The read
Using two AI assistants side by side tends to mean copy-pasting between chats, unclear ownership and claims of 'done' that nobody re-checked. I wanted them to work like a small team with clear roles, their own copies of the work and proof attached to every result.
Game plan
- One shared rules file both assistants follow: who leads, who works, and which decisions always come back to George (merges, deploys, pushes, spending, logins).
- Every job gets its own copy and branch, a written brief, and a reviewer who did not build it.
- A job board and helper script track each job from proposed to verified, and a job only counts as verified when the lead has re-run the checks itself.
- Used on real work, including the 2026 rebuild of this site.
Final score
A tested prototype that lets George hand bounded builds and reviews between two assistants without relaying prompts by hand, while keeping every release decision with him.
- Status: tested prototype as of October 8, 2026.
- Claude leading with Codex working has been proven end to end with real runs.
- Used to split the 2026 portfolio rebuild between Claude and Codex.
What it is
A working agreement plus a small toolset that lets Claude and Codex operate as a two-person dev team. The assistant George is talking to leads; the other one takes a bounded job in its own copy of the code and returns a result that the lead verifies before anything moves forward.
Where it stands
A tested prototype, in use on real projects. Codex leading with Claude working is partly proven and still being exercised.