I don't know if we are talking about the same stuff, also let me first state that I don't like Effect, Zod, and the like, so I'm really talking about types here.
If you get them out of the way, I literally need to do the type inferring myself, not a problem if I'm looking at a small and trivial function, but if I'm trying to understand a "conversation" inside my code I really want that job already done by the compiler, this is also to tell that more of the "dumb/mechanical" computation is done by the compiler/interpreter the better, this is why we always wanted to get rid of syntax because we want to read english but at the same time we cannot use english because we don't want ambiguity. Writing clean code allow for the right compromise. I believe LLMs can write clean code with the right instructions, the same you would tell a junior dev, but one thing they will never succeed (unless they litreally copy) is architecture, because that is not teachable by writing (instructions/guides/etc.), only by doing and feeling.
If you get them out of the way, I literally need to do the type inferring myself, not a problem if I'm looking at a small and trivial function, but if I'm trying to understand a "conversation" inside my code I really want that job already done by the compiler, this is also to tell that more of the "dumb/mechanical" computation is done by the compiler/interpreter the better, this is why we always wanted to get rid of syntax because we want to read english but at the same time we cannot use english because we don't want ambiguity. Writing clean code allow for the right compromise. I believe LLMs can write clean code with the right instructions, the same you would tell a junior dev, but one thing they will never succeed (unless they litreally copy) is architecture, because that is not teachable by writing (instructions/guides/etc.), only by doing and feeling.