Symphony
Symphony is OpenAI's experimental framework that turns project management into autonomous agent runs. Instead of supervising individual coding agents, teams assign work items and agents handle implementation end-to-end—with CI checks, PR reviews, and proof of work built in. It's designed for codebases already using harness engineering patterns.
Script: Haiku 4 Voice: Inworld TTS 1.5 Max
Transcript
Justy OpenAI just released Symphony. It's a framework that lets you hand work to autonomous agents and actually trust them to finish it.
Justy Instead of watching agents write code line by line, you assign tasks from a Linear board. Agents run the implementation, check it against CI, generate PR reviews, and land the code themselves.
Justy The proof is built in. You get CI status, review feedback, complexity analysis, even walkthrough videos. You manage work. Not agents.
Justy The catch: it's an experimental preview. Meant for trusted environments only. And your codebase needs to be using harness engineering patterns already.
Justy This isn't some vague research thing. There's a reference implementation in Elixir you can run today. Or you can build it yourself from the published spec.
Justy If you've got a codebase that's ready for this, go check the Elixir README on the GitHub repo. Set it up and test it. See what actually lands without supervision.
Justy This is the shift from managing coding agents to managing work. That's the next frontier.