To me the core purpose of code review is clear - knowledge sharing. If the only person who knows about a particular change is the person who wrote it, then you have a critical point of failure. If there is an issue in a subsystem whose owner is away or who has moved on, that issue is likely to take much longer to resolve if the person on the case is looking at the code for the very first time.

An AI maximalist might say that code review is no longer necessary because in the case that there is an issue in a subsystem nobody is familiar with, you can simply ask the AI to read that source code and come back with a report of where the bug is and a proposal of how to fix it. And, since code review is useless anyway, might as well take the human out of the loop entirely - just have AI immediately commit the change and push it to production and iterate if or when another issue emerges.

This is the dream of autonomous, self-managing systems! Of course this dream is decades old at this point, and despite developing ever more complex systems it turns out that we were never quite able to do away with humans altogether. Thus, code review still appears to be useful. But it's only useful if everybody goes into it with the mindset that the goal is knowledge sharing. If the outcome of a review is not that everyone comes out of it with a good understanding of the purpose and function of the code being committed, then imo it was a waste of time.