Taste in the AI era is not the ability to produce a good output. It is the ability to recognize, out of a hundred plausible outputs, which one should ship.
These are different skills, trained differently, and the second one has been quietly underpriced for a long time.
The skill of recognizing good comes from having seen a lot of bad. Not bad in one domain — bad across many. The person who has written mediocre marketing copy and mediocre research memos and mediocre product specs has, almost without trying, built a calibrated eye for what off looks like in each. They can't necessarily produce great work in any of those modes. They can almost always tell which AI output is wrong and why.
This is the part the generalist-vs-specialist debate has been getting backwards for a decade. The argument used to be that specialists win because they go deeper. That was true when the rare skill was producing. It is no longer true, because producing is no longer rare. The rare skill is choosing — which means the rare résumé is the one that has bounced through enough adjacent domains to have a calibrated no.
The people I trust most to spot a wrong AI output are people who would have been called unfocused ten years ago. They picked up the writing eye in one job and the analysis eye in another and the design eye in a third, and they were never the best at any of them. They are now extremely valuable, and most of them don't know it yet, because nobody told them their bouncing was the training.
I screen for this directly now. The people I want are the ones who can name the worst work they've ever shipped across three different domains, and articulate what was off about each. That's the calibrated no, made visible in an interview — and most of the people who have it still don't know it's their edge.