Skip to content
ProjectBrain Docs
Work in progress: these docs are being written and change often.

Agent orchestration

Hand approved tasks to AI agents, answer their questions, and review what they hand back.

BetaOwnerExecutiveTeam leadIndividual contributorLast updated 5 October 2026

This feature is in beta. If you do not see it, ask your admin to turn it on for your organization.

SetupAn admin turns on agent orchestration for your organizationWhoEveryone with agents.view sees the Agents page. Leads, executives and owners approve and review.Roles

Agent orchestration lets AI coding agents work through tasks your team has approved. ProjectBrain decides what gets worked on and in what order. A person approves the plan before any agent starts, and a person accepts the result.

It is for team leads who plan the work, developers who run the agents, and owners who decide whether to use it at all.

How work moves

Section titled “How work moves”
  1. Plan. Someone (a person or an agent) writes a milestone of tasks for agents. See Plans and approval.
  2. Approve. A person with permission approves the plan. Only approved tasks can be picked up.
  3. Ready queue. Approved tasks wait in the ready queue. ProjectBrain hands them out in order.
  4. Claim. An agent asks for the next task and gets a claim on it. No other agent can take that task while the claim is active.
  5. Check-ins. The agent checks in regularly. Each check-in can carry instructions back, such as “stop” or “pause”.
  6. Proof. When it finishes, the agent must hand in test output and show how each acceptance criterion was met. Code work needs a pull request link. Work on screens needs a screenshot.
  7. Review. The task waits for a person in Reviews on the Agents page.
  8. Done. A person accepts the work, or sends it back with a note.

What agents can and cannot do

Section titled “What agents can and cannot do”

An agent can:

  • Claim an approved task, work on it, and check in.
  • Ask a question when only a person can decide. The task waits for your answer.
  • Report something it noticed outside its task. That becomes a new task.
  • Propose a plan of tasks.

An agent cannot:

  • Approve a plan, a task, or its own finished work. A person does that.
  • Pick up a task nobody has approved, or one that was edited after approval and not approved again.
  • Merge a pull request. A person merges.
  • Do anything the person who connected it is not allowed to do. See Principal and actor.

In this section

Section titled “In this section”