Waterfall 2.0: Controlling LLM-Driven Software Development With Stage-Gated Discipline
DevOps.com, Friday, August 21st, 2026
DevOps.com argues stage-gated waterfall becomes useful again as a lightweight control structure over fast AI-generated work.
The article argues waterfall is becoming useful again in the LLM era, not as a slow bureaucratic process but as a lightweight control structure for fast AI-generated work.
Waterfall 2.0 follows five stages: vision, architecture decision records, prototypes, an architectural skeleton, and code plus tests.
The key inversion is that backtracking becomes a strength rather than a failure, because LLMs can regenerate earlier artifacts quickly. That makes revisiting an earlier stage cheap in a way it never was with human-written documents. A shared knowledge corpus keeps the stages consistent as work regenerates.