Taste is the answer our profession has reached for, and it is the wrong one. Not because taste does not matter. Because it is the thing a machine can hold.
Our profession has settled quickly on an answer to what AI does to it, and I think that answer is wrong in an instructive way.
The reply is taste
That reply is taste. Anyone can generate a screen now, so what remains is judgment, discernment, knowing which of the four options is the good one. David Hoang, who runs design for AI at Atlassian, has the line for it: when output is cheap, taste becomes more visible. Karri Saarinen at Linear locates it earlier, in understanding the problem well enough to know what should exist at all. Joel Lewenstein at Anthropic makes it a question of where a designer stays involved, and which problems still need a person at all.
None of them is wrong about the practice, and I would not argue with any of it as a description of the work. My objection is to what happens when a profession makes this its answer to an existential question.
Julie Zhuo has already said the uncomfortable part out loud. Taste, at least as most of us have it, is pattern recognition across a large amount of absorbed culture. That is precisely the kind of thing these systems are good at. What remains distinctly human, she argues, is not taste but agency.
I would put it more bluntly for our field. Taste is becoming the second seat at the table. It has the same shape as the first. A scarce-sounding property we assign to ourselves. Defended in conference talks, unfalsifiable from outside, and not something anybody else in the company has agreed to price. We spent fifteen years arguing that design deserved authority because design was strategic. We are now arguing that design deserves authority because design has taste. Both arguments are addressed to ourselves.
Opinion sorts people
What actually survives is narrower than agency and less flattering than taste. It is opinion.
Those are not the same thing, and the difference is the whole point. Taste is a claim about quality. This one is better than that one. It wants to be right, and it can be checked against a pattern, which is exactly why a machine can hold it. An opinion is a claim about what should exist, made by somebody who can be judged for making it. Whether it is right is not settled and mostly not settleable, and that is the property rather than the flaw.
Which is why opinion does the thing taste cannot. It sorts people. An opinion in the work is legible to those who share it and to those it offends, and both of those reactions are useful. Morality and ethics have gone the same way this decade. Contested, unresolvable from outside, and therefore the sharpest signal anyone has about who they are dealing with. Design is not exempt. A product with an opinion belongs to a culture and repels the rest of it. A product optimized only for quality belongs to nobody, which is a fair description of most generated work.
A model can produce an opinion. It cannot hold one, because there is nobody there to be wrong in public. Opinion costs something to have, and the cost is what makes it worth anything. That is the moat, and it is not craft.
What it costs to hold one
The reason this is not a comfortable conclusion is that opinion is expensive in a way taste is not.
Taste can be exercised privately. You can have immaculate judgment, apply it in reviews, and never once put your name against a position that could turn out to be wrong. Plenty of careers are built that way and they are not dishonourable. An opinion committed to a product is different. It is dated, attributed, and available to be held against you for as long as the thing exists.
Which is why most organizations quietly select against it. Not deliberately. A process that asks for consensus, a review that rewards the defensible option, a manager who wants the safe version by Thursday. None of that is anybody being cowardly. It is a system doing what systems do, which is reduce variance, and opinion is variance.
A standard encoded in a component survives the person who cared about it. A standard living in a head leaves when the head does. Taste is the first one only if somebody does the work of moving it.
So if you run a design organization, the useful question is not whether your people have taste. They do, or they would not have been hired. It is whether anything in the way you work makes it possible for one of them to be wrong in public, with their name on it, and still be there afterwards.
I've led design and product at Honor, Lyft, Grammarly and Pitch, and today I build at Noord. This began as two sections inside The Head of Design is not the head of design, which is about where design authority actually sits. Julie Zhuo on taste and agency is the argument this leans on hardest and then departs from. David Hoang, Karri Saarinen and Joel Lewenstein are quoted for the position I am arguing against, and none of them is wrong about the practice. Read one way this is a license to be contrarian, which is not what I mean. An opinion nobody would pay for is not an opinion, it is a preference.
© 2026 Renato Valdés-Olmos