Slop amplifies
who's to blame?
PM uses AI to write the spec. Dev takes that spec and feeds it directly to a coding agent. Two days later: “the feature is done.”
Let’s stop there and think.
The spec looked solid. Acceptance criteria, edge cases, user stories.
Clean formatting.
Confident language.
It was also missing the one business rule that everyone on the team knows but nobody ever wrote down. The one that sometimes come up in customer call. The one that lives in three people’s heads and nowhere else.
The coding agent had no idea. It wasn’t in the spec.
So who failed here?
The PM - they disabled thinking while using AI to speed up a process they were already doing badly.
The developers - they disabled thinking and assumed an AI-generated spec was as reliable as a human-reviewed one.
The process - team had no clear step where anyone asked: “what’s missing here that we all know but nobody said out loud?”
Soooo… everyone failed.Which is the worst kind of failure, because there’s nobody obvious to blame.
---
AI is very good at making incomplete thinking look complete.
A half-baked spec written by a human looks half-baked. You can see the gaps.
A half-baked spec written with AI looks polished. The gaps are invisible. Until the code ships and the customer calls.
The teams that get this right add one step to the pipeline:
A human reads the spec before it goes to the agent. Not to rewrite it. Just to ask: “what are we all assuming here that we didn’t write down?”
That step takes twenty minutes.
slow is smooth smooth is fast
Skipping it costs two weeks. Do the math.