How We Run Lullwood With Eleven Claude Agents and One Release Train

Description and use

Lullwood uses 11 AI agents on a shared repo. By restricting blocking reviews to high‑risk changes, the team cut review time and doubled release velocity.

Problem

Reviewing every change in a multi‑agent development environment consumed more effort than creating new features, limiting release velocity.

Outcome

By limiting blocking reviews to high‑risk changes, the team reduced review overhead, doubled shipping throughput, and maintained acceptable defect rates.

Workflow

• 1. Agents submit pull requests to the shared repo.
• 2. Automated checks run on all PRs.
• 3. High‑risk PRs are routed to a Code Reviewer for blocking review.
• 4. Low‑risk PRs merge automatically after checks pass.
• 5. Release train incorporates merged changes and deploys to Vercel.

Evidence

In one week the two agents building the game ran 803 sessions between them, and the two agents gating that work ran 853.

Source

Docs, tests and copy ship the moment checks are green, with no review.

Source

The engine's core simulation, persistence, anything touching secrets or merge rules, and release cuts still need a blocking review before they land.

Source