It has become our favorite answer to AI-generated sameness. The reality is good products come from judgment built on facts, and the agency to use them when the room prefers opinions.
Ask Rick Rubin what he does and he’ll tell you he has “no technical ability” and knows nothing about music. In a 2023 interview with Anderson Cooper, the producer said artists pay him for confidence in his own taste.
https://medium.com/media/1f83a53654c7bc936b099303acddda54/href
“The confidence I have in my taste and my ability to express what I feel has proven helpful for artists.” — Rick Rubin
Everyone now wants that job.
Just typing this makes me squirm because I believe the field should be research and data driven as much as it can be. With humans even.
“Taste is the new moat” might be the industry’s favorite sentence of 2026: artificial intelligence made competent screens cheap, so the one thing left to sell is the judgment to pick the good one. A ProductLed interview in January went further, telling builders not to validate before they build. It’s about 45 minutes in.
I decided not to show the video inline because it adds fuel to the fire and their use of typography isn’t to my taste.
It’s come up with couple of authors I follow — The taste paradox by Nicole Alexandra Michaelis is a good read over at UX Collective, and Xinran Ma had a interview recently about it in UX — and it’s in the zeitgeist.
My point is that it’s the wrong word.
The moat crowd is right that judgment matters; they’re wrong about where it comes from:
- Taste is a preference. It comes from what you’ve seen and who you admire, and two people with different taste can disagree with no way to settle it.
- Judgment is a claim you can check. It rests on principles research has tested, plus evidence about your users, so it can be explained, taught, and shown wrong.
Judgment needs agency too, because someone has to raise those facts when the most senior voice would rather trade opinions. What Nicole Alexandra Michaelis and Xinran Ma are talking about Judgement i.e. experience they can back with facts and experience.
The short rule: if you can’t name the fact or research behind a call, it’s still an opinion, which is otherwise known as taste.
Admit taste is an opinion
In 2001, researchers in Bordeaux and Montpellier gave a panel of 54 tasters a white wine colored red with an odorless dye. The panel described it with the vocabulary of red wine.
Nothing had changed but the color, and that was enough to overwrite their noses.
That’s the trouble with treating taste as a standard: it’s an opinion that feels like a fact. Two skilled designers can look at the same screen, reach opposite conclusions, and neither can prove the other mistaken.
Taste feels like perception, but a lot of it is expectation. Apple Music is probably the best example, it’s Apple, but still a challenging application to use established as fact.
For example, I like driving Audi TTs. A lot.
The proportions, the way one sits on a highway on-ramp, the complete impracticality of the back seat, the beautifully designed cockpit, the race to 110 miles per hour in California.
That last part is just storytelling, really.
I don’t expect any of that to be anyone else’s taste, and nothing is wrong when it isn’t, because my taste in cars only has to satisfy only me. A product has to work for thousands of people who never met me for it to be profitable, and that’s where a preference stops being enough.
My expectation is that Audi TT’s are designed for my taste, and it doesn’t fail me, ever.
Design has its own version of the wine study.
In 1995, Masaaki Kurosu and Kaori Kashimura of Hitachi had people rate 26 ATM layouts and found that how usable a layout looked tracked its beauty far more than its inherent usability. The field now calls this the aesthetic-usability effect, and it cuts both ways.
Generative tools make this worse because they turn out polished screens by default, so every option in the review looks finished. That’s the blind spot in the moat argument: the less polish costs, the less polish can tell you.
It reminds me of a conversation I had with Scott Jenson — Generative AI makes a lot of beautifully just above average screens. I made that argument they’re still better than a lot of portfolios I have seen, even if the screens have less depth.
This is why “does it have taste?” makes a weak review question: it has no failure condition other than an opinion.
Ruiqi Zhou called taste unfalsifiable in a February essay named Taste Is the New Moat — And That’s the Problem. It’s a great read, and points out that a standard you can never fail works more like a status marker, and status markers reward whoever already holds status. Usually, that’s product management.
Build judgment on laws that have held for centuries
User experience borrowed its foundations from people who ran experiments, and they had computers in only some cases.
Here are examples:
- Paul Fitts showed in 1954 that the time to hit a target depends on its size and distance using fighter cockpits for research, and William Edmund Hick showed that decision time climbs as options pile up. Both findings have held up for seven decades.
- Jakob Nielsen and Rolf Molich developed heuristic evaluation in 1990, and the ten heuristics are still the default checklist. Each turns an opinion into a checkable claim, like a button too small for a thumb.
- Dieter Rams and his ten principles took what could have stayed personal taste and wrote it down where other people could use it, which is why teams still audit work against them fifty years on.
They are normative rather than experimental, but a checklist can be argued with. I keep plain-language summaries of Fitts’s Law, Hick’s Law, Nielsen’s heuristics, and Rams’s principles on ux-guidelines.com, a reference site I maintain. Every law has conditions, and knowing them is part of judgment.
