Understanding the fix before you keep it
A fix that works but that you don't understand is a loan, not a solution — and the interest comes due the next time something nearby breaks. Because the AI does the typing, it's easy to accept a green checkmark and move on without ever knowing why the code is green now. Resist that. Before you keep a fix, make the AI account for it:
- "Explain why this fixes it, in one or two sentences." If the explanation is vague — "this makes it more robust," "this handles the case" — push back. A real explanation names the specific cause and how the change removes it.
- "What was the bug, in terms of cause and effect?" You should be able to repeat the answer back in your own words. If you can't, you don't yet understand your own program.
- "What else touches this code path that I should re-test?" This turns a fix into a regression check.
This is the understand-what-you-ship philosophy applied to its hardest case. You don't need to be able to write the fix from scratch — that's the AI's job. But you do need to hold the cause-and-effect story in your head, because you are the one who will be there at 2 a.m. when it breaks again, and the only thing that helps then is knowing how the thing actually works.