Will Kinsey

Agent Lab · Playbook 04 · Pattern: Lanes

One owner per topic

I run several agents. When more than one of them watched the same inbox and parent channels, I got duplicate checks and the risk of one agent acting on something another one was responsible for. The fix is boring and it works: every topic has exactly one owner.

Status
Live
Agents
Grok Bot (owner today), with history across ChatGPT agents and Hermes
Applies to
Email, parent channels, research topics
Last updated
October 28, 2026

The problem

Agents are cheap to start and easy to forget. Over a few weeks I had scheduled jobs in several tools. One check of my Codex scheduler alone found 31 saved job definitions, with 7 active at the same time. When the kids’ channel monitoring moved from one agent to another, the old two-hour check couldn’t be confirmed as shut down, so for a while two agents could have been running the same job.

Two owners for one topic means duplicate alerts at best. At worst, one agent marks an email read or files it before the agent that owns it ever sees it.

How it works

Agents and topics in the animation are examples.

Replay

  1. Name the ownerEach topic, such as school and kids, AI feature news or a home repair project, has one owning agent that reports on it.
  2. Everyone else stays outOther agents may see related email but don’t act on it or report on it.
  3. Hand off in writingMoving a topic means a written handoff: what the job does, its sources, its rules and its schedule.
  4. Turn the old job offThe previous owner’s jobs are disabled and confirmed off, not just forgotten.

Setup steps

  1. Inventory every scheduled job across every tool, including paused ones. Count what’s active.
  2. Write the ownership table: topic, owning agent, sources, where it reports.
  3. Give each agent the rules below, with its own topics filled in.
  4. For each handoff, use the checklist and confirm the old jobs are off before the new owner’s first run.

Copy-ready rules and handoff checklist

Written for this page from my working setup. A GitHub version will come with the skills shelf.

Topic ownership rules (give to every agent)

You own these topics: <list your topics>.
Other topics belong to other agents. The current table:
  <topic>  |  <owning agent>  |  <sources>  |  <where it reports>

For topics you own:
- Check your sources on your schedule and report in your channel.
- You are the only agent that marks read, files, labels or replies for this topic.

For topics you don't own:
- Don't act on related messages: don't mark read, file, reply or report on them.
- If you see something urgent, mention it once to Will and name the owning agent.

Handoff checklist (moving a topic to a new owner)
1. Write the handoff: what the job does, sources, allowlists, rules, schedule, saved state, known problems.
2. New owner reads it and does one supervised run.
3. Disable every job the old owner ran for this topic.
4. Confirm each one is off. "Probably off" is not off.
5. Update the ownership table and tell Will the date ownership changed.
6. Retired agent names stay retired: note them as legacy so nobody mistakes them for live services.

Guardrails

  • One owner, written down. If ownership isn’t in the table, the topic isn’t covered.
  • Non-owners don’t act. They don’t mark read, file, reply or report on another agent’s topic.
  • Off means confirmed off. A handoff isn’t done until the old jobs are disabled and checked.
  • Retired names don’t come back. Old agent names linger in files and notes; they must never be mistaken for a live service.

What broke and what I changed

  • Monitoring moved back and forth. The kids’ channels went from Grok Bot to a ChatGPT agent with a written handoff package, then back to Grok Bot on October 24, when the ChatGPT agent’s combined job and four standalone jobs were turned off.
  • A shutdown I couldn’t confirm. During the first move, the old check’s shutdown was recorded as unconfirmed. That’s why “confirmed off” is now a step, not an assumption.
  • Too many schedules. Several brief schedules and migration paths meant duplicate deliveries and gaps were both possible. An inventory is the first step of every cleanup.

Last updated October 28, 2026. Related: Playbook 02, five parent channels, one short update. Back to the Agent Lab