-
134 green tests and a broken tool
My coding agent reported 134/134 tests passing, but the tool could not do its one job. Why green tests lied, and the six-role pipeline where the writer never grades its own work.
-
I stopped being the message bus
What a year of AI agents changed in how I build: five shifts, from pasting text between two chats to one tested rules file for every agent.
-
My AI reviewer mistyped a hash. The fix was to stop asking it.
6 of 245 good approvals lost to one wrong character, and the rule I took from it: the model judges, the code copies.