Grok Bot automation: one front door, many agents
I built a team of AI agents on Grok Bot. I talk to one agent, the work goes to specialists, and nothing risky happens without my yes.
How it works
I talk to one chief-of-staff agent. It hands work to specialists for engineering, QA, copy, research, product and operations. Specialists never ask me for approvals. They go back through the front door.
Every assignment names the goal, the limits, what done looks like, what's off limits and who's next. Every handoff names the next owner and points to a file.
Every approval goes in one shared ledger with my exact words and the time. An agent acts only on an entry that matches its action and hasn't expired. If nothing matches, it asks the front door, and the front door asks me once.
Merges, deploys, publishing, DNS and outside messages wait on my yes. Spending money is banned outright. Agents can recommend, and I make every payment myself.
Long, repetitive local work goes to my A9 stack first. Hosted models take it only when the A9 is offline or the job is urgent.
As of Oct 2026, the operation runs 13 agents: the chief-of-staff agent and 12 specialists.
How the diagram reads, box by box:
- Me, arrow to the chief-of-staff agent on Grok Bot.
- Chief-of-staff agent, arrows to specialist agents: engineering, QA, copy, research, product and operations.
- Chief-of-staff agent and specialist agents, arrows to one shared approvals ledger.
- Specialist agents, arrow back to me for merges, deploys, publishing, DNS and outside messages.
- Specialist agents, arrow to the A9 local stack (Paperclip, Hermes, Ollama) for long, repetitive local work.
- A dotted arrow from the specialist agents to hosted models, used only when the A9 is offline or the job is urgent.
Built with a team of AI agents under my direction. I set the scope, made the product calls, and nothing shipped without my yes.