Skip to content

Comment on Why AI code fails differently: What I learned talking to 200 engineering teamsparent

Comments

The "in-codebase" approach is the right one, but a YAML file with plain text is a half-measure. The most reliable rule that "lives next to the code" is an architectural test. An ArchUnit test verifying that "all routes in /billing/* call requireAuth" is also code, it's versioned with the project, and it breaks the build deterministically That is a more robust engineering solution, unlike semantic text interpretation, which can fail

AboutSource Built by g1lg1l

Hackerly is an independent reader for Hacker News, built on the public HN API. Not affiliated with Y Combinator.