INFRA Signal 503
Anthropic's AI SDLC playbook requires local tests but skips what they verify against
↗
Anthropic published a playbook for restructuring the SDLC around coding agents with six stages and file artifacts, but its Stage 4 feedback loop leaves unspecified what infrastructure those agent verification tests should run against.
A human developer can judge how stale a fake service has become; an agent cannot. For services that depend on other services, databases, and APIs, the question of what tests run against is the hard part, and the playbook's artifact chain records decisions people wrote down, not what is actually running in the cluster.
Written by elseif from the cluster below · every claim links back to a sourceThe three things worth knowing
Anthropic's playbook defines six SDLC stages, each committing a file artifact (intent.md, spec.md, plan.md, etc.) for the next stage to consume.
Stage 4 requires agents to check their own work against tests, builds, or screenshot diffs, but does not specify what those tests should run against.
Agents cannot judge how stale fake dependencies have become, making verification against real infrastructure more critical for agents than for human developers.
THE CLUSTER