Do Brands Dream of Electric Customers?

I have spent years telling clients that a brand is not a logo. It is a system of judgements: what is shown and what is not, what is said and how it is said, what gets rejected even though it works. In other words, a brand is, fundamentally, an aesthetic position sustained over time.

Now I find myself reading a series of articles that force me to think otherwise.

In January 2025, Amber Atherton, an investor at Patron, published a piece titled Toward Computational Taste: LLMs, Aesthetics & Judgment. The thesis is simple and brutal: taste is no longer merely something people express. It is something AI models optimise.

She does not present this as a threat. She presents it as a business opportunity. Which is perhaps the most revealing part.

Taste was always power

Kant argued that taste was a universal judgement. Bourdieu corrected him: taste is social distinction, cultural capital, a way of saying I belong here and you do not. Editors, curators, art directors: they have all exercised that power for centuries. To decide what is beautiful is to decide what has value.

The internet democratised access, but it did not dissolve hierarchy. It merely shifted it towards algorithms. TikTok’s For You feed is not neutral. It is an invisible curator that learns your preferences and returns them to you, exaggerated. The result, as Atherton points out, is fragmentation and echo chambers: each person trapped inside their own taste, fed back to themselves until it curdles into caricature.

What comes next is different. LLMs do not simply amplify existing taste; they can shape it, personalise it and generate it. They can learn not only what you like, but the kind of judgement you exercise. From there, they can produce content that feels not just familiar, but intelligent.

What this means for branding

For decades, the value of the branding consultant has rested on a mix of judgement, experience and cultural sensitivity that is hard to replicate. We know how to read a context, spot an aesthetic inconsistency, propose a direction the client could not yet see. That required years of training and exposure. It was, in the best sense, an art with barriers to entry.

Atherton proposes something she calls Taste-as-a-Service: an API for taste that any platform could plug into in order to generate recommendations ten times more precise than current ones. Later, she goes further still, imagining models trained on specific aesthetic profiles — minimalist, maximalist, avant-garde — that automatically rewrite any content in that style. She calls it “Figma for taste”.

The question I cannot shake is this: if taste becomes computable and scalable, what happens to the value of those who have so far exercised it through their own judgement?

I am not saying that value disappears. I am saying its nature changes. And those who fail to see that in time will realise too late.

But there is something worse: they may stop seeing you altogether

Up to this point, the problem was that AI learns consumer taste and optimises it. But Bain & Company has now published a report that adds a more disturbing layer: in the agentic economy, the consumer may no longer even see your brand. They ask an agent, and the agent decides.

More than 60% of searches are already resolved through AI summaries without the user ever visiting a website. The old goals of SEO — rank highly, win the click, tell your story — are mutating into something Bain calls optimisation for generative engines. The aim is no longer to appear on Google, but to win the agent’s output. When someone asks ChatGPT which brand to buy, your name needs to appear in the answer. And appear properly.

What that means for branding is radical. If a visual identity once had to work on a billboard, a screen and a pack, it now also has to work compressed into a single line of text inside a list generated by AI. No one sees the logo. No one sees the tone either. What survives is whatever the model can summarise with credibility.

Bain puts it with striking coldness: in this new environment, trust stops being a feeling about a brand and becomes an attribute of its data. Not what you project, but what the model can verify about you.

Their historical analogy is the most honest one I have read about this moment. At the beginning of the twentieth century, mass production severed consumers from the origin of products, and branding emerged precisely to rebuild that lost trust. Now AI is severing the consumer from the traditional path to purchase all over again. Branding has to reinvent itself again too.

The difference is that this time the intermediary is not a shopkeeper, a shop window or a recommendation algorithm. It is a system with its own criteria for which brands deserve to be mentioned. And you did not design those criteria.

The danger Atherton names but does not quite develop

The most honest part of Atherton’s article is the warning at the end: the risk of AI is not that it imposes foreign taste on us, but that it freezes us inside our own. Rather than helping us discover what we might come to appreciate, it traps us in a loop of what we have already liked.

For branding, that is particularly serious. A brand that wants to lead culturally cannot be held hostage by the average taste of its audience. It needs tension. Controlled strangeness. The calculated misstep that opens up new ground. That is precisely what optimisation systems do badly: the optimal is not the original.

This is where I think there is still ground left for us. Not in competing with the machine on aesthetic efficiency, but in doing the opposite. Introducing the friction the machine avoids. Proposing what the algorithm would never recommend. Defending judgement against optimisation.

The question I am leaving open

Philip K. Dick spent his life writing about androids who did not know they were androids. Beings that felt, judged and chose with such conviction that they were indistinguishable from humans. The question that obsessed him was not technological. It was philosophical. If something exercises judgement with coherence and consequence, what does it matter what it is made of?

Computational branding places us in the same spot, but with the question reversed. We are no longer asking whether the machine can have taste. We are asking whether we, surrounded by systems that optimise, amplify and endlessly feed back our preferences from the moment we unlock our phones in the morning, still exercise judgement of our own. Or whether we have long been, without noticing, the android in the story.

Are we looking at the end of branding as a discipline of judgement, or at its clearest opportunity in decades?

My instinct is that the brands that survive culturally will be the ones with an aesthetic point of view so sharply defined that no model can flatten it into the average. The ones that remain unmistakable precisely because someone made decisions optimisation never would.

But I am wrong often enough. And I would genuinely like to know what people working in this make of it.

By Dimas Gorostarzu – Director at Harmon