I Use AI as a Rubber Duck, Not a Co-Pilot, and My Code Got Better
Explaining the bug to it fixes half of them before it says a word.
I stopped asking the model to fix my code and started explaining the bug to it out loud, in the chat.
Formulating the question forced me to state what I assumed, and the assumption was usually the bug.
When it does answer, I treat it like a colleague's guess, not gospel. Some guesses are wrong on purpose-adjacent logic.
The speed win isn't the patch. It's that I debug faster because I have to be precise to get a useful reply.
A rubber duck doesn't write your code. Neither should this one, most of the time.
