The AI-Native SDLC Starts with Your Infrastructure

Anthropic’s playbook defines six development stages for coding agents, each generating a machine-readable artifact (intent.md, spec.md, plan.md, CLAUDE.md, REVIEW.md). In stage 4 the agent validates its output against a test suite and build, but the playbook does not specify how those tests should interact with real services. The article recommends using mirrord to execute the agent’s code against actual staging databases, queues, and APIs, allowing verification of behavior in a live environment rather than with simulated mocks.