My take on agents solving frontier math problems:

Agents think differently than humans, so *their* frontier is different.

As in Project Hail Mary, each civilization made discoveries the other didn't.

The interesting part about agents solving the Jacobian conjecture, etc., is learning which kinds of problems are easy for them.

E.g., we learned they are better than humans at finding counterexamples.

If I was a math researcher, my priority would be to understand the nature of their frontier; the low-hanging fruit is in problems that are in ours but not theirs.