Your AI Is Grading Its Own Work. That's Why Your Codebase Is a Mess
Fable writes. Kimi reviews. I decide. My two-AI loop for shipping real production code: one model plans and builds, a different model from a different company attacks its work, and the human keeps exactly one job.
In February my main repo had 228 dirty paths in it.
"Dirty paths" is git-speak for files with changes nobody had reviewed, committed, or even decided to keep. My AI coding assistant had been implementing features directly in the shared main folder of the project, session after session, piling half-finished work on top of half-finished work, until the place where my production code lives looked like a workshop where nobody ever puts a tool back.
That same week I looked at what my code reviews cost. I was asking Claude to review Claude's own pull requests — and each of those reviews burned through a lot of tokens. Worse, the reviews had a...
Copyright of this story solely belongs to hackernoon.com. To see the full text click HERE
