From idea to production, and back again
Twelve stages. Each has an agent doing the work and, where it matters, a person approving. Production problems return to testing as new bugs, so the loop never stops. Select a stage for detail.
An agent for every role you would otherwise hire for
Each agent has its own instructions, its own sandbox and only the tools its job needs. Agents that review never see the reasoning of the agent that wrote the change. Select a team to filter.
Adding an agent
A new role is a folder: its instructions, its trigger, its tools and its human gate. The factory picks it up and puts it on the board. Shown as the design for the real implementation; not yet run on a live server.
agents/devops-release-manager/
SKILL.md # role, rules, what it may and may not touch
agent.yaml # trigger: "story merged to main"
# tools: github-actions, terraform, vercel
# gate: production release needs a human approval
Works with the tools you already have
One configuration file per client says which tools they use. Where no dedicated connector exists, standard formats (JUnit, SARIF, OpenTelemetry, webhooks, MCP) cover most tools. Badges show what is real today.
A smaller team, with people on the decisions
The factory takes the repetitive work in testing, build and operations. People keep the judgement calls. These are design goals to prove in a pilot, not measured results.
Testing team
- Agents run the suites, explore the app and file de-duplicated bugs
- A triage agent checks each report against the requirements
- One QA lead reviews verdicts and owns the requirements
Development team
- Agents plan, build fixes and open pull requests with release notes
- A few senior engineers review and merge
- Nothing merges without a person
DevOps team
- Agents fix red builds, prepare infrastructure changes and run releases
- Agents roll back a failed release and report it
- One platform owner approves infrastructure and production
Who decides what
| Decision | Agent | Person |
|---|---|---|
| What the product should do | Drafts requirements | Approves |
| Is this a real bug? | Gives a verdict with reasons | Overrides, owns requirement gaps |
| Code change | Writes and reviews | Merges |
| Infrastructure change | Prepares the plan | Approves the plan |
| Production release | Prepares, deploys, verifies, rolls back | Approves |
What we will measure
From recorded data only: bugs found, triage accuracy, share of pull requests that pass review first time, time from request to release, and deployments. Roadmap