The way I see it is this

- if AI is so powerful that it is a superset of human capabilities in that it can not only solve questions but also come up with useful abstractions, then we just simply don't need mathematicians!

- if AI is not that powerful but can automate in-distribution tasks, then that allows mathematicians to explore concepts much faster. Overall positive thing for mathematics

We are obviously in the second camp (at least for now). Your job as a mathematician is to offload as much in-distribution task as possible to the LLM and just focus on the rest.