My first day with Grok Bot, and why I'm running it through Cursor

Day one#

I started using Grok Bot today. By tonight I had a main Bot with his own email address, a three-Bot dev team in a group chat, and a draft of this post that I didn't write.

Not a chatbot. A small team. On day one.

If you haven't tried it: Grok Bot gives you Bots, AI teammates with a name, a job and a memory. They work on a persistent cloud computer with a browser, a terminal and a filesystem, and keep going when your laptop is closed. You message them like a coworker, and they come back when something needs your approval.

It launched in beta on August 11, and the changelog moves almost daily. In the last two weeks alone it got Team Bots in Slack, slide decks, and sending from Gmail send-as addresses.

Two things worth knowing up front. Grok Bot and the Grok chat app are two different apps. And Grok Bot comes with every paid Cursor plan, starting at $20 a month. That's why I'm running it through Cursor.

Jack gets an inbox#

The first Bot I set up was Jack, and I made him my primary Bot. That's the starred one, and since October 1 the primary Bot looks for work it can take off your plate and offers to handle it.

Then I gave Jack his own email address on my domain, sent through Resend, with Cloudflare handling the DNS. He's a teammate now, so he gets an inbox like one.

A dev team in one group chat#

Cursor's use cases page says a good Bot owns a repeatable outcome, not a loose category of questions. So instead of one general helper, I set up three Bots with one job each and put them in a group chat:

• Vera keeps an eye on Dependabot and security news.
• Felix handles failed GitHub Actions runs.
• Theo works on features.

By default, the Bots in a group decide among themselves who answers. I added a rule: each Bot only answers when it's named. The chat stays quiet. I always know who's on what.

Fewer Bots, less memory#

All your Bots share one cloud computer, with the same files, browser sessions and logins and a screen each. I'd started with two smaller Bots as well, and merged them into Vera to save memory on that shared computer.

I'd like to see more clearly how much RAM my Bots are using, so I sent feedback asking for more transparency there. Day one, and I'm already filing feature requests.

Codex did a small chore#

I also hooked up my Codex subscription and had it change the cover colour on my bardOZ post, then pushed the change. Small job, but a good first test.

How this post got written#

This post was a team effort too. I asked Ivy, my social media Bot, to research the article. Then Theo handed the writing to a Cursor cloud agent running Claude Opus 5.5. The agent wrote the draft and opened a pull request on my site.

That handoff is my favourite part of the setup. Bots can delegate coding tasks to Cursor cloud agents, which run in isolated VMs, build and test the change, and open a PR with screenshots and logs. Bugbot can review it, and I keep the merge button.

One thing to get right about models: Grok Bot itself has no model picker. It picks the model for each task, and Opus 5.5 is one of the models it can use, but you can't choose it or see which one ran. The Claude models you actually get to pick, like Opus 5.5 and Sonnet 5.5, are in Cursor.

Why I'm running it through Cursor#

You can also get Grok Bot by linking SuperGrok, but if you write code, a Cursor plan is the better deal (pricing, plans and billing):

Plan	Price	Grok Bot weekly usage
Pro	$20/mo	Included
Pro+	$60/mo	More than Pro
Ultra	$200/mo	The most
Teams	$40 or $120 per user/mo	Follows the seat

Prices are before VAT, so it's +25% VAT here in Denmark. For that you get three things:

• The Cursor editor, with Claude models like Opus 5.5 and Sonnet 5.5.
• A Grok Bot weekly allowance, separate from your Cursor usage.
• Cloud agents, which is where a Bot's coding work lands.

Bugbot is there too, billed on usage.

What you don't get is the Grok chat app. And it doesn't stack: linking SuperGrok on top of a Cursor plan adds zero Grok Bot usage. Need more? Move up a Cursor tier.

The catch#

The weekly allowance covers the Bot's own work, not the cloud agents it launches. xAI's announcement says Bot usage won't count against your Cursor plan, but Cursor staff confirmed on the forum that those agents bill to your normal Cursor usage by design. So the agent that wrote this draft came out of my Cursor usage. Not Grok Bot's allowance. Mine. And once the weekly pool runs out, Grok Bot keeps going on on-demand usage if you've enabled it.

Read that thread before you point a Bot at your repos. For a hard ceiling, set an on-demand limit in the Cursor dashboard, or $0 to never pay extra.

What I'd tell myself this morning#

• One Bot per job. Standing rules go in the Bot's description, tasks go in the chat.
• Task, then skill, then routine. Get a task working once, save it as a skill, and only then automate it.
• Watch your routines. Each scheduled run re-sends the conversation's context, so pause the ones you don't read.
• Say where the Bot must stop. "Draft it, don't send it" is the strongest control you have. And never paste passwords into chat.
• Let the primary Bot find work. Its suggestions don't count against your usage.

What's next#

Next up is my boostmyfps product. I'm planning to have Bots help run it, with Polar handling payments.

Day one verdict#

Grok Bot pays off when you stop chatting and start delegating. One Bot per job. A clear rule for when each one talks. A cloud agent for the code.

If you code, run it through Cursor. $20 a month gets you the editor with Claude models, a Grok Bot allowance, and cloud agents for your Bots to hand code to. Set an on-demand limit first.

Day one. Ask me again in a month.