Agent-to-Agent Pair Programming: Claude and Codex Working Side by Side
The Next Frontier of AI Coding: Agents Pair Programming with Each Other
Axel Delafosse has introduced loop a CLI tool that launches Claude Code and OpenAI Codex side by side in tmux, connected by a bridge that lets them communicate directly. The concept mirrors human pair programming, where one agent acts as the worker and the other as a reviewer.
Why Agent-to-Agent Communication Matters
The best agentic workflows increasingly resemble human collaboration. Cursor discovered this with long-running coding agents. Claude Code Agent Teams and Codex Multi-agent features follow similar patterns. When Delafosse tested Claude and Codex simultaneously for code review, they gave different feedback but when both agreed, teams addressed 100% of those recommendations.
How Loop Works
- Launches Claude and Codex in parallel tmux panes
- Bridges communication between the two agent harnesses
- Preserves context across iterations
- Keeps humans in the loop for steering
Implications
As multi-agent harness adoption grows, agent-to-agent communication is emerging as a first-class feature requirement. The future of agentic workflows may look less like magic automation and more like familiar teamwork.
Open source at github.com/axeldelafosse/loop