Checklist
AI software delivery checklist for production teams.
Use this checklist when evaluating AI agents for engineering work. The goal is to get useful acceleration without losing quality, review, security, or cost control.
Workflow readiness
Before agents run delivery work, make sure the work can be decomposed, staffed, tracked, and reviewed.
- The workflow has named stages and expected artifacts.
- Specialist roles are clear for planning, implementation, review, and validation.
- Human review criteria are explicit before execution starts.
Execution safety
Generated work should happen inside controlled environments with bounded access and inspectable outputs.
- Execution uses a sandbox or isolated workspace.
- Secrets and credentials are not exposed to generated artifacts.
- External writes require policy, approval, and audit controls.
Quality and operations
Production teams need evidence that generated work was validated and can be traced after the run finishes.
- Validation checks inspect implementation and review outputs.
- Artifacts and logs are durable across retries.
- Usage, cost, and run status are visible to operators.