SDLC

ISO/IEC/IEEE 12207 defines the SDLC as the normed frame of software production: requirements, design, development, test and QA, deployment, maintenance, measured by DORA metrics, with the stated purpose of building software correctly and reliably. Two accounts of what AI does to that frame are in open disagreement.

Olivier Rafal (WeNvision) holds that the phases themselves do not move: define, build, verify, deploy, maintain survive, and AI intensifies them. Louis Claxton, of Anthropic's Applied AI team, keeps the six stages and the control objectives while changing the execution mode, turning the process into a loop built on committed artifacts. SFEIR stretches the same sequence to eleven phases with three immovable human gates (Define, Plan, Ship). Chris Williams rejects the premise: the SDLC was shaped to counter human failure modes (ego, fatigue, forgetting), and a model's failure profile is different, so he builds the ADLC against eight catalogued model failures instead.

The recurring measurement is displacement. Claxton and SFEIR both locate the bottleneck in the stages still running at human speed on either side of the build. DORA 2025, cited by Paula Hingel and by SFEIR, correlates AI adoption positively with throughput and negatively with delivery stability. Marty Cagan's Four Big Risks expose the structural limit: the SDLC natively addresses technical feasibility alone, one risk out of four.

Whether the five invariant phases are a durable backbone or a habit inherited from human constraints remains open between Rafal and Williams.