OpenAI Symphony automates Codex work

OpenAI released Symphony, an open-source orchestration spec that turns Linear issue trackers into control planes for Codex coding agents. (openai.com) The company published the spec and a reference implementation on GitHub on April 27–28, 2026, licensing the repository under Apache‑2.0. (openai.com) (github.com) OpenAI’s blog and demo materials say some internal teams and early adopters saw up to a 500% increase in landed pull requests within three weeks of using Symphony. (openai.com) (helpnetsecurity.com) OpenAI describes Symphony as an agent orchestrator that reduces engineer context switching by polling a project board, spawning one Codex agent per active issue, and running each agent in its own loop. (openai.com) (infoworld.com) Technically, Symphony spawns isolated workspaces per ticket, monitors CI and review signals, and can land PRs after human approval; the implementation exposes a raw linear_graphql tool call to avoid leaking Linear tokens to subagents. (github.com) (helpnetsecurity.com) The GitHub repository has attracted thousands of stars and forks since the announcement and includes a SPEC.md and an Elixir-based reference implementation meant for experimentation. (github.com) OpenAI repeatedly frames Symphony as a reference design for teams that have adopted “harness engineering,” warning it’s an engineering preview intended for trusted environments rather than a managed product. (openai.com) (github.com) Vendor and independent writeups note the same trade-offs: Symphony automates the issue-to-PR pipeline but shifts risk to repository hygiene, CI reliability, and token handling in production deployments. (helpnetsecurity.com) (allthings.how) Linear founder Karri Saarinen and OpenAI engineers said usage spiked on some workspaces after the release, and OpenAI published guides and a spec for teams that want to run Symphony themselves. (helpnetsecurity.com) (openai.com)