Post #1464077
2026-03-07 18:00 UTC
@jschuster @jfdm @cross @lindsey @krismicinski @csgordon @jeremysiek
I think you're thinking about this incorrectly.
Agentic coding seems to work best when it can get feedback. So you don't need to generate code. You just need to check it against the behavioral spec.
I conjecture that tools for AI will actually generate BIG error messages, because people won't read them but the AI will, and the more info it can give as to what went wrong, the quicker the AI can fix the code to pass the check.
Replies (1)
-
@jschuster@hachyderm.io 2026-03-07 19:42
@shriramk @jfdm @cross @lindsey @krismicinski @csgordon @jeremysiek ah, I see. I was thinking more like expressing structural constraints in UML and passing those as extra constraints to the LLM, along with the rest of the prompt/spec. 100% idle speculation for me, I haven't dived too deep into this yet.