June 17, 2026
I run Claude Code and Codex side by side on the same repos daily. Skip the agent-to-agent chat idea -- the handoff is the repo plus a committed spec file (CLAUDE.md / AGENTS.md) as the shared source of truth, and git branches or worktrees so they dont stomp each other.
What I wouldnt do is run both as co-builders editing the same files in parallel. Two agents in one folder makes more merge cleanup than it saves. I keep one as the primary builder and the other as an independent reviewer/cross-check -- catches a lot more than letting them both write.
For sharing what Claude knows with Codex, dont try to pipe one models memory into the other. Put it in the committed plan/spec files and let both read from that.