I've found it useful in limited cases for writing optics which can be incredibly obtuse, sometimes boilerplatey, and yet ultimately accomplish what in other languages might be considered utterly trivial use cases... consider the following prompt and output:

    reply with one-line haskell source code ONLY: implement the function projectPair :: (Field1 s s a a, Field2 s s b b) => Lens s s (a, b) (a, b)

    lens (\s -> (s^._1, s^._2)) (\s (a,b) -> s & _1 .~ a & _2 .~ b)

... which can be shown to satisfy the three lens laws. If you can understand the types it is generally true that the implementation falls out much more easily, in a similar vein as "show me your tables, and I won't usually need your flowchart; it'll be obvious."

I suppose LLMs are good for this and other extremely constrained forms of boilerplate production. I consider it an incremental improvement over go codegen. Everything else I still tend to hand-write, because I don't consider source code production the bottleneck of software development.