OpenAI’s engineering team replaced manual Codex sessions with Symphony, an agent orchestrator that pulls work from Linear tickets and runs continuously. The shift eliminated context‑switching by letting agents auto‑generate task trees, manage CI, and resolve conflicts.

Building on a repo of fully AI‑generated code, the team redesigned workflows, invested in rigorous tests, and treated Codex like a teammate. Symphony’s DAG‑based execution lets agents tackle dependent tasks in parallel, boosting efficiency.

Teams reported a 500% rise in landed pull requests within three weeks, while product managers could file feature requests directly from the tracker. Engineers now focus on high‑impact problems, leaving routine implementation to the orchestrator.

The approach demonstrates that large monorepos can offload repetitive coding to AI, reducing human babysitting. Future work will refine guardrails and expand agent skills, but the immediate payoff is clearer delivery pipelines and higher output.