SDLC AI-native

Louis Claxton, of Anthropic's Applied AI team, published a six-stage playbook on 21 August 2026 (Plan, Design, Build, Test, Deploy, Maintain) in which the stages themselves are unchanged from the classical SDLC. What changes is the mode of execution: each stage commits the artefact the next one reads. Intent is captured by its original author as «intent.md», requirements and design merge into a `spec.md`, and the build freezes `plan.md` before any code is written. The commit chain becomes the audit trail.

Jason Clinton, Deputy CISO at Anthropic, framed the same shift in July 2026 as an Amdahl problem: with «Claude» writing roughly 80% of merged code and engineers shipping 8x as much code per quarter, review and monitoring become the bottleneck if they do not scale with production. His per-stage controls attach to named boundaries, which is «SFEIR»'s reading of why the cycle matters more than the controls: a gate is by definition something placed between two stages.

The measured claims come from vendors. Atlassian's data scientists Robbie Geoghegan and Fan Jiang report 19% more merged pull requests for repositories adopting «Rovo Dev», rising to 59-87% when three to five team members adopt it. Paula Hingel, citing the DORA 2025 report, notes that AI adoption correlates positively with throughput and negatively with delivery stability.

Gregor Hohpe supplies the caveat that cuts across all of it: an AI-driven SDLC punishes bad habits far faster, because it amplifies everything, dysfunctions included.