Taste Beats Competence

There is a moment in every discussion about artificial intelligence when someone says, with audible relief, that the model still cannot do everything, and holds on to whatever it cannot yet do as though that were the real insurance policy. I find that moment interesting, because it almost always grabs the wrong end of the thing. The uncomfortable news is not in what the machine does badly, but in what it has come to do excellently, and it does so for everyone at the same time.

Imagine there were a model that could genuinely run your company well. Not spectacularly, not visionary, simply solid, competent, at the level of a good managing director on a good day. That would be an enormous advantage for you, until you notice that the very same model is, at the very same hour, running your competitor's firm just as well, and his competitor's, and so on, all the way down to everyone who can pay for access. Competence that is available to everyone in identical quality is no longer competence in any economic sense, it is infrastructure, about as distinguishing as a power socket.

This is precisely where something tips that most careers and most organizations took to be their core. A whole generation learned that reliability, high quality, and reproducible excellence were the road upward, and that was true as long as those things were scarce. High quality without variance, however, creates no competitive advantage the instant a system copies it on demand into every firm, and what remains, once the reproducible becomes standard issue, is exactly the thing that cannot be reproduced.

The scarce thing is no longer competence, it is deviation

The industry has by now found a word for this, it calls it the taste economy, and the word is more precise than it first sounds. When producing answers trends toward zero, the contest shifts to the question of which answer you even want, and no model answers that, because it is not a knowledge problem. It is a question of taste, and taste is nothing other than the ability to pick, from a thousand technically correct possibilities, the one that is right, and to discard the other nine hundred and ninety-nine with a reasoning you cannot look up.

Deviation cannot be averaged. A model trained on the average of everything produces, by construction, the average, and the average is exactly what anyone fears who has ever watched mediocrity roll off the line in series. Taste is the only mechanism that reliably does not land on the mean, because it comes from stance, from a person willing not to let a certain kind of wrongness pass, even when it would be statistically inconspicuous.

From this follows something that organizations say reluctantly: the person with the taste becomes more important than the entire apparatus built to deliver the product. For decades the organization was the answer to the problem of reproducibility, it existed so that quality would not depend on any single person. But if quality is available anyway and only deviation is scarce, then the structure optimized for reproducibility turns into a cost center, and the one person whose judgment the machine does not have becomes the actual asset.

Why handmade gets more expensive, not cheaper

The second observation cuts deeper, and it sounds at first like the art market, yet it concerns every product a human stands behind. Whoever buys a painting does not buy the surface on the wall, they buy the story that a human conceived it, and that story is the real object of the transaction. Research on how people value AI works shows this with almost rude clarity: the moment people suspect a work was made by a machine, they devalue it systematically, and not gradually but almost on an all-or-nothing basis, as though it were a contamination.

Seen through behavioral economics this is not snobbery but a clean instance of a costly signal. A signal is credible only when it is expensive enough that a cheap provider cannot fake it, and human effort was always the most expensive signal of all, because it bundles time, skill, and risk that cannot be feigned. As long as effort was rare, it was merely a precondition. In a world where the machine delivers the effortless in abundance, effort itself becomes the scarce good, and scarcity whose price rises with demand rather than falling has a name, it is called a Veblen good. Handmade therefore gains value not despite AI, but because of it.

Anyone who thinks, in the behavioral field, about the difference between a thing and the narrative about the thing finds here the cleanest formulation of the mechanism. What was sold was never just the object, it was always the credible trace of a human judgment, and that trace becomes, in a sea of machine interchangeability, the only piece of information that still counts.

The danger is not in the machine, it is in character

So that none of this reads as good news for people with taste, here is the uncomfortable half. AI is not actually the problem, it is the maximal expression of a culture that had already become results-obsessed and effort-averse, and outsourcing the effort is therefore less a technical risk than a risk of character. A machine that takes every exertion off your hands is, for someone who wanted to cut the corner anyway, not a tool but a permission.

The empirically supported part of this worry concerns learning, and it is uncomfortably concrete. When you merely query answers, you believe you are learning without learning, because the machine delivers the result and not the path. Studies on cognitive offloading describe an illusion of competence, in which the fluency of a well-phrased AI answer is mistaken for one's own understanding, and in one of the cited studies participants felt more confident in their answers with AI assistance even when those answers were wrong. That is exactly the constellation taste forbids, because taste arises from the path, from the possibilities discarded, from the friction you spare yourself when you only collect the result.

Here the circle closes. Taste becomes scarce, valuable, and teachable in precisely the measure that the temptation grows never to acquire it, because the machine delivers the result anyway. Whoever outsources the path outsources exactly what would have made them distinguishable, and notices it too late, because the outsourced effort feels like a win.

What this concretely implies

If this holds, then the most important consequence is not a question of technology but of training. Taste used to count as an incidental achievement of a life, as something you develop on the side, if you are lucky, with the right people, around the right things. That incidentalness is no longer affordable. If taste is the only scarce thing, then it has to move from a hope to a training program, to a routine in which you practice, again and again, the reasoned decision against the obvious mediocrity, until the judgment is faster than the temptation to take the first thing on offer.

This is meant soberly and not as a promise of salvation. The condition is set, whether you like it or not, the machine stays, and it gets better. The only interesting question is therefore not whether you find this good, but what you make of it, and the answer is surprisingly old-fashioned: you cultivate the one thing that cannot be copied, and you do it deliberately, because the incidental no longer suffices as a strategy.