What you deliberately withhold from the machine becomes the most expensive part of your brand.
When the camera arrived, portrait painting should have died. It became expensive instead. A painted portrait today is a luxury precisely because nobody needs one anymore.
Same with watches. A twenty-euro quartz watch keeps time more accurately than any mechanical watch. Yet a mechanical watch costs more today than it did before the quartz crisis. And when music became endlessly copyable as a recording, the money moved to where music can't be copied: the concert.
It's the same pattern every time: When copying gets cheap, value shifts to what can't be copied.
This is happening to brands right now
AI makes the everyday surface of brand work cheap. Applying the logo, writing in the right tone, decent design for every channel: machines handle that now, in usable quality, for everyone. The kebab shop around the corner has a good menu these days. The average is rising.
In almost every discussion right now I hear the question: What can't the machine do? Today it gets asked defensively. It's meant to establish where the machine stays out and why your own job is safe. An exclusion question.
That will flip. The same question becomes a pricing question: What do we deliberately not give the machine, and what exactly is that worth? The line of defense becomes a price list.
But: Build the price list straight from that question and you walk into a trap.
The trap of "can't"
"What the machine can't do" is a list with an expiry date. Generative image models still don't reliably produce a defined brand gradient the same way twice, because every run starts from fresh random noise. And even a single color value often lands wide of the mark. A color benchmark from October 2025 gave nine image models the target color as a hex or RGB value across 11,500 prompts, and six of them hit it in fewer than ten percent of cases. What it tested was one specified color on one object, and the paper has not been peer-reviewed. Conclude from that that the gradient is now the human part of your brand, and the next model update leaves you without an argument. The machine's capabilities grow every quarter. Value built on its inability shrinks at the same rate.
And the old examples show something more precise. The quartz watch keeps better time. The camera portrays faster and cheaper. So the value of the mechanical watch and the painted portrait never sat in the machine's inability.
Where the value actually comes from
The value comes from a decision you can see. Someone made it by hand although the machine was available. And someone puts their name to it. The painted portrait is valuable because a human spent lifetime on it and signed it. The signature is half the price.
This already works today, in the middle of the AI era: Klarna moved customer service to AI and is now bringing humans back for part of it. "Talking to a human" is a promise the Klarna CEO now makes publicly. A human on the phone is a company decision there, and it visibly costs money.
This decision needs two things to carry value. It has to be made before anyone asks. And it has to be verifiable: Who was it, who stands behind it? Handmade without proof is indistinguishable from the machine's product. And what can't be distinguished won't command a different price for long.
The two lists
In practice, that means a brand needs two lists.
The first list holds everything the machine may deliver. Applications, formats, variants, the daily surface. This list should be long, and what's on it may become cheap. That's not a loss. Only when watches cost twenty euros did the mechanics itself become a status symbol.
The second list holds what you deliberately withhold from the machine. Not much, or it isn't a selection. Maybe the visual world. Maybe the one text the founder writes herself. Maybe the customer contact in the moment that matters. At Studio Ghibli, Hayao Miyazaki still draws his films by hand. His "The Boy and the Heron" from 2023 is billed by its distributor as a hand-drawn film, even though colouring and compositing at the studio have long since moved to the computer. When AI tools flooded the internet with Ghibli-style images in 2025, the original didn't get any cheaper. The style can be copied; the decision behind it can't.
The list goes beyond design. When The Cure toured North America in 2023, the band opted out of Ticketmaster's demand-based pricing, by then common practice in the concert business. The band wanted to decide for itself what an evening costs. Robert Smith called the model a scam and rejected the notion that artists have no say in what their own tour costs. Smith stood for it publicly. A year later, Rolling Stone called his approach the gold standard for how artists can keep concert tickets affordable. What the refusal cost, Billboard put a number on: the average ticket sold for $68.54, 37 percent below the average price of the top 50 tours in Billboard's midyear 2023 recap. The tour worked out anyway. At $37.5 million from 547,000 tickets across 35 shows, it was the most successful North American tour in the band's history, more than double the $18 million of 2016. Its value comes from the band demonstrably leaving the markup on the table, with a name answering for it.
And sometimes the second list is the entire business model. In 2026, Aldi publicly declared it would forgo app discounts and loyalty programs: one price for everyone, while others collect purchase data and personalize offers. Forgoing the data machine is the brand there. And because Aldi says it out loud, a habit becomes a decision.
Every item on the second list needs two entries: an address that stands behind it, and proof that it actually works that way. The address can be a name, like Miyazaki, or a company that declares it publicly, like Aldi.
Start with a single list: everything that makes up your brand, from the logo to the customer conversation. Then test what the machine can reliably do with it today. That shows you where it struggles. But the sorting follows a different criterion: The second list gets what becomes more valuable through a human, regardless of whether the machine could do it. Build the second list from the machine's weaknesses and you're back to the list with the expiry date.
The sorting leaves a remainder. The gradient that isn't important enough for anyone to have a human maintain it, and that the machine doesn't reliably render the same way twice anyway: it goes. The gradient becomes a fixed color value, and the next brand design doesn't define one in the first place. That, too, is a decision nobody had to make before.
The only dangerous thing is what lands on no list and doesn't get cut either. The unresolved, the "we'll see": the gradient that's somehow supposed to go through the machine and looks different every time. What remains is inconsistency, and it costs without earning anything.
From the end to the beginning
This reverses the familiar order of brand work. A brand used to be designed, approved, and poured into guidelines, and at the very end came the application. The machine was the last step in the chain. Now it's the first. You test what it reliably delivers before you define the base elements. And you check how it reads your brand today before you decide how the brand should be read tomorrow. A rebranding will start with this inventory, in design as in strategy.
Not consolation, a sorting
An honest postscript. Portrait painting becoming a luxury didn't help most portrait painters. Anyone who made a living from commissioned portraits lost their customers to the camera. A small top inherited the premium. The pattern is a sorting, and it's already running. Which is why the inventory is worth doing now, while you can still choose what goes on the second list, instead of explaining later why nothing is on it.
The price premium doesn't sit on what the machine can't do. It sits on what you decided not to give it.
Terms from this text: To the glossary →