POLYREQ Blog

Notes on the ADLC.

How to keep understanding, governance and proof when AI writes your software. Essays from the team building the governance layer for AI-driven delivery.

ADLC · Series part 4

An agent builds exactly what you asked for

A requirements defect reaches production about four and a half times more often than a coding defect. Everything in your pipeline checks code, and what checks requirements checks the writing. Why agents made that gap matter more.

Aug 25, 2026 · 8 min read  ·  Read the essay →
ADLC · Series part 3

The anatomy of the cost, and the two ways understanding gets lost

Design, code, tests and review can all do their job perfectly and still ship the wrong system. Understanding is lost twice, once when the requirements start wrong and again at every handoff between phases. What that costs, and why AI makes both worse.

Aug 18, 2026 · 11 min read  ·  Read the essay →
ADLC · Series part 2

One change request, two lifecycles: the enterprise path to the ADLC

Coding is around 12 percent of a change. Follow one change request through today’s AI-assisted lifecycle and through the ADLC, where the impact map is a query, business approves a prototype, agents get briefs, and delivery returns evidence.

Aug 10, 2026 · 8 min read  ·  Read the essay →
ADLC · Series part 1

From SDLC to ADLC: the software lifecycle when AI writes the code

AI is making writing code cheap and helps with design and analysis too, but that help has holes. It is only as reliable as the picture of the system it works from. Where the classic SDLC breaks under agent-speed delivery, and what the ADLC replaces it with.

Aug 3, 2026 · 8 min read  ·  Read the essay →