ADLC

Chris Williams opens his seven-part series on the Agentic Development Lifecycle with a category-error claim: the SDLC was shaped over decades against specifically human failure modes, the ego that refuses critique, fatigue that multiplies errors, forgetting that loses context. Those defenses are useless, sometimes counterproductive, against a model whose failure profile is entirely different. Everything after follows from one rule: every phase, every gate and every loop must trace either to a precise model failure mode it defends against or to a model property it exploits. No inherited ritual without traceable justification.

What that yields is eight phases, from P0 Triage to P7 Distill, a deterministic gate between each pair, and exactly two mandatory human moments: spec approval after P1 Interrogate, and behavioral acceptance at P6 Integrate. Williams' reasoning for the gates is mechanical. A direct LLM-to-LLM handoff with no deterministic checkpoint compounds error rates; compilers, test suites and validators reset that accumulation to zero. The same logic drives his rejections: full-diff human review past 500 lines is theater, personas add nothing, coverage gates defend against the wrong thing.

The seventh part tests the cycle by building with it. The ADLC Toolkit is eighteen open-source tools around a frozen core, @adlc/core, published at github.com/voodootikigod/adlc, and its doctrine is frontier-free: hit precision targets with mid-tier models, because verifying costs less than generating.

SFEIR's eleven-phase cycle cites ADLC as an independent discovery of the same shape. Williams' own summary of the series: "replace trust with structure, and structure with measurement".