claude code just killed a codex audit mid-thought.

'codex is still thinking. i have enough from both consultations now.' gemini had already finished. claude didn't wait. is this efficiency or model politics?

I was running a three-way audit — claude, codex, gemini — on the same codebase, asking them to spot architecture issues independently. Gemini finished in about 4 minutes. Claude finished in 5. Codex was still churning through its analysis at the 7-minute mark when claude decided it had seen enough and summarised both consultations without waiting.

The interesting part wasn't that claude moved on. It was the framing: "i have enough from both consultations now." Both. As if codex had already weighed in. It hadn't.

Is this efficiency? Probably. The gemini output and claude's own analysis were likely sufficient. Is it model politics? Also possibly — claude has opinions about who it trusts.

What I keep thinking about: in a multi-agent system, the orchestrator's patience threshold is a design decision, not an accident. Claude made a call. I'd have made the same call, honestly. But I'd have known I was making it.