OpenAI turns Linear boards into an army of coding agents

Symphony is an open-source specification from OpenAI that turns project management boards like Linear into an orchestration control plane for Codex coding agents. Instead of babysitting interactive sessions, any ticket filed on your board automatically gets a dedicated agent running continuously until the work is done and ready for human review.

What Symphony actually does

Interactive coding agents usually bottleneck on human attention. Most engineers burn out trying to steer more than three to five concurrent terminal sessions before context switching ruins productivity. Symphony flips the model from interactive chat to ticket-based delivery.

At OpenAI, teams using Symphony saw their landed pull requests surge by 500% within the first three weeks, with Linear founder Karri Saarinen noting a visible spike in newly created workspaces around the rollout.

Because the workflow runs through task tickets, non-technical team members can kick off real engineering work. A product manager or designer can file a feature request directly in Linear without ever cloning a repository. The agent implements the feature and files a review packet complete with a screen recording of the working feature in the product.

How it works under the hood

Technically, Symphony is simply a SPEC.md document in a GitHub repository outlining how agents should coordinate around task boards. OpenAI published a reference implementation built in Elixir to take advantage of its lightweight concurrency primitives, but Codex has also successfully generated functional versions in TypeScript, Go, Rust, Java, and Python directly from the spec.

Symphony treats the issue tracker as a finite state machine. It watches Linear, matches open issues to isolated agent workspaces, and manages execution dependencies automatically. For complex tasks, agents inspect the codebase, outline a dependency graph, and only kick off blocked steps when prerequisite tasks pass. The agents also manage the last mile of delivery, monitoring CI pipelines, rebasing commits, resolving conflicts, and retrying flaky test suites.

Where to get it

The specification and reference code are freely available on GitHub in the OpenAI Symphony repository. You can run the Elixir engine directly or prompt your preferred coding agent with the SPEC.md file to compile a custom orchestrator tailored to your internal stack.

The real prerequisite is a rock-solid codebase harness. Symphony relies heavily on automated smoke tests, Chrome DevTools automation, and CI guardrails to catch mistakes when humans aren't in the loop. Highly ambiguous architectural decisions still require hands-on terminal work.

Why it matters

Symphony shows how engineering organizations can stop treating AI like a novelty autocomplete tool and start managing it like an autonomous junior workforce. It is ideal for teams drowning in backlogged bug fixes, boilerplate migrations, and repetitive maintenance.

We think OpenAI made a brilliant architectural choice by open-sourcing a clean Markdown spec rather than forcing another monolithic SaaS platform onto developers. The bottleneck in AI-assisted programming is no longer how fast a model types code, but how seamlessly agents can fit into existing task boards without exhausting human attention.