OpenAI Open-Sources Symphony, a SPEC.md for Autonomous Coding Agent Orchestration

"OpenAI Symphony is an agent orchestrator that uses project-management tools, like issue trackers, as a control plan to coordinate multiple coding agents. Instead of developers managing interactive coding sessions, Symphony manages "tasks" by assigning each one to a dedicated agent that works autonomously to completion. Once a task is finished, a human is in charge to review the resulting output."

"Each engineer would open a few Codex sessions, assign tasks, review the output, steer the agent, and repeat. In practice, most people could comfortably manage three to five sessions at a time before context switching became painful. Beyond that point, it became harder for engineers to remember which session was doing what, monitor stalled agents, and maintain a clear mental model of ongoing work."

"So, instead of organizing work around individual coding sessions, each aiming at eventually merging a PR under explicit human supervision, Symphony structures its workflow around the core deliverables of a project as its workflow building blocks, including issues, tasks, tickets, and milestones. Symphony continuously watches the task board and ensures that every active task has an agent running in the loop until it's done. If an agent crashes or stalls, Symphony restarts it. If new work appears, Symphony picks it up and starts organizing work."

"In this model, agent work is not longer tied to PRs. An issue might instruct an agent to analyze the codebase and generate an implementation plan, then break it down into a tree of tasks that Symphony can schedule across other agents. Similarly, if an agent identifies an opportunity for optimization or refactoring, it can open a new issue on its own. In both cases, a human developer remains responsible for reviewing these generated issues before Symphony a"

OpenAI Symphony is an agent orchestrator that coordinates multiple coding agents by using project-management tools such as issue trackers as a control plan. Instead of developers running interactive coding sessions, Symphony assigns each task to a dedicated agent that works autonomously until completion. A human then reviews the resulting output. Symphony was created to address a human attention bottleneck caused by context switching across multiple coding sessions. It structures workflow around project deliverables like issues, tasks, tickets, and milestones. Symphony continuously monitors the task board, ensures agents run for active tasks, restarts agents if they crash or stall, and picks up new work as it appears. Agent work is decoupled from PRs, with agents generating plans, breaking work into task trees, and opening new issues for optimization or refactoring, all subject to human review.

Read at InfoQ