Symphony

OpenAI's experimental tool that turns project work into isolated, autonomous implementation runs — monitor Linear board, spawn agents for tasks, manage work instead of supervising coding agents.

Links

• GitHub: https://github.com/openai/symphony
• Spec: https://github.com/openai/symphony/blob/main/SPEC.md
• Reference implementation: Elixir-based
• Related: harness-engineering-openai - Prerequisite approach

Overview

Symphony monitors work boards (e.g., Linear) and spawns autonomous agents to handle tasks. Agents complete work and provide proof: CI status, PR review feedback, complexity analysis, walkthrough videos. When accepted, agents land the PR safely.

Key shift: Engineers manage work at higher level, not supervise coding agents.

Status: Low-key engineering preview for trusted environments. ⚠️ Experimental.

Requirements

Works best in codebases that have adopted harness-engineering-openai practices. Symphony is the next step: moving from managing coding agents to managing work itself.

Setup Options

1. Make your own: Tell your coding agent to implement from SPEC.md
2. Use reference implementation: Elixir-based (see elixir/README.md)

Demo

Video shows Symphony monitoring Linear board, spawning agents for tasks, agents providing proof of work, and landing PRs safely.