Taste Is the Last Moat

The skill that everybody mocked for three decades has, quietly, become the most valuable thing a working creative can have. Taste, for years, was the embarrassing accusation; if you didn’t have a metric, if you couldn’t justify your choices on engagement data, if you said “I just like that one better,” you were considered slightly suspicious by the data-driven creative-industry adults. Taste was the thing the creatives talked about and the marketers gently overrode. Taste did not scale. Taste was not measurable. Taste was a luxury that you got to indulge after you had hit your KPIs. The whole posture of the late-2010s creator economy was that taste was nice but secondary; what mattered was production volume, distribution, and feedback-loop optimization. The taste people lost that argument so completely that most of them stopped having it.

Then the machines started producing things, and the argument flipped overnight, and almost nobody is talking about how violent the flip was, so I am going to talk about it.

When you can generate a thousand competent images in an afternoon for the cost of an espresso, the skill that matters is not generation. The skill that matters is choosing which one of the thousand to use, which seven hundred to throw away, which two hundred to revise, which hundred are too obviously machine-shaped, which forty are interesting in a way the algorithm did not intend, which three are actually worth showing to anyone, which one is the actual answer to the brief. That decision sequence is taste. And taste is the only thing in the entire creative production pipeline that has not been substantially automated, because taste is not a function. It is a judgment, made by a specific person with a specific history of looking, against criteria that are themselves the artifact of years of education-by-immersion that no model can shortcut.

This is what people mean, even when they don’t have the language for it, when they say AI-generated work feels “off.” The work is not technically incompetent. The work is, increasingly, technically excellent. What it lacks is the editorial intelligence that makes choices, and choices are what make a finished thing legible as art rather than as the output of a machine.

Steve Jobs got there in 1995

There is a 1995 PBS interview with Steve Jobs that you can find on YouTube in six minutes. The relevant section is about Microsoft. Jobs says, fairly bluntly: “The only problem with Microsoft is they just have no taste.” Then, leaning into it: “They have absolutely no taste. And what that means is, I don’t mean that in a small way, I mean that in a big way. In the sense that they don’t think of original ideas, and they don’t bring much culture into their products.”

In 1995 this was a slightly weird thing for a CEO to say. By 2026 it sounds prophetic. Microsoft’s products were technically capable; what they lacked was the editorial judgment that turns capability into objects people love. Jobs’s bet for Apple was that taste, specifically taste, was the defensible advantage no competitor could replicate by adding features. He was right for twenty years.

The applicability of this to a 2026 working creative is direct. Your competitors are no longer other working creatives. Your competitors are AI tools that can produce capable work in seconds. The market position you can defend, the one nobody else (no human, no machine) can produce in the volume you can, is the specific judgment your taste produces. Not the work itself. The choosing of which work goes out under your name.

What taste actually is

Taste is widely treated as innate or mystical. It is neither. It is, in fact, a measurable cognitive faculty, and we know roughly how it develops.

Taste develops through high-volume, high-context exposure to good work, in the field you care about, over years. The “high-volume” part is what gives you a wide reference set. The “high-context” part is what teaches you why the good work is good (not just what it looks like). The “years” part is what consolidates it into intuition you can use without conscious effort. The faculty is real. It just takes time, and most people will not invest the time, which is why taste remains scarce.

Susan Sontag’s Notes on Camp (1964) is, among other things, a treatise on taste. She does not define camp as a checklist; she defines it as a sensibility, which is a particular kind of trained perception. Her ability to do so was the result of decades of obsessive looking, reading, and conversation in specific aesthetic communities. The essay is a working demonstration of taste in action. The reader can almost feel the cognitive faculty being deployed, paragraph by paragraph, against examples Sontag had spent her life accumulating.

This is the part of the taste argument that’s hard to operationalize. You cannot read three articles about taste and develop it. You cannot take a six-week course in it. You cannot purchase it from a creator-economy product. You can only develop it the way Sontag did; by looking at thousands of objects, in their contexts, over years, with the specific intention of building your own internal grading system. Most contemporary working creatives have not done this. Most haven’t even started.

Why taste is uniquely AI-resistant

I want to be specific about why taste, of all skills, is the one machines have the hardest time with.

AI systems are trained on the output of human taste, but they don’t reliably reproduce the judgment behind the output. They produce, instead, a statistical median of the outputs they’ve seen. The median is, by definition, not what the most discerning humans would have chosen. The median is what the average discerning human would have chosen, which is a different thing.

Practically, what this means: an AI tool can generate a thousand images “in the style of Wes Anderson” and they will all look like the average of Wes Anderson’s known visual choices. None of them will look like an actual Wes Anderson frame, because Wes Anderson’s frames are the result of his specific aesthetic decisions, which are by definition deviations from the average of his style. The deviation is the taste. The average is not.

