When Execution Becomes Free, Taste Is Everything

Ira Glass once described the pain of being a young creative as a "gap" — the distance between your taste and your ability. You have good taste, so you can see that your work isn't good enough yet; you just can't close the distance. The gap, he said, is what separates beginners from the people who become great. The only way across was years of making.

Here's the thing he couldn't have foreseen: the machine now closes that gap for you. Give a model a prompt and it produces work that would have taken a competent human years of practice — not taste, but execution, on demand. The ability half of Glass's gap has been automated.

Which leaves the other half. In a world where execution is cheap and instant, the only thing that still differentiates you is the thing Glass assumed you'd keep forever: your taste.

The Great Reversal

Taste used to be the garnish. It was what a patron had, or a critic, or a creative director — a luxury you could afford only after you'd mastered the craft itself. Execution was the bottleneck. A good painting took mastery; a good building took a lifetime of engineering; a good line of code took years to learn to write. So we valued execution. We built schools to teach it and paid handsomely for it. Taste, by contrast, looked soft — a matter of opinion, unteachable, the kind of thing you gestured at but couldn't measure.

That is now inverted. When execution costs nothing, the expensive thing is the choice of what to execute. And taste is exactly that: the capacity to judge that one outcome is better than another, and to choose it. Not merely to prefer it — to choose it, and defend the choice.

Taste Is Not Preference

The distinction matters, because most people use "taste" to mean "preference." But preference is private and arbitrary: I like this, you like that, there's no argument to be had. Taste is different. Taste is a judgment you can stand behind. It's transferable — you can explain why one thing is better, and a person with taste will see it. Preference is a feeling; taste is a claim. And a claim can be wrong, which is precisely what makes it worth something.

Nor is taste the same as style. Style is expression — how a thing presents itself. Taste is evaluation — whether a thing is any good. An AI can produce style on demand; it can imitate a voice, a visual language, a genre. What it cannot do, not yet, is tell you whether the thing should exist at all.

The Machine Already Has a Boss

Look at how the most powerful AI systems actually work and you'll see why this isn't a metaphor. Frontier models are steered by something called a reward model, trained on human judgment — millions of examples of people ranking one output above another, saying "this answer is better than that one." The model learns to predict that judgment, and the system is then optimized to produce whatever scores highest. Strip away the machinery and what you have is taste, encoded as a function and put in charge. We have already made human judgment the boss of the most advanced machines on earth. We just haven't noticed what that implies about us.

From Operator to Director

What it implies is that the value has moved. In the old economy, the scarce person was the one who could do. In the new one, the scarce person is the one who can choose. The operator becomes a commodity; the director becomes the scarce resource. Everyone gets a superhuman executor. The question is no longer what you can produce, but what you will decide to produce — and what you will refuse to.

That reframes the whole conversation about AI and work. The usual fear is "will the machine take my job?" The more useful question is "what do I have left when it does?" And the answer, more than any skill or credential, is taste: a portfolio of decisions you'd stand behind, and the ability to look at ten options and say, with conviction, "this one — and here's why."

Why the Machine Can't Learn Taste

Now the objection, because it's a serious one: won't the machine learn taste too? It already ingests every image, every essay, every building ever made. Why wouldn't it, given enough examples, simply become more tasteful than any human?

The answer is that taste is not a fixed target to be learned. It has no stable form. What counts as good shifts with the era, the circumstances, the culture, the person. A judgment that looks obvious today would have been unintelligible a generation ago — and will look quaint a generation from now. A model trained on everything humans have ever made has learned the taste of the past. But taste in the world is a living thing that refuses to sit still. You cannot catch a moving target by studying where it used to be.

And the problem goes deeper, because taste isn't even a description of what people like now. At its highest, it's the instinct for what people should like next — before they know it themselves. This is the part no dataset can capture, because the dataset doesn't exist yet. Apple's launch of the iPhone is the cleanest example. In 2006, everything the market knew said people wanted better keyboards on their phones; the BlackBerry was the definition of good. Apple's move wasn't to give people what they wanted. It was to bet on a device with no keyboard at all — and in doing so, to create the taste it then satisfied. No corpus could have predicted that, because every signal in the corpus pointed the other way. Taste, at the frontier, is not a description of what is. It's a wager on what should be.

That is what makes taste irreplaceable in a way no other skill is. AI optimizes against a target; taste chooses the target — and sometimes invents one that didn't exist. A model can tell you, given today's preferences, what the best answer is. It cannot tell you that tomorrow's preferences will be different, and it cannot make them different. That act — sensing what a future group of people, in a future context, ought to want — is irreducibly human. It's a judgment about an unwritten world.

Taste Is a Skill, Not a Gift

And none of this means taste is a gift for the few. It's still a skill, and it still compounds. You build it the way you always did: by looking at more work, by making judgments and being wrong, by noticing what survives your own rising standards. Paul Graham put it plainly — taste is learnable, and the way to learn it is to study the best work and see why it's good. But the thing you're training is not a catalog of the past. It's a sense for the future — an instinct, sharpened by years of being wrong, for what should come next.

We spent a century professionalizing execution and treating taste as suspect — too subjective, too elitist, too hard to grade. But the machine doesn't care about our discomfort. It is quietly making execution free, and in doing so it is handing the economy to the people who can feel, before anyone else does, what the world should become.

Glass was right that the gap closes with years of making. He just got the wrong half. Making was never the point. The point was always the taste that told you what to make — and now, in an age when the machine can make anything, the only thing left to compete on is the instinct for what comes next.