OpenAI just open-sourced Symphony — a tool that reads your issue tracker and assigns AI coding agents to every ticket autonomously — hitting 20K GitHub stars in 4 days. It solves the biggest bottleneck in AI-assisted development right now: orchestration. If you're using Cursor, Claude Code, or Codex and babysitting agents one at a time, this changes everything.
Symphony is a long-running Elixir daemon that polls your Linear board, spins up isolated workspaces, and launches Codex agents in parallel — up to 10 concurrent agents by default. The real product isn't the Elixir code though — it's the SPEC.md, a protocol-level spec for building orchestrators in any language. Five serial tasks that take 10 hours wall-clock time finish in roughly 2 hours with parallel execution. Same code, same review process, same quality — one-fifth the elapsed time. But it's still early: Linear-only in v1, security concerns with filesystem access, and the Codex API is evolving fast.
Does this sound like the way forward?