I'm under impression that current AI cannot handle 10% of edge cases that human can. Worse AI feels like to hallucinate so it's even harder to identify whether the edge cases happen.

How do companies plan to work around or mitigate that? Or they're betting on them being acceptable soon?