OpenAI released a new open-source tool called Symphony on March 5, 2026. It’s a framework that lets AI agents handle software coding tasks automatically. The system connects project management tools, like Linear, to AI models. This allows open issues to trigger agent workspaces on their own.
Symphony turns project boards into control systems. When an issue is marked “Ready for Agent,” the system picks it up and starts working. It creates a separate, sandboxed workspace for each task. This means one failing task won’t affect others running at the same time. The system can handle hundreds of tasks at once.
Symphony transforms project boards into living control systems — where a single status change can launch an autonomous agent to work.
The tool is built using Elixir and Erlang, which are programming languages known for handling many tasks simultaneously without breaking. A core piece called the Orchestrator acts like a manager. It checks issue states every 30 seconds and keeps everything in sync. If something fails, the system retries automatically using a method called exponential backoff.
Each agent writes code, runs tests, and creates a walkthrough showing its work before anything gets merged. The system also watches for human updates. If a person changes an issue, the agent stops working on it. Finished tasks move to review, and blocked tasks pause on their own.
Teams control agent behavior through a file called WORKFLOW.md stored inside the project. This file acts like a rulebook. It includes instructions, prompts, and settings that tell the agent how to behave like a human engineer. Symphony uses PostgreSQL via Ecto to persist state across all active implementation runs.
Right now, Symphony officially supports Linear for issue tracking and OpenAI‘s models for the AI work. Other integrations from the community are starting to appear. The system does require a well-organized codebase to work effectively. Agents can crash and need restarts, so the fault-tolerant design is important.
Symphony’s goal is to shift developers away from writing every line of code themselves. Instead, they’d focus on reviewing and directing the work. It’s described as a step toward truly autonomous software engineering. OpenAI reported a 500% increase in landed pull requests within three weeks of implementing Symphony across their workflows. OpenAI’s release marks a notable moment in how software teams might work in the future.
References
- https://www.youtube.com/watch?v=BX3D8tF7EVg
- https://www.marktechpost.com/2026/03/05/openai-releases-symphony-an-open-source-agentic-framework-for-orchestrating-autonomous-ai-agents-through-structured-scalable-implementation-runs/
- https://sjramblings.io/openai-symphony-autonomous-agent-orchestration/
- https://www.youtube.com/watch?v=aeFyQn_z_D8
- https://openai.com/index/open-source-codex-orchestration-symphony/
- https://sf.aitinkerers.org/technologies/openai-symphony
- https://openai.com/index/next-phase-of-enterprise-ai/