Hick’s Law describes a choice among equally likely options, and a 1985 study of its exceptions found cases where adding choices barely slowed people down. Fitts measured people tapping a stylus between two targets, not a thumb stretching across a phone.
A law applied outside its conditions is just another opinion with a citation.
Principles don’t settle everything.
At Microsoft, Ronny Kohavi’s team reported that only about a third of well-run experiments improved the metric they were designed to move. The laws tell you what’s likely, and testing tells you what happened.
User research earns its place next to the laws for the same reason.
A law is general and your users are specific, so only watching them shows which steps they repeat and where they give up. Teresa Torres makes the cadence the point, defining continuous discovery as weekly contact with customers by the team building the product, rather than a study that lands after the decision.
Test beauty with the people who will use it
None of this means beauty is decoration. The same aesthetic-usability effect explains why users forgive attractive products, and appearance shapes trust and first impressions.
Don Norman gave the effect a mechanism in Emotional Design: a positive mood broadens thinking and makes people more forgiving of small friction. Jony Ive’s teams built the most admired hardware of their era and also shipped the butterfly keyboard, which Apple settled a class action over for $50 million in 2022 after years of complaints.
Taste operating at that altitude still needed someone to check whether the thing worked.
The mistake is assuming the designer’s eye can stand in for the audience’s. Nielsen Norman Group showed the same five screenshots to 229 adults aged 18 to 25 and 228 aged 35 and older,
The results?
- The younger group rated the flat designs slightly more attractive.
- The older, slightly less attractive.
Whose taste was right? Each group’s, for itself.
Beauty is in the eye of the beholder, especially in enterprise software, where the person who admires the demo is rarely the one who lives in the product eight hours a day.
Beauty is a user outcome, so measure it like one.
That’s where the line between fact and opinion falls here. What users report about a screen is a fact about your audience, while a design lead’s verdict is one more opinion, however practiced the eye behind it. The survey used a modified version of the Microsoft Desirability Toolkit, which asks people to pick words that describe what they see from a fixed list.
Score task success and time on task separately from those ratings, so a polished prototype can’t inflate the verdict on whether the thing works.
Teach the principles before the portfolio
In 2009, two researchers who had spent careers on opposite sides of this argument wrote a paper together. Daniel Kahneman had documented how judgment fails, while Gary Klein had defended expert intuition in the field.
Their joint answer on when intuition deserves trust came down to two conditions:
- The environment has to be regular enough to learn about who they’re designing for
- The person needs a real chance to learn it
Design taste rarely gets that chance. A designer ships a screen, moves on to the next project, and never finds out whether the choice helped anyone.
Confidence grows with years, but accuracy only grows with feedback.
Tell a designer two years out of school to trust their taste, after a portfolio course, maybe an internship, and very few closed loops between a decision and its outcome. What they have is exposure to showcase sites and admired apps, which produces taste that matches the feed, not the user.
Principles are the shortcut experience can’t hand out, and they come with standing. A junior designer who knows Hick’s Law can push back on a longer menu in her first critique, and a new hire who has studied the heuristics can flag a missing undo before anyone senior does.
Facts give people the agency that seniority usually keeps for itself.
Give everyone the agency to argue from facts
Taste has a politics problem.
When the standard is taste, the taste that counts usually belongs to the most senior person present. Ron Kohavi named it in a 2008 Stanford talk, Listen to Your Customers not to the HiPPO: the HiPPO is the highest-paid person’s opinion.
A taste culture is a HiPPO culture with better typography.
The fix isn’t to ban taste from the room, it’s to stop one person from owning it without evidence.
Everybody on a team has taste, and the new designer’s is worth as much as the vice president’s, which is exactly why it makes a poor standard: nothing decides whose version wins except rank. Facts don’t work that way, since a law or a session recording says the same thing no matter who reads it out.
Taste cultures have a patron saint they misappropriated: Steve Jobs.
He told BusinessWeek in 1998 that designing by focus group is hard. People, he said, “don’t know what they want until you show it to them.” It gets quoted as permission to skip research, though the same answer described watching trends and studying Apple’s installed base.
But even he met his match in Paul Rand.
When Steve Jobs needed a logo for NeXT in 1986, he hired Rand, whom he openly called the greatest living graphic designer. The friction came immediately: Jobs asked for a few options, and Rand refused, saying he would solve the problem and Jobs would pay him, and if he wanted options he could hire someone else.
Jobs agreed to the terms and the $100,000 fee. Rand returned with one design, the tilted black cube, presented in a hand-bound booklet that walked Jobs through the reasoning until the logo felt inevitable.
Jobs loved it, and later admitted that you don’t hand a genius a menu.
Because of this, Apple did research all the time and that gave Jobs credibility to make judgement calls. Those that claim they have the same truly understand the company, nor how to combine taste with judgement.
