The Taste Test

My mum is a textile artist. Extraordinary, in the specific way that people who've spent a lifetime developing a singular craft become extraordinary. My kids create too, in art, in music and in performance with the ease of people for whom it is simply the natural response to being alive. They can create, it makes them happy, they improve the world a bit, and they can tell.

I can only tell.

I have spent thirty years in an industry built on creative work, sitting in rooms with some of the most gifted people in the world at making things. I can spot the real thing the moment it arrives. I can inspire it in others, judge it against a brief, protect it from the people who want to sand it down. But the act of genesis, that particular collision of anxiety, self-doubt, and horsepower that produces something that didn't exist before, I have never been able to do it myself.

Which makes me exactly the kind of person who should be relieved that AI can now fill that gap as I put my petty jealousies aside.

I'm not. Here's why.

Last week I wrote about the fastest horses and the gap between companies cutting for AI's potential versus its performance, and the industry's error of treating efficiency as the destination rather than the entrance. The automobile question was where the value was going. Today I want to be specific about one part of that answer. Why creativity remains irreducibly human. And why the case is stronger and more precise than most people have articulated.

Three reasons. Each distinct. Each matters.

AI is extraordinary at patterns. This is not a criticism, it is the entire architecture of the technology. Large language models, image generators, code systems, all of them are, at their core, pattern recognition systems trained on human-generated material at unprecedented scale. Exceptionally good at predicting what comes next, given what came before.

Standout creative work does not confirm what comes next. It disrupts it. The campaigns that still get referenced, the work that shifted a category, the creative ideas that actually entered culture rather than passing through it, none of them arrived by finding the most likely next step. They arrived because someone looked at what the category expected and went somewhere else. Because the departure from the dominant pattern happened at the right cultural moment, with enough internal coherence that it felt inevitable in retrospect.

AI's version of surprising is the statistical average of surprise. Its version of unexpected is precisely what an audience has most often encountered before. Ask it to break a mould and it produces the most common version of mould-breaking. It optimises for the central tendency of its training data, which means it cannot generate the genuinely discontinuous idea. Only the next plausible step on an existing trajectory.

There is something further that the industry hasn't fully absorbed. The best creative disruption requires cultural timing or a felt sense of what this specific moment is ready for, often before the moment itself knows. That requires living inside the current culture, not training on its past. AI trained on yesterday's creative output cannot feel today's pressure building. It can describe the last zeitgeist with precision. It cannot anticipate the one that hasn't happened yet. That intuition is not in the training data. It is in the human who is alive right now, in this specific year, carrying the particular weight of this particular moment.

I want to be precise here, because this is usually said loosely.

Taste is not the ability to run ten thousand creative variants and identify which one generated the highest click-through rate. AI can do that. It is getting better at it. That is not taste.

Taste is the judgment that precedes the test. The capacity to look at two options, before any data exists about them, and know which one is right. Not which one will win the test. Which one is worth making. Which one has something to say.

This is the form my contribution has always taken. I watched enough creative work developed, rejected, approved, and later regretted to understand something clearly: The best creative directors I worked alongside had an almost physical response to work that was right, and an equally physical response to work that was technically fine, technically competitive, but hollow at its core. No click-through rate could tell you which of those two things you were looking at. Only their taste could.

This is not mysticism. It is expertise built over years of exposure to what moves people, what ages well, what connects with something real in human experience and what merely simulates the form of connection without achieving it. AI can produce the form. It cannot evaluate whether it has achieved the substance. And it cannot tell the difference between the two, because telling the difference requires having actually cared about it, across years of living with people for whom it mattered.

When modern AI image generators produce an image, they don't work the way you experience seeing. They are not rendering a scene.

They begin from noise. Literally from random statistical distribution and then iteratively refine it toward a visual pattern that matches learned associations between images and text from their training data. They are finding the shape in the noise that statistically fits the description.

The model has never seen light hit a surface. It has never experienced depth, weight, shadow, or the way colour changes across a day. It operates in a learned mathematical space where these things exist as relationships between numbers. Encoded, not experienced. It produces images that look like things because it has learned what things are supposed to look like in images. Not because it understands what those things are in the world.

It is working in a flat space. Not even a space but a set of learned correlations. Without physics. Without gravity. Without air. Without the embodied experience of what it is to exist in three dimensions and move through them.

When people call the outputs magical, they are recognising something genuine as the technical execution is often extraordinary. But the system has no access to the experience that makes magic feel like magic to a human being. It has learned what magic looks like. That is not the same thing.

Will embodied AI change this, with robots that see, move, sense, and accumulate physical experience of the world? Almost certainly, in time. But that is not what people are calling magic today. Understand what you are working with. And keep being magical yourself.

This is the hardest to articulate and the most important.

The best creative work does not just disrupt a pattern or demonstrate taste. It says something. It takes a position on the human condition. It makes a claim about what matters, what is true, what deserves attention, what we should feel. The campaign that names a truth the category has been avoiding. The brand idea built on an actual belief about how the world should be.

AI can produce the rhetoric of conviction perfectly. The sentence structures, the declarative confidence, the tone of someone who holds a view. It is fluent in the language of belief. It does not believe anything.

Conviction requires having something to lose. It requires that you actually hold the position and that it costs you something to state it and that you would stand by it when challenged, that it reflects something concluded from living rather than computed from training data. AI has no stakes. It has no beliefs. It has learned what believing sounds like, and it produces that sound with remarkable accuracy.

This matters because the most commercially powerful creative work is almost always anchored in genuine conviction: a brand that actually stands for something, expressed through work that communicates it with confidence. That conviction has to originate somewhere human. If it doesn't exist upstream, no amount of executional fluency downstream will manufacture it.

AI will continue accelerating creative execution. Variants at scale, optimisation at speed, technical production at a fraction of previous cost. These are genuine capabilities and the right places to deploy them.

What AI will not do is originate the creative idea worth having. It cannot sense the pattern worth breaking. It cannot exercise taste before data exists. It cannot draw on the embodied experience of what this human moment requires. It cannot hold a genuine conviction about what the work should say.

The danger is a specific confusion: mistaking AI's ability to produce creative work quickly for AI's ability to produce creative work that matters. The first is unambiguously true. The second depends entirely on the human judgment upstream.

I said at the start that I am the one in my family who can tell.

I have come to understand that this is not the consolation prize.

My mother has spent a lifetime developing the judgment to know when a piece of work has achieved what it was reaching for. My children built it from making things, failing, trying differently, and accumulating a felt sense of what right looks like. That process, slow, effortful, rooted in the full experience of being human, is precisely what produces the taste, the conviction, and the cultural timing that the technology cannot simulate.

What AI liberates, for those willing to use the liberation correctly, is more time for exactly that. More attention upstream. More space for the question of what is worth making, not just what can be produced.

The people who will matter most in the creative economy over the next decade are not the ones who can generate the most. They are the ones who can tell.

Brandflow is written by Justin Billingsley, who has spent his career on all three sides of the industry's table: senior client, global agency leader, technology founder. First published 23 March 2026 in the Brandflow newsletter on LinkedIn.