AI coding agents need applicability checks, not just retrieval of architecture decisions
Overview
Theo Valmis argues in O'Reilly Radar that AI coding agents need to know which architecture decisions govern the artifact being changed, not only which decisions are relevant.
In a controlled repository he describes, an agent retrieved a decision banning PostgreSQL and ORMs but still ignored it, because nothing marked that decision as governing the storage module.
Valmis proposes separating applicability from retrieval, and using deterministic enforcement only for decisions precise enough to check mechanically. He treats the remaining decisions as guidance.
Written by AI from the articles below · updated Oct 13, 7:09 AM ET
Check the sources:
Article timeline
The articles in this story. Times are ET.
- O'Reilly RadarBlogRetrieving architecture decisions is not the same as knowing they apply
AITheo Valmis argues that AI coding agents need to know which architecture decisions have authority over the artifact being changed, not just which ones are relevant. In a controlled repository, a retrieved decision banning PostgreSQL and ORMs was still ignored because nothing signaled that it governed the storage module. He proposes separating applicability from retrieval and keeping deterministic enforcement for decisions precise enough to check, with the rest treated as guidance.
Heat trend
Not enough continuous observations to show a trend yet.