Comment by jmole

The meta here is to use LLMs to make things simpler and easier, not to make things harder.

Turning tokens into a well-groomed and maintainable codebase is what you want to do, not "one shot prompt every new problem I come across".