OpenAI Symphony — open-source orchestration spec for coding agents

Summary

OpenAI released Symphony on April 27, an open-source spec (and Elixir reference implementation) that turns project management boards (Linear demonstrated) into control planes for coding agents. Every open issue gets an agent, agents run continuously in isolated workspaces, humans review the results. 20.5K GitHub stars, 1.8K forks. OpenAI reports 500% increase in landed PRs during internal deployment. Positioned as a reference implementation, not a maintained product — “a low-key engineering preview for testing in trusted environments.”

Implications

Symphony represents a new orchestration layer between project management and individual agent sessions. Combined with Codex’s persisted /goal workflows and permission profiles, OpenAI now has three levels of agent integration: individual session (Codex CLI) → project persistence (/goal) → portfolio orchestration (Symphony). This is the first vendor-published architecture for running coding agents at portfolio scale rather than session scale.

• Lifecycle thread: extends the agent lifecycle from “per-session” to “per-issue” — agents don’t stop when the session closes, they run until the issue is resolved
• Token economics thread: continuous agent runs consume tokens at scale; the efficiency gains from GPT-5.5 (40% fewer output tokens) become load-bearing at orchestration scale
• Enterprise deployment thread: issue-tracker integration moves agents closer to existing enterprise workflows (Linear, Jira, GitHub Issues)
• Competitive thread: no equivalent from Anthropic, Google, or Cursor yet — first mover on the orchestration layer