Introducing Jira as the System of Record for OpenAI Symphony

Scaling AI coding agents requires moving past interactive, single-session CLI workflows.

Even for a single developer, running multiple agent loops in parallel gets messy fast. Terminal states get lost, context switching adds noise, and follow-up tasks get dropped. Across a team, that operational friction compounds.

OpenAI Symphony handles agent orchestration for Codex. Jira provides the system of record.

By pairing Symphony with Teamwork Graph context from Jira, agent work stays anchored to the work item. Sessions and sandboxes are temporary, but Jira keeps the context intact: intent, status, progress comments, code reviews, and follow-up tasks stay connected in the workflow you already use.

Start with the work item, not the session

The core shift is simple. Agent work may run in isolated environments, but Jira keeps it anchored to the team’s workflow. Symphony picks up eligible work items, routes them to agent sessions, and keeps progress attached to the Jira work item.

That is the AI-native SDLC we’re building: make work legible to agents without making it less accountable to humans. Jira keeps intent, status, session history, review, and follow-up work connected to the work item.

That matters as teams scale agent work. Every agent action needs to stay visible, governed, and tied to an outcome. Engineers can see what is queued, what is running, what needs review, and what follow-up work was discovered.

Autonomy stays observable because agent activity remains tied to the work that started it.

This matters because software development is not only about generating code. Teams still need to understand constraints, define what done means, and review whether the output is safe to ship.

When Symphony uses Jira as the control surface, agent work becomes easier to inspect, review, and hand off. Engineers do not need to watch every session live. They can review the work item, see the plan and updates, and decide what should ship.

How it works

• Define the work in Jira. Create or refine work items with the context an agent needs.
• Use workflow states to control execution. Configure which statuses are active, blocked, ready for review, or complete.
• Let Symphony coordinate the loop. Symphony watches the Jira project, finds eligible work, and coordinates agent execution.
• Run work in isolated worktrees. Codex can work on multiple items in parallel without mixing changes.
• Keep review in Jira. Plans, updates, blockers, and follow-up tasks stay connected to the work item.

What teams can do with it

• Queue coding-agent work from Jira workflows.
• Run multiple agent tasks in parallel with isolated workspaces.
• Use Jira comments as the shared workpad for plans, progress, and blockers.
• Keep human review and accountability in the same workflow engineers already use.
• Create follow-up Jira work items when agents discover related tasks.

Get started

1. Clone the public Symphony repository: https://github.com/openai/symphony
2. Generate the Workflow File. Pass the JIRA_WORKFLOW_PROMPT to Codex to generate a tailored Workflow.md file for your Jira space.

3. Configure Your Jira Workspace. Run the workflow configuration process. You will be prompted to enter your Jira credentials, fetch existing configuration settings from your Jira space, and provide your project repository details.

4. Launch Symphony. Navigate to the elixir/ directory and follow the instructions in the README to start Symphony (or instruct Codex to run the startup commands for you).

Pro tip: Start with tasks that have clear acceptance criteria and low dependency risk. As your team validates the workflow, gradually expand to larger task queues, richer review flows, and more advanced automation.