ArticleSeptember 7, 2026Free to read

Grok Bot Has a Surprise!

Centered on Bot, not Chat.

Originally published . English translation: . Read the Chinese original.

Original video in Chinese.

Key Takeaway

  • Grok Bot is currently the product that comes closest to an “AI employee”: it centers on Bots rather than a chat box, and the context for the same kind of task accumulates over time on the same Bot; at the same time, each Bot is equipped with a long-running cloud-computer work environment, does the work on its own, and the user only takes over when necessary.
  • Grok Bot and Codex complement each other: Grok Bot is an “AI employee” (it has its own office, stays online long-term, and works autonomously); Codex is an “AI assistant” (it listens to instructions and helps in your office). Right now I use two main Bots, one for topics and one for GEO, and then later I can add Bots for development and investing.
  • This is the right direction for building a personal sovereign AI: there’s no need to create a large number of Bots; just consolidate similar tasks into one role. Bots can also be grouped together to collaborate. The product thinking is ahead of the model capabilities themselves, and it’s worth trying for both bosses and entrepreneurs.

Every boss should try Grok Bot. Because among the products I’ve used, at least, it’s the first one that truly feels like an AI employee.

Every entrepreneur should also try Grok Bot. In the Agent wave, they’ve really gotten the product right.

In this video, I’ll talk about why I strongly recommend it, and how I’m using it right now. In fact, I’ve already posted several notes about it in the Knowledge Planet over the past few days.

In my view, Grok Bot got two things right:

First, it centers on Bot, not Chat.

The chat box is suitable for Q&A; that’s a product of the chatbot stage. But once you reach the Agent stage, if you still use a chat box and have to start a new conversation every time, what do you do with all that scattered context?

So Grok Bot centers on the Bot—you create a Bot; whenever you have a need, you come back to the same Bot.

All the information for the same kind of task is gathered in the same Bot. Different Bots handle different tasks separately. It’s like everyone in a company doing their own job. That’s what an AI employee should look like.

Second, it equips the Bot with a long-running work environment.

Grok Bot doesn’t fight you for your computer; it works on a cloud computer. The setup is actually pretty powerful, and I praised it in a post before.

Each Bot has its own work interface. Bots under the same account share a continuously running cloud computer. They execute tasks on their own; you don’t need to keep watch.

By default, the user only sees status and previews, and only takes over full screen when necessary. For example, logging into an account, and then handing it back once that’s done.

If you share a computer with it, the advantage is that the materials are all local and the AI can get started right away. But the downside is that you have to set permissions to stop it from doing random things; in addition, you have to prevent the computer from sleeping and interrupting tasks.

You can see that Grok’s model capabilities may not be the very best, but Grok Bot’s product thinking is very advanced. In some respects, it even feels more like a complete AI employee than Codex.

So I’m using both products now. As I said in Knowledge Planet newtype:

Grok Bot is an AI employee: it has its own office, stays online long-term, and tinkers away on its own.

Codex is an AI assistant: in my office, it listens to my instructions and helps me work.

I didn’t create a huge number of Bots. In fact, my core needs right now are only two things: one is topics, and the other is GEO.

The first Bot I created was for topics. All of my past materials are on Dropbox, so I gave it permission to look through everything. In addition, I gave it my personal website, X, and YouTube accounts so it could know what happens after content gets published.

With that foundation, I had it give me a comprehensive diagnosis and propose optimization suggestions. To be honest, the judgment it gave was very close to my own.

Besides video content, I also had it help me monitor top creators on X to see which topics are worth following up on, and to provide specific post content.

The second Bot is dedicated to GEO. I had it open the back ends of Google and Bing, and I logged in myself. That way it can see the site’s specific data. After setting up scheduled tasks, it can also regularly check the data and give optimization suggestions.

Before, I used to manually download data reports and then hand them to Codex for analysis and processing—I actually don’t really like Codex moving my browser around. Now that Grok Bot has a long-running work environment and can check the back end by itself, it saves me a lot of trouble.

From my actual use, I’ve found that many tasks can in fact be consolidated into one role, one Bot. There’s no need to make too many; in fact, that just consumes resources. Right now these two are the main ones, and if I add two more Bots for development and investing, that’s enough.

If you don’t want to create Bots yourself, you can also check the official Marketplace, where there are already quite a few ready-made Bots.

Also, in addition to using these Bots separately, you can create a group, add them all in, and let them message each other and share context. For example, some can handle gathering materials, while others write reports, and so on.

So, as I said earlier, I’m using both Grok Bot and Codex at the same time. One works for me, and one helps me.

I plan to run this setup for a while. I feel like this will be the right direction for building a personal sovereign AI.

OK, that’s it for this episode. If you want to understand AI, want to reclaim individual sovereignty, and want to find like-minded people, come join our newtype community. See you next time!