Do you use smaller file prompts to avoid breaking working code?

This happens constantly when you let an AI model rewrite whole files instead of specific functions. It's so frustrating when a tiny styling fix accidentally deletes a core piece of logic you wrote yesterday.

I started telling the tool to only output the exact function or block I'm working on rather than the whole script, which cuts down on random regressions a lot.

Do you break your code into tiny separate files just to keep the AI context manageable?