Anthropic's AI-Native SDLC Has Three Controls. It's Missing a Fourth.

Anthropic unveiled an AI-Native SDLC playbook recently, contending that code generation is no longer the bottleneck in delivery. The stages preceding it, such as planning, review, security sign-off, and release, continue at human speed. When most of the diff is written by agents, the disparity must resolve in one of two ways: review queues escalate until gains in speed diminish, or code is shipped under-reviewed.

Both scenarios are undesirable, especially for regulated entities, and even suboptimal for non-regulated organizations. To address these issues, the playbook proposes a structural solution rather than a procedural one. Every stage in the process archives an artifact that the subsequent stage can access: intent.md, spec.md, plan.md, the diff and its tests, the PR with review findings, and the incident record.

This chain of commits transforms into the audit trail, enabling human focus to concentrate at the gate between stages, rather than being spent re-deriving context at the outset of each stage. While the playbook offers valuable insights, it appears to be lacking a crucial control surface, a gap that its logic reveals.