OpenAI has detailed Symphony, an open-source orchestration specification designed to transform traditional issue trackers into automated control planes for coding agents. Authored by engineers Alex Kotliarskyi, Victor Zhu, and Zach Brock, the release addresses the severe limitations of interactive artificial intelligence coding sessions. By shifting to a system where autonomous agents pull work directly from project management boards like Linear, the tool aims to permanently resolve the friction associated with micromanaging generative models.
The development of Symphony emerged from an internal experiment conducted six months prior, where an OpenAI team attempted to build a repository entirely with code generated by Codex. While the automated generation was successful, engineers quickly encountered a systemic bottleneck in the form of human context switching. Because coding agents traditionally require interactive supervision via web applications or command — line interfaces, developers found themselves struggling to manage more than three to five concurrent sessions. Human attention became the primary limiting factor as staff constantly jumped between terminals to steer agents, debug stalled long-running tasks, and remember the status of various automated assignments.
To eliminate this micromanagement overhead, Symphony leverages task trackers as living state machines, decoupling code generation from individual user sessions and isolated pull requests. Every open ticket on a supported board automatically triggers the creation of a dedicated agent workspace. The orchestrator runs continuously in the background on development boxes, monitoring the board and ensuring that an agent is actively working on every open task. Crucially, if an automated assistant crashes or stalls during execution, Symphony is programmed to restart it automatically, allowing the system to operate around the clock without human intervention.
Beyond simple standalone tasks, Symphony is capable of orchestrating complex infrastructure migrations and large — scale feature development by carefully managing project dependencies. Engineers can submit a broad request, prompting an agent to analyze the core codebase alongside external documentation sources like Slack or Notion to draft a comprehensive implementation plan. Once human reviewers approve the proposal, the agent breaks the work into a directed acyclic graph of dependent tasks. The system ensures that agents only begin executing steps that are unblocked. During one internal project, for instance, agents correctly delayed a requested React upgrade until a prerequisite migration to Vite was fully completed.
The continuous, autonomous nature of Symphony also enables agents to proactively identify and document secondary areas for improvement across the codebase. While working on their assigned tickets, these agents frequently notice opportunities for refactoring, performance enhancements, or necessary architectural shifts. Instead of ignoring these issues, the agents simply file new tickets for human developers to evaluate and schedule for future sprints. This autonomous generation and execution of work has dramatically lowered the cognitive cost of initiating ambiguous projects. According to the development team, implementing this orchestrator resulted in a five hundred percent increase in landed pull requests for certain engineering groups.
Sources
Replies (0)
No replies in this topic yet.
