Markus Eisele Urges 5-Step AI Coding Workflow With Temporary Change Briefs

1 articles · Updated · O'Reilly Media · Aug 17

Eisele argues AI-driven software teams should treat specifications as temporary “change briefs” that define only the next delta, then discard most prose after release.
His case is that code, tests, schemas, policies and telemetry are stronger, more maintainable sources of truth than sprawling Markdown specs, which consume context windows and quickly drift from production behavior.
The proposed 5-step workflow starts from existing code and operational evidence, defines intended changes and unknowns, adds research or stronger verification only when risk justifies it, then reviews implementation and preserves durable facts in native artifacts.
Eisele warns against agent-generated, exhaustive plans that can run to 1,000 lines, saying they raise human review burden, still diverge from implementation, and recreate waterfall-style documentation failures.
He cites research showing targeted clarification improved ambiguous-requirement repair by about 31%, while one benchmark found generated tests could double repair precision—supporting small briefs plus independent evidence over longer specs.