Any given AI to date has had weaknesses where some humans do better.

If this remains true indefinitely, humans can always fill that niche and be rich.

If it doesn't, you're wildly underestimating how badly it can go wrong for humans, because "things humanity collectively can do which robots controlled purely by AI currently cannot do" currently includes "making compute and robots for the AI".

If AI gets good enough to close this loop, the "minimally necessary human buffer" is "zero"... and that includes the would-be-rich.