I do not think I had a single university course that was not ultimately assessed through supervised written tests. That matters, because homework was never a reliable anti-cheating mechanism. Even before AI, you could just pay someone smarter to do your homework for you, or ask older students for their assignments (and sometimes even their test questions, there's even a pg essay on this [0]).

You can cheat as much as you want on homework, but it won't help you on supervised written tests. At some point you have to sit down, unaided, and show that you can solve the problems yourself. So I do not see how AI substantially weakens the signaling value of a degree, at least in systems where the degree is backed by in-person written assessment. It may make take-home coursework less meaningful, but that was already the weakest part of the signal.

[0] https://paulgraham.com/lesson.html