paneherd

Guide

Run Codex and Claude Code together: a practical workflow

By Eren Çolak, who makes Paneherd · Updated 24 October 2026

Running Codex and Claude Code on one project works well if you keep them out of each other's files, give them different jobs, and watch both. Here is a workflow that does that. For the general version with other CLIs, see running several AI coding agents side by side.

1. One worktree each

Two agents in one checkout will eventually overwrite each other. Create a worktree per agent:

claude --worktree feature-api        # Claude Code makes its own
git worktree add ../app-tests -b tests
cd ../app-tests && codex            # start Codex inside the second one

Merge the branches when each is done. Details: Claude Code worktrees.

2. Shared instructions

Codex reads AGENTS.md and Claude Code reads CLAUDE.md. Put the rules both need (how to run tests, code style, folders to avoid) in one of them, and have the other file say to follow it, so they do not drift apart.

3. Give them different jobs

4. Watch both

Both agents stop and wait for approval, and a waiting agent does nothing. In VS Code, Paneherd opens each as a terminal in a grid and shows which are working, waiting or finished from their own hook events, sends one prompt to both, and keeps Claude Code's and Codex's usage limits at the top of its sidebar. The free version runs up to four panes at once.

Claude Code and other agent sessions in a grid in VS Code, with the Paneherd sidebar showing their status.

5. Spend two plans, not one

Claude Code draws on your Claude plan and Codex on your ChatGPT plan, so when one is near its limit the other keeps going. Check both before starting a long task: Claude Code, Codex.

Frequently asked questions

Is it worth the extra setup?

If you regularly hit one tool's limit or want an independent review, yes. For a single small task, one agent is enough. See Codex vs Claude Code.

Install Paneherd free