BMAD

Philippe Martin's February 2026 note frames BMAD, Breakthrough Method for Agile AI-Driven Development, as a plan d'urbanisme for agentic AI inside the SDLC. His counter-image is vibe coding: every developer negotiating alone with a contractor, extensions everywhere, no overall plan, wiring ready to catch fire. BMAD organizes agents IA as a disciplined construction crew instead. One clarifies a need into a real specification rather than a vague prompt, an architect proposes options aligned with standards, artisans generate code, tests and documentation, controllers track defects and regressions. The human stays architect in chief, holding the vision, the non-negotiable constraints and the structural calls.

Marco Mornati put that promise through a one-day test in March 2026, running specialized AI personas (analyst, product manager, architect) across market research, a PRD, epics and stories, technical architecture and UX design. He used Stitch for the interface to avoid the generic aesthetic of AI-generated screens, and gemini-cli for exploration to save his costlier tokens for development. His conclusion: architecture is where human expertise stays indispensable. The figures he cites show the gap: 93% of developers use AI assistants and 27% of production code is AI-generated, yet coding is 26% faster and delivery only 8-10% faster, while unsupervised AI code introduces 1.7 times more defects.

For a DSI carrying legacy, Martin's downstream zone matters most: agents map the code, flag risk areas and propose progressive refactorings. Neither note settles how juniors will learn without writing code by hand, or what remains of the product manager's role once technical profiles can run product preparation themselves.