§ Agent control room Snapshot as of 8 November 2026
My AI agents, in one register
This is the register I use to govern the AI agents I run. It is the same dashboard the free guide below helps you build, with the sensitive parts removed. My private control room holds much more: the accounts, files and systems each agent can reach, and the fine-grained detail of what each one can do. That gives me a clear overview of everything every agent has access to. This page is a public snapshot of it, and I update it after each monthly review.
-
(Hermes)JobWork notes and briefingsGateDrafts and alerts only. I read each brief before a log goes up. She never archives or deletes without my written approval.Review due -
(OpenClaw)JobRuns a team of 7 agentsGateWorks alone on his own sites and repositories. No access to my personal accounts.Review due - CEOraJobManages 2 sub-agentsGateCan push code alone. Cannot delete repositories.Review due
-
PipJobDaily assistantGateAsks me before each send. Once I say yes, it can send anything as me.Review due - Weekly content agentJobWeekly contentGateSchedules posts only after my permission. I spot-check one voiceover in seven.Review due
How to read this page
Each row is one AI agent I run. The columns tell you what it does, where it has to stop and ask me, and whether I have checked it lately. That is all a register needs to do.
What is an agent?
AI that takes actions, not only answers. It sends the email, edits the file, posts the update or runs on a schedule. A wrong answer costs a re-read. A wrong action can cost a client.
What is a gate?
The point where an agent must stop and ask me, written so a stranger could check it. "I confirm every email before it is sent" is a gate. "A human reviews important output" is a slogan.
What do the colours mean?
Reviewed Checked in the last 31 days, with nothing unknown.
Review due Not checked in the last 31 days, or something is still unconfirmed.
No gate The agent has no written gate.
Why is everything amber?
I built this register on 8 November 2026, and none of these agents has had its first monthly review yet. When I run that review, the chips and the date at the top change. I would rather show you an honest amber than a green I have not earned.
How it stays current. After each monthly review I update my private register and re-publish this page. The date at the top is the date of that update, not the day you loaded the page. The review itself takes about 10 minutes per agent. The article AI agent governance lists the six checks and tells the story of the incident that made me start.
Left out on purpose: the accounts each agent can reach, the models, the tools and the schedules. Publishing those would be a map for anyone trying to get at them. Every agent has one owner: me.
§ Free guide
Build your own agent control room
The Agent Control Room guide is a PDF you drop into Claude. It interviews you, then builds your agent register, a dashboard like this one, and a monthly review. Made for small teams and enterprises early in their journey.