This is a structural limit of statistical generation, not a temporary technical limitation. As long as we’re training models on existing output and asking them to produce more, the output will trend toward the median of the training set. Taste, defined as the ability to recognize and produce specific deviations from the median that are aesthetically valuable, is the inverse of the operation an LLM is performing. You cannot improve a tool to do the inverse of its core operation; you can only build a different kind of tool.

What this means for working creatives, specifically

If you are a working designer, illustrator, writer, photographer, filmmaker, or any other creative whose income depends on producing work people pay for; the pricing structure of your career over the next five years is going to bifurcate.

The lower tier (work optimized for the average) will be priced down toward AI-generation cost, which is approximately zero. This tier is gone. Don’t compete in it. The clients in this tier do not know taste from prompt-output and will go with whichever is cheaper. They are not your customers anymore.

The upper tier (work that demonstrates specific aesthetic judgment) will be priced up, because the supply is scarce and the demand from clients-who-can-tell is increasing. The clients in this tier are the ones who can recognize the difference between average-of-style and deviation that earns its place, and they will pay substantial premiums for it because they understand that their own brand depends on visible taste.

Your job, as a working creative in 2026, is to position yourself in the upper tier and develop the specific taste-faculty that justifies the position. This is not a marketing exercise. It is a development exercise. You build the taste by looking, with discipline, at thousands of objects, over years. You position by demonstrating the taste in everything you put out under your name.

If you want to see the upper tier in action, look at the case studies of book-marketing campaigns that have actually worked and notice that the consistent variable across them is editorial intelligence applied at every stage; the cover, the copy, the launch sequencing, the press, the partnerships. None of those choices are AI-generatable, because each requires a human-scale judgment about what this specific book needs in this specific cultural moment for this specific reader. The judgment is the entire product. The judgment is what the publisher is paying for when they hire the team.

The investment to actually develop it

If you want to develop taste deliberately, the practice is roughly:

Read more than you produce. Look more than you publish. Listen more than you broadcast. The ratio matters. Working creatives who consume widely (books, films, exhibitions, gigs, bad museum shows, good museum shows, archival materials, internet culture, weird Reddit threads, specific working artists’ Substacks) develop taste faster than those who don’t. The taste develops in the gap between your input and your output. If your output volume is too high, your taste stagnates. The most-published creatives are not the highest-taste creatives. The taste-rich creatives produce less but produce better choices.

Specifically: spend at least 20% of your working week on inputs that are not directly tied to a deliverable. This is the part nobody schedules. It is also the part that, over a five-year horizon, separates the tasteful working creatives from the high-volume mediocre ones.

Develop a private grading system. When you encounter work in your field, grade it. Not aloud, not on the internet, but in a notebook for your own eyes. Why is this piece good? What did the maker get right? What did they get wrong? What would you have done differently? The grading system, applied consistently for years, becomes your taste. It is your internal IP. Nobody can replicate it, because nobody has been grading the same set of objects with your specific eye for the same length of time.

Find the small audience of people whose taste you trust, and consult them. Almost every working creative I admire has a small group of two or three people they trust to give honest aesthetic feedback. The group is private, slow-moving, and weighted heavily in their decision-making. Not focus groups. Not online comments. Two or three specific people whose taste they have come to respect, and whose feedback shapes the work before it goes public. Build that group. Protect it.

Refuse to produce work below your taste line. This is the hardest one and the most consequential. There will always be a financial reason to take the work that is below your standards. Take less of it. Better yet: design your career so the volume of below-line work decreases over time, and the volume of at-line work increases. The arc of a serious working creative is the slow refusal of work below their own standards, until the only work they produce is at the top of what their taste can reach. That arc is what builds a career that compounds rather than one that flattens.

The closing argument

Taste is the only thing the machines can’t reach, and it is also the only thing in the creative production pipeline that has been radically democratized by the same technological shift that threatens working creatives. The same internet that distributes AI tools also distributes art history, archival material, working-creatives’ notebooks, behind-the-scenes documentaries, criticism, and the entire global record of taste-making across centuries. Anybody with a phone has access to more cultural inputs than the wealthiest collector of 1985 could have assembled.

The democratization of inputs and the automation of outputs together create a strange new condition: the bottleneck in creative work is now the judgment connecting the two. The faculty that processes the inputs and chooses among the outputs. That faculty is taste. That faculty is human. That faculty, as far as I can see, is going to stay scarce, and the working creatives who have it are going to be paid more, not less, in the AI era.

The art-school joke about taste was that it was an unprofitable affectation. The art-school joke is over. Taste is the profitable defensible position now. Develop it carefully. Protect it from market pressure to commoditize. Use it as the primary deliverable nobody else can replicate.

The machines can produce a thousand competent things in an afternoon. They cannot produce a fortieth one that is the right one. That choice is your job, your moat, your business, and increasingly the only thing you have that is genuinely worth selling.

Take it seriously. Almost nobody else is.