25:05Alex Finn · Tutorial
Grok Bot: From Beginner to Expert in 25 Minutes
A hands-on tour of Grok Bot's zero-config, cloud-first multi-agent setup, and the six-bot roster running one founder's day-to-day business.
August 17thThe argument in one line.
Grok 4.7 is the first model trained specifically to run inside an AI agent harness rather than a chat window, and pairing it with a three-role bot team turns everyday coding and project work into something that runs itself for days at a time.
Read if. Skip if.
READ IF YOU ARE…
- You're already running an AI coding agent (Cursor, GrokBot, Claude, ChatGPT) and want a repeatable structure for handing it real project work instead of one-off prompts.
- You manage multiple projects or areas of a business and want tasks triaged and executed without babysitting every step.
- You're deciding between AI agent harnesses (GrokBot, Muse, Hermes, ChatGPT) and want a concrete comparison of what each is actually good at.
SKIP IF…
- You want a deep technical explainer of Grok 4.7's architecture or benchmarks — this is a workflow video, not a model-card breakdown.
- You don't use a project-management tool like Linear or Notion and have no interest in adopting one — the workflow depends on it.
The full version, fast.
Grok 4.7 is the first model trained specifically for GrokBot's agent harness rather than for chat, and it now beats Opus 5 at half the price while costing the same as Grok 4.6. The core workflow: set up a project-manager, developer, and designer bot for each major area of your work, feed the project manager an idea, have it break the work into tasks inside Linear or Notion, then tell the bots to pull tasks off the board, execute with Cursor's cloud agents (to save on GrokBot usage), review each other's work, and open pull requests. Run correctly, the bots keep working and talking to each other for days with zero further input.
Free for members
Chat with this breakdown — free.
Sign in and you get 23 free chat messages on us — ask for the hook, quote a framework, find the exact transcript moment, generate a markdown action plan. Bring your own key when you want unlimited.
Create a free account →Where the time goes.
00:00 – 00:29
01 · Intro
Cold open framing Grok 4.7 as the first model trained natively for an AI agent harness, previewing the setup and workflow to come.
00:29 – 02:49
02 · What is different
Grok 4.7 released same-price as 4.6, trained specifically for GrokBot, framed as the opening move in a broader race among AI companies to own the agent harness rather than the chatbot.
02:49 – 03:26
03 · Vs other models
Price-to-performance comparison: Grok 4.7 sits above Opus 5 at a fraction of the cost; positioned as the daily-driver model versus higher-end options like Fable/Astra-tier models.
03:26 – 07:48
04 · GrokBot set up
Walkthrough of the three-bot structure (project manager, developer, designer) set up per major life/work area, and why that split beats improvising threads in a plain chat tool.
07:48 – 14:05
05 · GrokBot workflow
Live demo of the workflow: switch Cursor cloud agents to Grok 4.7, hand the project manager bot an idea, have it write issues into Linear, then let the bots pull tasks off the board and work autonomously for days, wrapping with a note on usage/cost tradeoffs.
Lines worth screenshotting.
- Grok 4.7 was trained natively for the GrokBot agent harness rather than for chat, making it built to act autonomously rather than just answer questions.
- Grok 4.7 costs the same as Grok 4.6 while benchmarking above Opus 5 at half of Opus 5's price.
- AI companies increasingly care less about people using their chatbot and more about people using their agent harness, since the harness sees emails, purchases, and every task a user hands it.
- A three-bot structure of project manager, developer, and designer per life area lets a harness route tasks to the right specialist without manual triage.
- Routing coding work through Cursor's cloud agents rather than the chat platform's own execution saves on the platform's usage limits, since the IDE's own credits absorb the compute.
- A detailed backlog of tasks in a tool like Linear lets a bot team work for days without new input, because the bots pull the next task themselves instead of waiting on a human.
- Bots assigned distinct roles will ask each other for permission and second opinions on their own, producing a review loop with no human in it.
