Grok Bot
// September 2, 2026
Grok Bot is SpaceXAI's product for creating named AI teammates with jobs. Each Bot works from a persistent cloud computer with a browser, filesystem, terminal, connected apps, and memory. You delegate an outcome; the Bot uses the available tools and returns the result. (Official overview)
This page is based on SpaceXAI's current product material and documentation. I will update it as the product and my own field evidence develop.
What makes it different
- A job, not an empty chat. A Bot has a name, a focused role, and a description of what it owns.
- A persistent computer. Files, browser sessions, preferences, and working context can remain available between tasks.
- Work across tools. Bots can use connectors, MCP servers, APIs, a browser, and computer interaction when a direct integration is not available.
- Reusable execution. A successful task can become a skill. A reliable skill can later run as a scheduled routine.
- Visible control points. Sending, publishing, purchasing, deleting, changing permissions, and touching production should remain behind explicit approval.
A sensible first job
Start with one repeatable outcome that produces something you can inspect. A research brief, meeting-preparation pack, inbox triage draft, or weekly product-feedback summary is easier to test than a vague instruction to "help with operations."
Define five things before the first run:
- The finished result.
- The sources and tools the Bot may use.
- The rules it should follow when information is missing or stale.
- The actions it must prepare but not execute without approval.
- The evidence it should return so you can check the work.
Run the task once, correct it, then test it against a second input. Save it as a skill only when the output is dependable. Add a schedule after that. (Skills, routines, and automations)
The shared-computer boundary
Multiple Bots created by one user share the same computer, including files and logged-in browser sessions. SpaceXAI explicitly says this is not a security boundary between Bots. Treat the computer as one workspace: connect the smallest useful set of accounts, keep privileges narrow, and avoid mixing contexts that should not be visible to one another. (Approvals, security, and privacy)
Why I am watching it
Grok Bot sits close to the work I am doing with Juno and the Institute for Zero-Human Companies: give an agent a durable job, useful context, reliable sources, and clear approval boundaries, then learn from what happens when it meets real work.
The product also creates an interesting education challenge. People do not need another tour of buttons. They need to see a complete job, understand why it worked, know where judgment stayed with a person, and have a pattern they can adapt.
Read my practical introduction to Grok Bot, use the job-design framework to plan a first workflow, or see my independent Grok Bot Academy proposal for the complete beginner-to-advanced learning system.
Open Grok Bot