The AI-Native SDLC Starts with Your Infrastructure

Anthropic's AI-native SDLC playbook breaks the software lifecycle into six stages, each producing a file. But it leaves out a key question: what environment does the coding agent run against when it checks its own work? If tests run against fake local services, passing only proves the code works against fakes, not against real cluster services. MetalBear's mirrord lets the agent's code run against real staging services instead of local mocks. The code still runs locally, but environment variables, secrets, and outbound calls go through the cluster network, and cluster traffic can be routed to it.