- A chat-only model is better suited to hands-on work you want to verify step by step, while an agent-harness model is better suited to repeatable tasks you're comfortable delegating fully.
How to turn an AI agent harness into an autonomous team
AGENT WORKFLOW
Splitting an AI agent harness into project-manager, developer, and designer roles per area of work lets the bots plan, delegate, and ship for days without direct supervision.
- Assign a project manager, developer, and designer bot to each major area of your work so tasks route to the right specialist automatically instead of you triaging everything yourself.
- Feed a project idea to the project manager bot and have it break the work into individual tickets inside a project-management tool like Linear or Notion before any coding starts.
- Route coding work through a cloud IDE's own agents rather than the chat platform's execution, since that shifts the compute cost onto the IDE's usage allowance instead of the harness's.
- Let bots pull tasks directly off the shared board and message each other for review before opening a pull request, rather than waiting for you to assign each step.
- A detailed, well-written backlog is what lets a multi-agent setup keep working unattended for days at a time — the depth of the task list, not just the model, drives the autonomy.
- Check a model's price-to-performance chart before assuming the newer or more expensive option is the better pick for a given task.
- Reserve a chat-only, step-by-step model for hands-on work you want to verify closely, and save the autonomous agent harness for repeatable, lower-oversight tasks.
- A model trained specifically to operate inside an agent harness can outperform a general chat model at the same task and price point.
Terms worth knowing.
- Agent harness
- The software environment an AI model runs inside when it acts autonomously across tools and tasks, as opposed to a plain chat interface where it only responds to prompts.
- GrokBot
- Grok's AI agent platform, where multiple bots can be set up with distinct roles, memories, and tool access and left to work on tasks with minimal supervision.
- Cursor cloud agents
- Cloud-hosted coding agents inside the Cursor code editor that can write and submit code changes on their own compute, separate from a chat platform's own usage allowance.
- Linear
- A paid project-management tool used to track work as discrete tickets called issues, which an AI bot can read from and update as it completes tasks.
- PR (pull request)
- A proposed code change submitted for review before it's merged into the main codebase.
Things they pointed at.
05:20toolCursor cloud agents
09:58toolLinear
09:58toolNotion
Lines you could clip.
01:40
“I don't know really what the point of using Opus 5 is, but I am sure we will hear more from Anthropic soon.”
03:30
“I believe everything we see moving forward from these companies are gonna be all about the harness.”
04:07
“I think Grokbot is that fastball down the middle pitch you rely on that just gets it done better than all the others.”
10:10
“Your Grok bots can now just work off this board. They can go autonomous.”
Word for word.
Read-along
Don't just watch it. Burn it in.
See every word as it's spoken — crank it to 2× and still catch all of it. The same dual-channel trick behind Amazon's Kindle + Audible.
metaphor
Grok 4 .7 just released and it's not only an excellent model, but is the first model trained to be used in an AI agent harness. That's right, this model is literally built for GrokBot. In this video, I'll cover why this model is so good, show you how to use it, and most importantly, show you my GrokBot setup and workflow that will skyrocket your activity and makes the absolute most out of Grok 4 .7.
This one's a banger, so let's lock in and get into it. it. So Elon did it.
He dropped the long awaited Grok 4 .7 today. It's the one we've been waiting for. Let's talk about what is different with this model.
The number one news story here for me, and after this, we'll go straight into Grokbot. I'll show you my setup. I'll show you which bots you need.
I'll show you the workflows for using this model, all of that. So stick around. But the number one story here is that this model was trained to be used natively in Grokbot.
That means it was built to be your personal assistant. It was built to do things, is built to be proactive, is built to execute on tasks.
It was built to be just as good, if not better than human beings at doing work on your computer. And this is huge news because I personally believe Grokbot was the best AI agent harness. And that was on Grok 4 .6.
That was before this upgrade. Now, there are a lot of other agents out there. There's now Muse.
