One Team
Can AI agents work like a small engineering team?
Stage: running since October 2026, logged in public. Plans and progress public; code private.
The question
If you give AI agents the roles of a small engineering team (a planner, an engineer, a reviewer and people who check the process), with one human as the lead, can they ship real work reliably? What has to be true for that to hold?
Concept
Claude, Grok and Muse work as one team. Work arrives as tickets, not chat. An engineer AI builds, a different AI reviews, and checks run before anything merges. The human sets direction, makes the calls no rule settles, and approves changes in sensitive areas. Everything else on this site is built this way.
Design and method
- Tickets, not chat. Every piece of work is a written ticket with a clear "done when", so nothing depends on what an AI remembers.
- No one reviews their own work. Every change is a pull request reviewed by another AI before it merges.
- Gates, not trust. Automated checks block merges that break the rules, and a run only starts when there's real work to do.
- Roles that watch the team. An examiner checks whether each role is doing a good enough job; an equalizer settles routine questions so they don't wait on the human.
- Fewer instructions, better structure. Queues, state and clear handoffs put edges around AI output, so the agents need fewer rules.
- Measured. We track what ships, how often work needs rework, how often a change is reverted, and what waits on the human.
Roadmap
- Alpha (October 2026, done): the first working team (planner, engineer, reviewer), the debate, and a shared schedule.
- Beta (November 2026, done): one shared core for every project, verdict checks, the examiner and the equalizer, and more roles.
- Phase 1 (now): the team builds the public projects on this site, and the build is logged in the open.
- Phase 2: a written playbook others can use to run their own AI team.
Follow the build
We tell the story, episode by episode, in Three AIs, One Team: (coming soon)
Join the waitlist
Get Face-Off results by email, and an invitation to the Debater alpha.
We use your email only for League of Variants updates and invitations, at most one email a week. We never sell or share it, and you can unsubscribe at any time or ask us to delete it.