Gaining and successfully communicating insight and understanding from a proof you discovered is additional work that human mathematicians do. It's not just some side-product of proof-finding (at least not to the degree usually needed to publish). That AI models don't provide this is mostly proof that the model wasn't asked to do this work. Either because the prompter didn't know or didn't care

But there are also plenty of examples of humans providing technically correct proofs without any elaboration. Usually they get ignored, unless they are famous or the problem they solved was famous