OpenAI has entered the agent orchestration space with Symphony, an open-source framework written in Elixir that transforms project management tasks into isolated, autonomous implementation runs. Rather than building yet another coding assistant, Symphony takes a fundamentally different approach: it monitors project boards, picks up tasks, spawns agents to handle them independently, and delivers completed pull requests with comprehensive proof-of-work documentation. Released under the Apache 2.0 license and currently in engineering preview, Symphony represents OpenAI's vision of how AI agents should fit into existing software development workflows. ## Key Features ### Autonomous Implementation Runs Symphony's core innovation is the concept of an "implementation run" - an isolated, autonomous agent session that takes a task from a project board, works on it independently, and produces a deliverable. Each run operates in its own context, preventing cross-contamination between tasks and enabling true parallel execution. ### Proof of Work Documentation Every implementation run produces a comprehensive documentation package that includes CI status reports, PR review feedback summaries, code complexity analysis, and walkthrough videos. This directly addresses a major concern in autonomous coding: how do you trust and verify what an agent has done? ### Work Board Integration Symphony integrates with project management tools like Linear to discover and claim tasks. It monitors project boards for new work items, evaluates whether they are suitable for autonomous execution, and begins implementation runs without human intervention. ### Safe PR Landing After an implementation run produces a pull request and it passes acceptance criteria, Symphony can autonomously merge the PR, closing the loop from task creation to code deployment. ## Technical Architecture Symphony is built primarily in Elixir (94.9%), which is a natural fit for the framework's concurrency requirements. Elixir's BEAM virtual machine excels at running many lightweight processes in parallel. | Approach | Description | Suitable For | |----------|-------------|-------------| | DIY Build | Use SPEC.md to implement a compatible system | Teams with custom requirements | | Reference Implementation | Follow elixir/README.md for setup | Quick evaluation and testing | The project follows a specification-driven architecture, with a detailed SPEC.md document that describes the system's behavior and contracts. ## Conclusion Symphony is one of the most architecturally interesting agent orchestration projects to emerge this year. By focusing on autonomous task execution with built-in verification rather than interactive coding assistance, it addresses a different and arguably more impactful segment of the AI development workflow.