There's a few others. I think Grokbot has been my daily driver for weeks now since it released. So this upgrade is awesome and I'll show you a lot of the awesome things it can do.
Some other amazing things, same price as Grok 4 .6. Grok has been the most cost -efficient model out there for months and months and months. That only continues.
They did not raise the price. They improved performance, same price. From a benchmark perspective, I'll show you those in a second, it's better than Opus 5 and it's half the price of Opus 5.
is pretty mind -blowing. At this point, I don't know really what the point of using Opus 5 is, but I am sure we will hear more from Anthropic soon. This is also SpaceX AI's first entry into what I am calling the great agent race.
What do I mean by that? I believe... all of these AI companies now care less about you using their models and more about you using their AI agent harnesses.
There is so much more to be won by SpaceX having you use Grokbot than it is just by using their chatbot because now they can see everything you're purchasing, emails, all of that. There's so much more data to be had from these AI agent harnesses. So I believe this massive race for the AI agent harness has begun and this is their first entry into it.
And I believe everything we see. moving forward from these companies are gonna be all about the harness. Grock bot muse all that rather than just the model.
So this is big from a performance perspective I like this chart because you can see price to performance here as you can see opus 5 grock 4 7 It is above opus 5 and basically medium and above and it is significantly cheaper So there's pretty much no reason to use opus at this point fable 5 1 They never claimed this was gonna be a fable or astra level model Although elon's saying the next one might be that fable or astra level model, but this is going to be your daily daily driver, your main workhorse.
When you need tougher tasks, you go with Fable and Astra. When you're looking to just kind of do your basic everyday tasks, Grok47 at this moment in time, we got a very exciting two weeks ahead of us, is the winner for those kind of daily tasks. Now let's get into the meat and potatoes here.
This is what you've been waiting for. This is the actual helpful part and that is GrokBot using this brand new model in GrokBot. I get a lot of questions around AI agents.
Should I be using... Hermes, Grokbot, Muse, all of these. We talk about a lot of different AI agents on this channel.
I personally, over the last few weeks, Grokbot has been my daily driver. I think Muse is stronger in a lot of different ways. I think chat GPT work is stronger in a lot of different ways, but for getting just basic tasks done, basic knowledge work tasks done, firing off code really quick, understanding your emails, doing a little bit of research.
I think Grokbot is that fastball. down the middle pitch you rely on that just gets it done better than all the others. That's why I've been relying on Grokbot.
So what I'm going to show you in a second is my workflow with the brand new Grok 4 .7 inside Grokbot. What I'm doing to get way more work done, get way more code out there, get much, much more done. But just from a high level perspective, Muse is great.
I just think Grokbot is built more for the power user. I'm a power user. Odds are, if you're watching this channel, you're a power.
user. I think there's just better for power users because you can have all of these different agent personas, right? With their own personalities, their own tools, their own memories for the power user, the non make me a reservation user.
I think this is the best setup you can have, right? I think news is great, but I think it's built for the, like the make me a reservation crowd, the crowd that like wants to save like eight seconds and have their AI agent go and make a reservation. I'm not one of them.
I make my own reservations, but when I want to get hard, work done. When I want to pump out tons of PRs and tons of code, I go to Grokbot.
I'm not going to Muse for that. So that's why I still recommend using Grokbot. Now let's talk about using Grok47 in Grokbot.
First, I'll show you my setup. Then I'll show you my workflow using Grok47, using the cloud cursor agents, which I think are excellent. So my setup is this.
As you can see it over here on the left, basically for every important area of my life, whether it's for Henry and intelligent machines, the startup I'm building, whether it's for my content team, whether it's for product research, whether it's for my community, for all the different important areas of my life. Another project I'm working on right now, I basically have three main bots for every area.
The project manager, the developer, and the designer. I think for every important area of your life, you should make these three bots inside a Grok bot. The reason I do that, I think this is a perfect way to separate your team.
separate their duties I find that they are really good at determining which tasks to give to which bot really well when you have these three roles so what you want to do if your setup is not perfect just yet go in think of the four to five important areas of your life then Build those sections on your left here and then put the three bots in each.
