fundamentally LLMs don't, of course, have *sensibilities*. they can output code that looks superficially clean and organised but which has no *meaning*. good human devs have a sense of what's pleasing, what's understandable and maintainable. good code is satisfying, like a good crossword clue.
