Sous les briques, le soleil

A note for future me.

18 months of coding with AI agents. My current stack: Claude Code with Opus 4.5 for writing, Codex with GPT 5.2 for reviewing PRs.

Opus feels different. I trust its output in ways I didn’t expect. Codex catches what I miss during code review—things I’d be embarrassed to ship.

Also trying oh-my-opencode and opencode in parallel. Interesting results. Orchestrating multiple agents unlocks capabilities none of them have alone.

I’m curious how long this setup lasts before something better appears.