When Everyone Can Generate, Taste Is the Job
As AI makes generating anything nearly free, taste becomes the scarce skill: calibrated judgment about what's good and why. Here's what taste is and how to actually build it.
Ask a model for ten logo concepts and you’ll have them before your coffee cools. Ask for twenty headlines, five architectures, a first draft of almost anything, and it arrives, competent and instant. Which sounds like a superpower until you notice the new problem hiding inside it. You now have twenty options and no idea which one is good. The generating got free. The choosing got harder. And choosing well was always the actual job.
Here’s the claim I want to defend: as generation collapses toward zero cost, the scarce and valuable human skill is taste. Not the ability to make something, but the ability to know what’s good and why, and to pick it out of a pile of plausible mediocrity. The person who can do that is about to be worth a great deal more than the person who can merely produce.
What taste actually is
Let me be precise, because “taste” sounds like a vibe and it isn’t. Taste is calibrated judgment about quality in a specific domain. Two words are doing the heavy lifting.
Calibrated means it’s been tuned against reality, not just asserted. A person with taste doesn’t only feel that one option is better. They can usually tell you why, and their why holds up when you test it against what actually works in the world. Specific domain matters just as much: taste doesn’t transfer freely. A brilliant editor of prose may have no taste at all in typography or system design. It’s earned one field at a time, which is exactly why it can’t be faked and can’t be downloaded.
And here’s the thing people miss. Taste was always the senior person’s real value, long before any of this. Think about what actually separates a senior engineer from a junior one. It usually isn’t that the senior writes code the junior couldn’t eventually write. It’s that the senior looks at three working solutions and instantly knows which one will rot in six months and which one will hold. That’s taste. The junior can generate. The senior can judge. We paid for the judgment all along and mostly pretended we were paying for the output.
Why AI raises the price of taste instead of lowering it
The intuitive fear is that AI commoditizes everything human, taste included. It’s the opposite, and the reason is simple once you see it.
A model can produce an enormous volume of competent, plausible output. What it cannot reliably do is know which piece of that output is genuinely excellent for your specific purpose, in your context, against a standard that may not be written down anywhere. It regresses, by design, toward the most likely thing. The most likely thing is, almost definitionally, average. Very good average, often. But average.
So the bottleneck moves. When making something took a week, the constraint was production, and the person who could produce was the prize. When making something takes a minute and you get fifty of them, the constraint becomes discernment. Someone has to look at the fifty and say “that one, because of this, and throw the rest away.” That someone is the bottleneck now, and the bottleneck is always where the value pools.
“Infinite competent output doesn’t make judgment cheap. It makes judgment the only thing that’s scarce. When anyone can generate fifty options, the person who can tell which one is good becomes the most valuable person in the room.
”
This is the same shift I’ve argued lives at the center of the leverage stack: the tools multiply your output, so what you multiply matters more than ever. Ten times the production with poor taste just means ten times more mediocre things in the world, faster. Taste is the multiplier’s multiplier. It decides whether all that leverage builds something worth building or just fills the internet with more competent noise.
The trap: outsourcing your taste to the model
Here’s the failure mode I’m most worried about, and it’s subtle because it feels like efficiency.
When the model is good enough, it’s tempting to stop judging entirely. It offered a headline, the headline is fine, ship it. It suggested a structure, the structure works, use it. Do this for a while and something quietly erodes: your own standard drifts toward the model’s default, which is the statistical middle of everything it was trained on. You stop choosing and start accepting. And because you’ve stopped choosing, your taste stops getting the reps that keep it sharp. It doesn’t just fail to grow. It fades.
The result is a slow regression to the generic mean, dressed up as productivity. Everyone using the same tools, accepting the same competent defaults, producing work that is individually fine and collectively indistinguishable. In a world where everyone can generate, sameness is the real risk, and outsourced taste is how you get there without noticing.
Taste is learned, not born
Now the liberating part, because “taste” gets talked about as though some people are simply born with it and the rest of us should give up. That’s wrong, and it’s provably wrong.
Consider how anyone becomes a sommelier. Nobody is born able to distinguish a great wine from a merely good one. They taste hundreds, deliberately, alongside someone who can name what they’re noticing, until distinctions that were invisible become obvious. The same is true of a chess player who can glance at a board and feel that a position is strong. Decades of research into expertise, going back to the classic studies of how chess masters perceive the board, found that masters don’t calculate more moves than amateurs so much as see the position in meaningful chunks built from thousands of hours of pattern exposure. That perception, that instant sense of good and bad, is taste. And it was entirely built.
So taste is not a gift. It’s a trained perceptual system. Which means it responds to training. The ingredients are known: high volume of examples, active comparison, articulating the why, and a feedback loop against reality. Miss any one of those and you get exposure without calibration, which is how someone can consume a thousand things and still have no taste at all.
How to actually build taste
Here’s the regimen. It’s not complicated, but it demands the one thing scrolling doesn’t: active attention instead of passive consumption.
Flood yourself with the best examples in your field
Taste starts as exposure, but only to the good. Go find the best work in your domain, the acknowledged great designs, essays, codebases, products, and study them at volume. Not casually. On purpose, with attention. You’re building the reference library your judgment will later draw on. You cannot recognize excellence you’ve never carefully looked at, and most people have looked at far less great work in their field than they think.
Compare two things and force a verdict
Exposure alone doesn’t calibrate. Comparison does. Put two pieces of work side by side, ones that are close in quality, and force yourself to decide which is better. The difficulty is the point. Easy comparisons teach nothing. It’s the near-ties that sharpen you, because they make you find the small thing that actually tips it. Do this constantly, with anything: two openings, two layouts, two solutions to the same problem.
Articulate the why, in words, every time
This is the step almost everyone skips, and it’s the one that turns a vague feeling into real taste. Don’t just decide A is better. Say why, out loud or on paper, in a full sentence. “A is better because it makes the one important thing obvious and lets the rest recede.” The act of putting the judgment into language is what converts a gut reaction into a principle you can reuse. A taste you can’t articulate can’t be trusted, taught, or improved.
Get feedback from reality and from people with more taste
Your verdicts have to meet the world or they drift into private fantasy. Ship the thing you judged best and watch what happens. Show your pick to someone whose taste you respect and listen when they disagree. Every gap between your judgment and a better one, or between your judgment and what actually worked, is a calibration event. Taste built in isolation goes strange. Taste checked against reality gets true.
Use the machine as a generator, and stay the judge
Here’s how taste and AI actually fit. Let the model produce the options, freely and in volume. Then do the human part: judge them hard, pick deliberately, and reject most of what it gives you. Used this way the machine sharpens your taste, because judging fifty options a day is enormous practice at discernment. Used the lazy way, accepting its defaults, it dulls your taste instead. The tool is the same. The disposition toward it is everything.
Do this for a while and you’ll notice the shift. Where you used to see one competent option and shrug, you now see the flaw in it immediately, and you know what would make it better. That noticing is the whole asset. It’s what makes you the person who can point the powerful tools at something worth building instead of just producing more.
Taste pairs with two skills I’d put right beside it. It needs agency to act on, because a refined judgment that never ships is just a private opinion. And it rests on genuine understanding rather than borrowed answers, which is its own discipline worth protecting when the answer is free. Generation is solved. What’s left, the choosing, the knowing-why, the refusing-to-settle, was always the human part. Now it’s the only part. Go build the eye for it.