The project manager, the developer, the designer. Then your chief of staff can easily message each one. The project manager is going to be your executive on each of those teams.
So you can create a group chat called the exec team, which has all your project managers in that one chat. So if you ever need status updates or things like that, you can just go to your exec team chat and get the status updates from the project managers. This is not a setup you can do in Muse.
This is a setup you can... technically do in like chad gpt and claude but you would need to like spin up threads and then pin the threads and then rename them with names so it's kind of like you're here grokbot just made for this use case and that's why it's still my daily driver by the way if you are a grok fan you need to immediately subscribe and turn on notifications because there is way more exciting grok content coming i filmed a video over the weekend which is my most anticipated video I'm ever putting out that's coming out soon.
So make sure to turn on notifications for that. I'm really excited about this one. It's going to be a lot of fun.
Subscribe, turn on notifications, and leave a like if you learned anything at all. Also, one last plug, number one AI newsletter on planet Earth, completely free. Link for that is down below.
Click to sign up for that. You won't regret it. So let's go into the workflow.
I personally am building my app, Henry Intelligent Machines, almost completely with Grokbot at the moment because I have a really good workflow going. What is this workflow? Well, here's what makes Grokbot so good.
It is really good at being autonomous and proactive, right? Chad, GBT six Astra best model on planet earth. Now, without a doubt, the best at building things, all that, but it's not.
autonomous. It doesn't just do work, right? You have to tell it exactly what to do.
It does what you tell it to do really well, but it doesn't autonomously pick up tasks and do things while you're not looking. Grokbot does that really, really well. When you connect your Grokbot to the cursor cloud agents, to the team I showed you earlier, developer, project manager, designer, and then to some sort of knowledge manager, whether it's Notion or Linear that has tasks in it, you can basically just go to your team say hey start building and they'll go and just build for hours and hours if not days and days like they've doing for me here so here's the grok47 inside grokbot workflow that's absolutely blowing my mind first of all make sure you're on grok47 in your cursor cloud agents all you have to do that say hey change our cursor cloud agents to grok for seven.
You hit enter on that. Whoever you're talking to, whatever grok bot you're talking to is going to go change it to grok for seven. It's not that by default for some reason.
So make sure you do that. Once you do that, we can start setting up this workflow. It's been amazing for me.
That's helped me pump out PRS. So you go to your agent and you give it the idea of what you want to build.
I want to build my own agent harness, right? Whatever you're building at the moment. For me, I'm building an AI agent.
So I give it to my agent here and I say, this is what I want to build. You give it details. Then you say, create issues inside linear slash notion.
for this project. So here's what we're doing. You're giving details about whatever you want to build to your Grok bot.
Your Grok bot is taking all those details and breaking your project down into tasks. These are called issues in linear. I don't know what they're called notion.
They're just tasks and notion. And they're breaking it down into little building blocks for your project. And then they're going to go and take that and put that in that project management tool.
You can use linear notion for this notions like completely free linear. I think I'm paying like $20 a month. but as like a really nice UI.
So it's completely up to you. But as you can see here, I have a ton of different tasks and issues in here that my Grok bot made. So once you have that, your Grok bots can now just work off this board.
They can go autonomous. So now you can say, and go to your Grok and say, start taking tasks off the board. and working on them.
Review all your work and verify before you create a PR. Then you hit enter on that. What's going to happen is your GrokBot, which it's been doing it for me for literal days now.
This is happening over days here. It's just, as you can see, there's zero messages from me here. It is just working autonomously, literally for days, taking issues and building them out.
And what's amazing here, it's what makes GrokBot so good. they just autonomously talk to each other, the different bots. I haven't sent messages to this bot for weeks.
