How developers at SpaceXAI use Grok Bot

Welcome back. If you’ve been hearing about SpaceXAI's Grok Bot (or seeing it all over your timeline) and not quite sure how to set it up or get the most out of it, this email is exactly what you’ll need.

In the next 5 minutes, you’ll learn how to set up your first Grok Bots and hand them the time-consuming parts of your dev work with a few simple prompts.

So, what’s Grok Bot? You can think of Grok Bot as a highly capable engineering intern with a computer of its own in the cloud. A bot has a name, one job, and a memory that carries over between tasks. If you pay for Cursor Pro, you already have it, with its own weekly usage pool separate from your Cursor allowance.

The one thing to know before you connect GitHub: Every bot on your account shares a single cloud computer, so logging into one bot logs you in to all of them. Don't store any credential on it that one of your bots shouldn't use.

Separately, keep bots off your own machine. Set local execution to "Never allowed" under Settings → General → Agent in the Grok Bot desktop app.

Get started: Install the desktop app, create a Bot with a single job, and send it a task that includes an outcome, a source, a constraint, and the point at which it should stop. Run it manually once before turning it into a routine. This tutorial walks you through the process.

Once you have installed Grok Bot, here are three workflows you can deploy right away:

The problem: Right now, you are the layer between a task and your coding agent (e.g., Claude Code, Codex). You gather context, write the prompt, watch the run, review, and merge.

The fix: Matt Palmer, who works on Grok Bot at SpaceXAI, moved that loop into a single bot called Loops. It pulls context from teammates' commits, drafts the prompt, launches a Cursor cloud agent, and watches the transcript. Review and merge stay with you.

Palmer says, “I talk to Loops, and Loops talks to a coding agent.”

Set it up: Install the Loops template with the GitHub and Cursor plugins, and name one repo in its description, then send this:

This is our first Saturday Special. It’s a quick 5 min email designed to show how to use the latest and greatest AI tools to code faster. Let us know what you think of this email in the poll at the bottom of this email (or by replying to this email).

We’re writing this email just for you, so your feedback will shape what we write next!

The problem: You open a PR at 6 p.m. and by morning it has failing CI, a merge conflict, and three unanswered comments from Bugbot or whichever review agent you run.

The fix: Lingxi Li, another SpaceXAI engineer, runs a routine every 30 minutes that checks each open PR for failing CI, conflicts, and unresolved findings, then sends the cloud agent to fix them.

A separate 3 a.m. audit sweeps dead code and bundle size, so most mornings include a fresh cleanup PR.

Set it up: Tell the bot that owns this job:

Use Test run before you enable it. A test run does real work, so keep merges behind approval.

The problem: When a Bot works through a web app, it clicks like a person. Each click costs tokens, and the browser is the slowest path to anything.

The fix: Sam Sokolin from SpaceXAI has the Bot record network requests from one run and turn them into a script. On the next run, it uses the same auth to call the API directly, skipping the UI.

His Bot, Not My Tempo, does the conversion for you.

Set it up:

How to start this weekend. Use a single repository, a focused bot, and one task that provides verifiable proof. Save it as a skill only after the result repeats, and schedule it afterward. The quickest sign that the system works is waking up on Sunday to a PR marked “Ready for Review” that you didn’t request.

You can also reply directly to this email if you have suggestions, feedback, or questions.

Until next time — The Code team