I think one of the underrated skills in AI coding is knowing when to say NO.
The model will happily add another abstraction, helper, wrapper, package or whole bloody subsystem if you let it๐
Sometimes the best prompt is basically: absolutely not. Do it simpler.