They're just talking to each other, the project manager, the researcher and developer, taking different tasks, asking each other for permission, asking each other for second opinions and building out my project. And because I have hundreds of tasks on my linear that I had them write out and develop in detail, they can just go and go and go and go for days on end.
they do it completely on time. So now I know what the first pushback you'll give me is, Alex, what about usage? What about usage?
Yes, Grok bot usage isn't the best in the world. It's not the best in the world out of all the AI agent harnesses. That's probably Muse.
Muse just gives you a crap ton of work for free. But here's what you do. As long as you make sure they're using cursor cloud agents to code, then you save on usage because it's not Grok doing the work.
It's your cursor credits that are doing the work. And if you're paying for a cursor plan or a Grog plan, you get cursor credits with it, right?
So you're saving usage because the cursor cloud agents are doing it. I've been doing this workflow and I'm getting so much work done autonomously. It's like actually ridiculous.
This is the only AI agent harness I use that's so good at autonomous work. Like if I'm doing hands -on work, like hands -on, I want to make sure it's done well. Yeah, I'm using ChadGBT because I do it step -by -step.
It's not autonomous, but it's super smart. But if I have kind of basic, again, , build this out, change this, put this, then I have Grokbot doing this. I can just do it autonomously.
You can just chug through these tasks so good. It is an amazing setup. It's an amazing workflow.
If you aren't using Grokbot like this yet, you need to be because now with Grok 4 .7, it's doing it so much better. It's so much more accurate. The code is so much better.
I am really enjoying Grok. rock bot. I just think it's an excellent user experience.
And now the brain's just got a nice free upgrade, right? It's the same price as rock four, six. You just wake up today and your rock bots are smarter than yesterday.
That's pretty good. What are you doing? Rock Bob, what's your setup?
Let me know down below. I'm very curious. And what AI agents do you want to hear more?
If you want more muse tutorials, do you want more rock bot tutorials? You want more Hermes tutorials? Let me know down below in the comments.
I make all my videos based on your feedback. So I want to know what you hear. I hope this was helpful.
We live in the most exciting time ever. These next two weeks are going to be crazy, by the way. These next two weeks are going to be insane.
We have so much coming out. Dev Day is next week for OpenAI. I guarantee you we get a crap ton of releases.
I'm hearing birdies about different things. I think a lot of other companies will be shipping things next week and this week are going to be probably the two most exciting weeks in the history of AI. You need to lock in.
You need to turn on notifications. I will have videos about all of them as they come out.
This is the number one destination in the world for AI news and most importantly, practical AI advice for getting the most out of these magical tools. Hope this is helpful. So grateful you'd watch my content.
I'll see you in the next video.
The Hook
The bait, then the rug-pull.
Grok's newest model wasn't trained to chat, it was trained to run GrokBot itself, and that reframes the whole release: same price as the last version, benchmarked above Opus 5 at half the cost, and built to be proactive rather than just responsive. The rest of the video is the payoff — the exact three-bot setup one creator uses to turn that model into an autonomous team that plans, codes, and reviews its own pull requests for days at a stretch.
Named ideas worth stealing.
05:35model
Three-bot team per life area
- Project Manager
- Developer
- Designer
For every important area of work or life, set up a project-manager, developer, and designer bot inside GrokBot. The project manager triages and hands off to the other two, and a cross-project 'exec team' chat of all the project managers gives a single place to check status.
Steal forany multi-agent AI workflow where tasks need to be triaged and delegated without constant manual routing
How they asked for the click.
VERBAL ASK
07:27subscribe
“Subscribe, turn on notifications, and leave a like if you learned anything at all. Also, one last plug, number one AI newsletter on planet Earth, completely free.”
Delivered as a quick mid-roll ask right after the setup explainer and before the hands-on workflow demo, paired with a plug for his free newsletter; repeated again in the outro.
MENTIONED ON CAMERA
FROM THE DESCRIPTION
PRIMARY CTAWhere the creator wants you to go next.
OTHER LINKSAlso linked in the description.
Visual structure at a glance.
