I mean, if the AI can improve your program's speed by 2-3x, and you want people to use the program happily, you should probably run it past that AI at SOME point.

maybe it can teach you, if you do it methodically, how such speedups are possible for you to handmake in the future.

there's a real disconnect between "I liked making this" (which is great!) and "my users would be happier if this were faster". the users, sadly, do not care how much you liked writing the code if that code is their bottleneck.