Executive Summary
OpenAI has announced Symphony, an open-source specification designed to orchestrate AI coding agents by using a project management board as a control plane. This system automatically assigns open tasks to autonomous agents that work continuously until completion, aiming to solve the bottleneck of engineers needing to micromanage individual AI sessions. By treating the issue tracker as a state machine, Symphony enables a new workflow where agents handle the bulk of implementation, allowing human engineers to focus on higher-level review and more complex problem-solving.
Key Takeaways
* Product Name: Symphony
* Primary Function: An agent orchestrator that turns a project management tool (like Linear) into a control plane for assigning and managing tasks for AI coding agents.
* Key Features & Capabilities:
* Continuously monitors an issue tracker and assigns an autonomous agent to every open task.
* Automatically restarts crashed or stalled agents to ensure continuous work.
* Agents can break down complex tasks into a dependency tree of sub-tasks and execute them in parallel.
* Capable of creating new follow-up tasks for refactoring or improvements they identify.
* Manages the full pull request lifecycle, including watching CI, rebasing, resolving conflicts, and retrying flaky checks.
* Format: Symphony is released as an open-source specification (`SPEC.md` file) rather than a pre-built software tool, defining the problem and intended solution for agents to follow.
* Stated Goal: To eliminate the human bottleneck of context-switching between multiple interactive agent sessions, enabling a massive increase in development output (citing a 500% increase in landed PRs for some internal teams).
Strategic Importance
Symphony signals a strategic shift from viewing AI as an interactive copilot to managing it as an autonomous agent workforce. By open-sourcing the specification, OpenAI is establishing a blueprint for the future of agentic software engineering, positioning human developers as high-level supervisors rather than hands-on implementers.