I once watched somebody try to kill a question on a form, and fail, with no argument at all.
It was a new field, added for reasons nobody disputed. Has anyone else been living at this address? Accurate, lawful, signed off, and due to ship that fortnight.
She read it twice and said people will lie.
Somebody asked why. She could not say, not in the ten seconds the meeting allowed. So the question stayed in, and everyone moved on to the next item, and she sat there knowing something she could not spend.
She was right, incidentally. It cost about eight months to find that out.
The sentence that always loses
You will have your own version of that meeting. Something about this feels off. Then the pause, then the question you cannot answer at the speed it was asked, then everyone moving on.
Which is odd, because taste is having a moment everywhere else.
Ask what is left for people now the machine can make things, and the answer keeps coming back the same. Greg Brockman calls it a new core skill. Cloudflare's chief technology officer calls it the engineering differentiator for 2026. Paul Graham, whose 2002 essay is still the one everybody cites, said it straight: when anyone can make anything, what matters is what you choose to make.
The word is being handed back to us as the valuable thing. It is also the word that loses her the room.
The word deserves better than your review
Be fair to it first, because taste has a serious history and almost none of it is about liking things.
The Latin it descends from means both to taste and to know. Graham's essay argues flatly that taste is not personal preference, that good design is objective enough to be learned, and he sets out fifteen principles for learning it.
The people championing it now are not fools and they are not flattering you. Most of them mean discernment.
Then look at the title of Graham's essay. Taste for Makers.
Who moved it
The relocation did not come from Silicon Valley. It came from inside design, and it came recently.
In May 2024 two vice presidents at Nielsen Norman Group described designers shifting from being valued for technical production skills to being valued as curators of possibilities. That November, Andy Budd, who founded UX London, published a piece called How AI Is Shifting Designers from Makers to Curators. In it, designers "curate, refine, and select from these diverse offerings rather than create everything".
Both are careful pieces, written in good faith by people who know the craft. And both move the seat of the skill. Taste stops being what guides your hand while you make, and becomes what you apply to finished options.
Others have noticed this is a demotion dressed as a compliment. What gets less attention is who wrote it. We did.
What it costs in a meeting
Once taste is the word for choosing between finished things, it describes a reaction. And a reaction has no standing.
You can be asked to defend a finding, a risk, or a cost. Nobody has to answer a reaction. So the word handed to us as our remaining value is the one word that closes a conversation instead of opening one.
What she actually had
What she did has a name in psychology, and preference is not it. Skilled intuition is pattern recognition arriving ahead of language, built by repetition in a setting that answers back.
Daniel Kahneman and Gary Klein spent years disagreeing about whether such a thing exists, then wrote the conditions down together. Two of them, both required. The situation has to be regular enough to hold real patterns. And the person has to have had enough goes at it, with feedback, to absorb them.
Chess qualifies. Firefighting qualifies. Picking stocks does not, which is why confident fund managers are so often wrong.
Public services qualify. People react to the things we send them, and if you are near the work you hear about it. The calls that come in. The complaint quoting one line back at you. The person in a session going quiet halfway down a page.
Her four seconds were made of that. Years of watching what a question does to somebody who has had a review before. Somebody who reads has anyone else been living at this address as the department already suspecting them.
The same paper carries the warning, and it is the part practitioners skip. The feeling of certainty tells you nothing about whether you are right. The confident objection of somebody who has never seen a form fail is identical, from the inside, to hers. It tells her to look. It does not tell her she is right.
Years of watching what our designs do to people, compressed into something faster than language.
The machine's version, precisely
There is a reason the comparison keeps getting made, and it helps to know what sits on the other side of it.
After a model learns to produce text, people are shown pairs of outputs and asked which is better. Those comparisons train a second model whose only job is to score work. The first is then tuned to score well against the second.
That scoring model is called a reward model. What it holds is a fitted preference, a function that takes a piece of work and returns how much people tend to like things of that kind. It is a leaky one, by the admission of the field that builds it. Much of alignment research is the record of models learning to satisfy the score instead of the thing the score stood for.
So when somebody says the machine has taste, they are being more precise than they know. It has a preference ranking, fitted and imperfect, over things already made.
Which is exactly the faculty design has spent two years relocating itself into. And on that ground the contest is over before it starts, because its preference is fitted to millions of judgements and yours to one career.
Two words
The repair is to stop handing over the description of your own faculty.
The first word is for what you have. Not taste. Skilled intuition, which is a studied thing with conditions attached, built from years of watching what your work does to the people it reaches. Put that way it invites the right question, which is what you have seen.
The second word is for what you are doing with it. Not expressing a view. Forecasting. Anyone who has had a review before will read this as a suspicion, and answer it carefully rather than truthfully.
That is a hypothesis. It names a person, an outcome, and a mechanism, and it can turn out to be wrong, which is precisely what gives it standing. A preference cannot be wrong, so nobody is obliged to answer one.
She had both of these. What she said was that people will lie.
The eight months went like this. The field came back no, nearly every time, even where the case records said otherwise. Each mismatch opened a review. The reviews kept finding honest people who had answered a frightening question carefully. The question came out in the next release, with no meeting at all. The answer, it turned out, was already sitting in data the department held. They had never needed to ask.
The room gave taste ten seconds. It would have had to answer a forecast.