once, but it'll do that once you

13:16 actually start using it. I have Google Calendar added. I have Slack. So, I can invoke uh Grockbot straight from Slack, which is quite nice. I have notion. And they have so so many already default

13:29 working with this. And you can see I've already added Box so I can read from all of my documents, my contracts. Shout out to my friends at Box. So, one of the most popular use cases I think most

13:41 people will get from Grockbot is routines. And these are tasks that happen on a recurring basis, a schedule. All right. So, you're going to rightclick on your agent. You can click,

13:51 it's kind of weird. It's kind of roundabout. You click edit profile, which you can give it a name, title, and description. But to get to routines, you click this little back button. And then

14:00 you can click this create routine. So here's an existing routine that I've already created. This is called computer cleanup. It runs every Monday at 8:00 a.m. And this is for the computer

14:09 cleanup agent. So if I click into it, I said look around my computer for files you think are stale and I can clean out. Don't actually clean it out. Just suggest. So basically what I wanted to

14:20 do is optimize my hard drive. Look for anything that I don't need anymore. Suggest the things that I don't need. And then I can just tell it, okay, go ahead. Throw it in the trash for me. And

14:29 so that'll run once a week. Super useful. But you can do anything on a recurring basis. You can say, "Okay, every single morning I want a summary of all the important emails that you see

14:40 that I need to answer right away." And so all of that you can create with simply a name, a natural language instruction that will go to the agent. And then of course you can set it to

14:51 active or inactive. You can delete it and you can test run it as well. All right. Next are tasks. These are really cool ways to teach your bot to do something. I think Codeex has something

15:04 similar, but basically if we click into this window in the top right, which is this agent's own environment, we just click open. And now this is its own computer that it can use. So I can click

15:16 around. There is this teach a task button right here. All right. So let's say I want to teach it a task. Every single week I want it to look up this camera and give me an update on the

15:25 price. So, all I have to do is click teach task. Okay. So, again, I'm in its own operating system in its own environment. I'm going to grab this URL. I'm going to place it right here in this

15:36 spreadsheet I made. Then I'm going to go back. I'm going to grab the price. I'm going to paste it right there. $599. Okay. And so then I click stop. And it's now going to learn from the

15:48 demonstration that I just gave it. And so let's see what happens. So it says new bot working. It's thinking. And what it's going to do is actually create a skill, a reusable skill based on what I

15:58 just showed it manually, which is really cool. So there we go. Saved. Log Amazon product to camera prices sheet. That is the skill. Open an Amazon product page. Copy the URL. Switch to the camera

16:12 prices Google sheet. Paste the URL into column A. And so it's watching the video and it's saying, "Okay, you put $5.99 as the price. Let me update that skill." All right. And there we go. So it

16:23 finished. Now it also copies the price. So that's it. You can give it anything. You can really allow it to learn from any knowledge work, any task that you're doing. And hopefully you can automate it

16:35 pretty easily to reduce any of those tedious tasks that you're working on. So we created a skill from the thing it learned, but you can also add your own skills. And one of my favorite skills is

16:46 the humanizer skill. Basically makes AI writing look less like AI slop. and I said install this skill. Hit enter and let's see what happens. So look, it first checked the plug-in marketplace

16:59 and it couldn't find it. So I'm installing it as a private skill from the repo. So just like that, very easy. And there's a lot more you can do with Grockbot and just continue to make it

17:10 more powerful. I actually covered a bunch of skills that you can apply to Grockbot right now in this