Even Steve Jobs’ famous 2×2 Matrix Strategy is rooted in science — Paradox of Choice.
The larger problem is survivorship bias, another old law. We remember the founder whose certainty paid off and forget the rooms where the same certainty sank the product with dollar amount attached to it.
Agency with receipts is the fix.
Session notes, retention data, and a heuristic all lose to a senior “that feels off,” unless the team agrees that facts outrank opinions.
With that agreement in place, the quietest person in the room can win by naming the law a design breaks or the data point it ignores. Peter Merholz argues that design executives earn that standing by making design their secondary concern.
The fair counterargument belongs to Doug Bowman, Google’s first visual designer.
In 2009 he left and explained why in Goodbye, Google: a team was testing 41 shades of blue, and he had been asked to prove whether a border should be three, four, or five pixels wide.
His complaint went past the trivia: with nobody senior who understood design, he argued, the company ran out of reasons for its decisions, and data became “a crutch for every decision” that kept anything daring from shipping.
$200 million later. Hard to argue that as an outcome.
Measuring everything is its own way of dodging judgment, so decisions slow to the test calendar, and designers who could settle a border width from a principle leave for places where that counts, as Bowman did.
Aim evidence at the choices that change what users can do. A corner radius needs no significance test, since a law settles it faster.
Choose facts over opinions
Judgment is knowing which facts apply, and where they stop applying: a law, a heuristic, a published study, or research on your own users. Where taste says “this feels right,” judgment points out that the target is too small to tap reliably and that Fitts showed why in 1954.
Agency is saying that out loud in the review, even when it cuts against the most senior opinion in the room.
Judgment tells you which facts matter, and agency is using them when the room prefers an opinion.
The new reality:
- Generative tools can produce fifty layouts before my morning coffee.
- Judgment cuts the pile fast, dropping the ones with targets too small to hit, menus too long to scan, and flows that ask people to hold too much in their heads.
- What’s left goes to research, where users settle what the principles can’t.
Teams that run interviews and usability sessions on a schedule build a store of facts they can reach for in any review, and teams that skip them are left with preferences and a deadline. Dan Maccarone spent a year rebuilding his studio’s process around these tools and found that it didn’t make the work faster, which he counts as the point: what changed was where the thinking had to happen.
Rubin’s approach works because a record has no task to fail. A designer who borrows that job description claims the authority without the conditions that make it hold.
Judgment also improves in a way taste doesn’t, if somebody writes the call down. Jeff Gothelf formalizes that in the Lean UX canvas: state what you believe will happen, then name the evidence that would tell you whether you were right. It’s the equivalent of the Change My Mind meme, and it’s brilliant.
Picture two designers.
- One moves a filter panel above the results because Hick’s Law favors fewer competing choices on the way to what people came for, predicts lower search abandonment, and checks the numbers six weeks after launch.
- The other spends the same six weeks studying award-winning screen shots.
Only the first one learned something new about their users, and only their judgment got sharper.
Conclusion
Taste isn’t going away, and it isn’t the real enemy.
Opinions shape style, brand voice, and the small choices no principle covers. The trouble starts when an opinion gets treated as a fact and wins arguments it hasn’t earned.
The moat story is half right.
Artificial intelligence does make average output cheap, and plenty of products will look alike because of it. The scarce skill isn’t picking the prettiest variation, it’s knowing which ones break what the field already knows, which ones deserve a test, and what the result means for the next decision.
The wine panel in Bordeaux wasn’t foolish; its members were human, and so are we. That’s the whole reason design has laws and methods at all.
Human perception drifts toward what it expects to see, and principles are how a field keeps itself honest across decades of new tools.
So learn the laws, read the research carefully, and give your team the agency to use both. Then let your users decide the rest. Judgment you can explain, backed by the nerve to raise it, will outlast taste you can only defend, and teams built on it will ship the products everyone else ends up copying.
Action Items
- Ask “what fact supports this?” instead of “does it have taste?” In critiques, every significant call should rest on a principle, a study, or something your users did. If nobody can name one, label the call an opinion and flag it for testing.
- Teach the laws before the look. Give junior designers Fitts’s Law, Hick’s Law, cognitive load theory, and Nielsen’s heuristics before mood boards. Principles hand them decades of feedback, and the standing to use it.
- Make it safe to push back with facts. Say plainly that a cited principle or data point outranks a job title, and back the junior designer who uses one against you. Agency that gets punished once rarely shows up twice.
- Test where the facts run out. Brand feel, new interaction patterns, and cases where two laws conflict all need research. Run a usability session, a few interviews, or a first-click test that week, and log the prediction behind it. Leave the border widths alone.
Taste is not a design methodology was originally published in UX Collective on Medium, where people are continuing the conversation by highlighting and responding to this story.