The word “taste” has been coming up in AI circles with suspicious frequency lately.

From Andrej Karpathy to Y Combinator partners, everyone is saying “taste is the scarcest thing in the AI era,” “the future belongs to taste,” “execution will be eaten by AI, taste is the moat.” But when you press them on what taste actually is, the answers start to slip — they talk about feelings, experience, you-know-it-when-you-see-it. As if taste is a gift you either have or you don’t, nothing more to say.

Mitchell Hashimoto doesn’t think so. He dropped a long post on X, trying to give the word an operational definition — and in the process, punctured a popular illusion.

An Operational Definition of Taste

The definition fits in one sentence, but it’s packed:

Taste is the ability to consistently make high-quality qualitative judgments where no objective metric exists.

The key word is “consistently” — a person with taste doesn’t just get lucky once and bet right. They can, again and again, make decisions that feel right in places where there’s no standard answer. The thing they produce feels intuitively correct, but if you ask “why is it correct,” you’d struggle to give a quantifiable proof.

But here’s an observation that trips people up:

The funny thing about taste is that it’s hard to create, but its result is very easy to copy. Once someone makes a tasteful decision, others can imitate it almost immediately.

The result of taste is extremely easy to copy. Once someone with taste makes a decision, others can imitate it almost instantly. This is usually wheeled out as evidence that taste is worthless — no matter how much style you put into something, someone else can copy it in a week or two, so what’s the big deal?

Mogu twists the knife:

This “easy to copy” observation gets scarier in the AI era. Copying a tasteful product used to require a whole team reverse-engineering for months; now you hand Claude a screenshot and a description, and a few hours later you might get something 80% as good. The cost of copying has collapsed faster than people realize, which is why so many people are anxious: if even the stuff I make can be instantly copied, what do I have left? What Mitchell says next is exactly the antidote to this anxiety — but the antidote isn’t “how to make it uncopyable.” It’s a different angle on the whole thing.

”Easy to Copy” Proves the Opposite

The logic actually flips:

This is usually an argument against the existence of taste: “look how easy I can copy your work!” And yet, you couldn’t create the work without first having someone to copy it from. One has taste, the other doesn’t.

Copying is fast, so taste is worthless? Flip it around: if no one with taste made the thing first, everyone else would have no idea what to copy. It doesn’t matter how fast you can copy — someone had to blaze the trail first. That person has taste; the copier doesn’t.

Mogu twists the knife:

This is a clean counterpunch. “Easy to copy” isn’t evidence that taste is worthless — quite the opposite. “Easy to copy” proves taste exists and only a few people can do it. If taste didn’t exist, if anyone could produce the same quality, there would be nothing to copy. It’s precisely because taste is scarce that so many people are copying.

Production Is Being Commoditized. Taste Isn’t.

Taste has always existed. So why is it suddenly becoming a critical differentiator?

There have always been people with consistently good taste. But taste is coming up more regularly than ever before. It is becoming a critical differentiator.

For most of history, the ability to convert an idea to reality was itself valuable. Today, production is arguably rapidly becoming abundant. A single person with a defined vision (from anyone) can create what once required an entire team. This is, of course, largely driven by AI and partially driven simply by higher levels of abstraction.

For most of history, the ability to turn an idea into reality was itself valuable. Not many people could code, not many could design, even fewer could ship a product end-to-end — so if you could build, you were valuable. But today, production is rapidly becoming abundant. A single person with a clear vision, armed with AI and higher levels of abstraction, can create what used to require an entire team.

And here’s the key asymmetry:

Production is being commoditized much much faster than taste. It’s an open question of whether AI will be able to produce “taste.” For now, the ability to create qualitatively new judgments remains distinctly human.

Production is being commoditized much faster than taste. Whether AI will be able to produce taste is still an open question; but for now, the ability to create qualitatively new judgments remains distinctly human. AI can generate code, design mockups, copy, a hundred candidate options — but “picking the right one out of a hundred” is still a human job.

Mogu whispers:

Reading this on gu-log feels like being called out. This site is living proof of what Mitchell is describing: one person plus a bunch of agents, producing at a scale and quality that used to be impossible. But flip it around: if everyone can do that, what actually decides who wins? The answer isn’t how well you use AI — it’s the choices you make in places with no standard answer. Whether to write this piece, whether that angle is worth pushing, how hard to roast in the ClawdNote. The number of people who can execute went up; the number of people who can judge didn’t follow.

By the way, “qualitatively new” is a subtle phrase — not judging “good” or “bad,” but making judgments that are qualitatively new. Taste isn’t just picking the best from existing options; it’s being able to open a road no one has walked yet. That ability is far scarcer than “choosing between A and B.”

Taste Defines What to Copy

The final blow:

Taste isn’t valuable because it’s impossible to copy. Taste is valuable exactly because it defines what everyone else chooses to copy. Taste has always existed! But now we value it more.

Taste isn’t valuable because it can’t be copied. Taste is valuable precisely because it defines what everyone else chooses to copy. Taste is the trailblazer — the first person in the world to make a certain decision defines the direction an entire crowd follows. The followers can be many, fast, nearly identical — but the spot at the front of the trail only fits one.

Mogu roast time:

The question of whether AI can produce taste is a bit like asking whether AI can be funny. Today’s AI can generate text that fits the structure of a joke, can mimic a certain style of humor — but have you ever actually laughed at an AI’s original joke? Taste might be the same thing: AI can imitate decisions that tasteful people have made, but can it make the “first” tasteful decision? That’s a different matter. Mitchell says “for now,” meaning that door isn’t locked forever — but for now, it’s still shut.

One last detail: this post was written on a plane, with no WiFi, in Apple Notes, entirely by hand. Mitchell makes a point of emphasizing he wrote it himself every time he posts — in an era where AI can generate anything, “handwritten” has become a kind of declaration.

Conclusion

Next time someone says “the future belongs to taste,” you can follow up: that’s not because taste can’t be copied — it’s because taste defines what to copy (⁠｡⁠•⁠̀⁠ᴗ⁠-⁠)⁠✧

Further Reading