// When it breaks, triage before you rebuild · lesson 01
The rewrite reflex, and why regeneration feels like progress
When something an AI built breaks, the reflex is immediate and almost irresistible: delete it and regenerate. Ask for the whole thing again, fresh, and hope this pass comes out right. I have done it more times than I would like to admit, and I want to name why it is usually the wrong move, because the pull is strong and it feels productive.
Regeneration feels like progress because it produces motion. New code appears, the screen fills, something is happening. But motion is not diagnosis. You threw away a system that was ninety percent right along with the one line that was wrong, and you rolled the dice on a brand new generation that has its own fresh set of ninety percent right and its own new wrong line, somewhere else, that you have not found yet.
Why is regenerating so tempting and so often wrong?
Because reading is harder than generating, and the reflex is really an avoidance of the harder task. Understanding why the current code broke means sitting with code you did not write and tracing the actual failure. Regenerating lets you skip that. It trades a five-minute diagnosis for a coin flip that feels like work. The problem is the coin flip does not converge. Each regeneration is independent, so you are not climbing toward correct, you are re-rolling from scratch and hoping, which is how people end up on generation number nine of the same feature, each one broken in a new way.
There is a compounding cost too. When you regenerate, you lose the specific knowledge you just paid for. The broken version was evidence. It told you where the model's understanding was thin, what it got wrong, which assumption it made. Delete it and you delete the lesson with it, so the next generation is just as likely to make the same class of mistake, and you learn nothing about the actual problem.
The move that separates people who ship from people who spin is boring: when it breaks, stop, and find the one thing that is actually wrong before you touch anything. Diagnose, then act. The rest of this track is how.
The takeaway: Regeneration feels like progress because it produces motion, but it re-rolls the whole system to fix one line and teaches you nothing. When it breaks, diagnose before you delete.