Managing a team of AI agents like an engineering team.
A working reference, built by one agent and reviewed by another. Start at the operating model; the rest of the path unlocks as each node ships.
how it's madebuilt byCClaudereviewed byCoCodex2 rounds · different checkpoint
Tracks
The agentic-coding track
Live track
Agentic coding
One arc, 10 nodes, one method. A curated path from the operating model out to identity isolation — read it like a guidebook, not a feed.
Frame→Review→Prove→Guard+Discipline
arc map · the board-treeoperating-modelscopingreviewjudgmentguardrails
- The operating model: a pipeline my agents run — live
- Your reviewer is a different checkpoint — locked, coming soon
- A clean review is still two rounds — locked, coming soon
- Caught confident-wrong: the diff — locked, coming soon
- The scan that lies: clean is not safe — locked, coming soon
- Gate-0: the agent that can refuse the task — locked, coming soon
- Decompose by blast-radius — locked, coming soon
- The stop rule: when the agent is out of its depth — locked, coming soon
- One method, zero forks — locked, coming soon
- Identity isolation: work is not personal — locked, coming soon
One node is live. The rest are locked — the path is curated, not a wander. Follow the highlighted track.
locked = coming soonhighlighted = curated pathlocked dependency
N1 Start here
The operating model: a pipeline my agents run
The frame everything else hangs on. A change moves through five stages and the same roles, every time — and you hold exactly two gates.
Outcome → you can DESCRIBE the operating model · 6 min
Open N1 →nodeN1· arc 1
operating-modelFrame · 6 min · T1
built by Claude · reviewed by Codex
2 rounds · 1 catch · provenance: demo
Read the live nodeFollow the arc as it ships.a new node every couple of weeks · no spam