Every Interface Has an Accent

On generative interfaces, cross-cultural design, and why AI that builds the screen might quietly export one region’s mind to the whole world

By Lucrezia Spapperi Gestri. Brand Communication and Marketing Digital Strategist. Europe and Asia-Pacific.

Reading time: around 11 minutes

The first time I really felt it, I was somewhere in China with a phone in my hand, about to back out of something, and instead I confirmed it.

The confirm button was sitting where I expected cancel. The two were swapped. My hand had committed to the exact thing I meant to escape, because my hand was following a layout from another country, one that lived in my muscles and not on this screen.

That was the small version. The larger version arrived every time I opened one of the WeChat mini programs that everyone around me used without a second thought. To them it was effortless. To me it was a maze. Layers inside layers, actions where I did not expect them, a density and a logic that plainly worked beautifully for hundreds of millions of people and left me lost on the very first screen. Nothing was translated wrong. Every single word made sense. And I still could not find my way.

Because the thing that was foreign to me was never the language. It was the system. The whole grammar of how you were supposed to move through it.

That experience is the reason I have never been able to take the industry’s favorite word at face value. Because for most of the industry, “localizing” a product still means translating the words. You take the interface, send the strings to a translator, swap the copy, maybe flip the layout for a right to left language, and everyone agrees the thing is ready for the new market. Job done.

That was never how it looked to me. Not out of some special conviction, just out of plain good sense. If you do this work with any strategy at all, and especially if you have ever been the confused foreigner lost on the first screen, you know that localization is not a translation problem. The words are the easy part, the visible part, the part everyone can point at and feel productive about. The real work is underneath, in the system. The rhythm of the thing. How much sits on the screen at once, where the eye is meant to go, whether the design trusts you to explore or takes you by the hand, whether empty space reads as elegance or as something missing.

That is the part that actually decides whether an interface feels foreign or feels like home. And it is the part the briefs kept leaving out.

The industry translated the words and left the system in its original language. And it mostly got away with it, because the system doesn’t announce itself. It just feels normal to the people it was built for and slightly off to everyone else, and because there was usually a human team in each market quietly fixing the layout even when the brief only said “translate the copy.”

I’ve been thinking about this a lot lately, because of what is happening to the interface itself. It’s dissolving. And I think that dissolving is about to take that quiet human fix away, and make a problem I have been pointing at for years impossible to ignore.

The interface is becoming a liquid

Let me explain what I mean by dissolving, because it sounds dramatic and it’s actually just where the design world is going in 2026.

For thirty years the interface was a stable object. Someone designed a screen, and that screen was more or less the same for everyone who opened it. You could argue about the buttons. But the buttons were there, in a fixed place, for all of us.

That is ending.

The new idea is called generative UI, and the shift is bigger than it sounds. Instead of a designer building one screen with some variations, an AI generates the interface on the fly, in real time, differently for each person and each moment. Not a template with the words swapped in. A genuinely different layout, a different structure, a different set of interactions, assembled right when you need it and gone after. People in frontend are calling it the biggest change since we moved from jQuery to React, which if you’re not technical just means “the ground is moving under the whole practice.”

Around it there are two more things happening. One is agentic UX, where you stop navigating a product and instead hand a task to an agent that does it for you, so the design goal quietly changes from “keep them engaged” to “get them out of here fast.” The other is what people call zero UI, the interface that disappears entirely, voice and context and prediction doing the work, no screen at all. Some analysts think invisible interactions like that will be most of how we deal with software within a couple of years.

So put those together. The screen is becoming fluid, personal, generated, and sometimes absent.

And here is the question that keeps me up. If the interface is generated fresh every time by a model, then somebody’s assumptions are doing the generating. Whose?

First, a smaller worry: what happens to brand

Before I get to culture, there’s a nearer problem that anyone in branding should sit with, and it’s this. A brand used to have a body. It had a screen you designed, a space you controlled, a set of moments that were yours. The color, the type, the little interaction that felt like you and no one else.

When the interface becomes a liquid that an agent pours differently every time, a lot of that body goes away. If a user never really visits your product, if an agent fetches what they need and hands it over in its own neutral wrapper, then where does your brand live? Not on the screen. There often isn’t one.

This is the same worry I keep circling in everything I write now. The more the machine mediates, the more it strips away the sensory, designed, human layer where affection actually forms. The interface dissolving is just that worry arriving in the one place designers thought was safe, the layout itself.

I don’t have a clean answer. I think brand has to retreat to the few places that stay solid, the physical product, the human moment, the voice and meaning that survive being poured through a neutral container. But I’ll be honest, I’m still working this one out. It’s new for all of us.

Now the bigger one: the system carries a culture

Here’s the part I actually wanted to write about.

When we say an AI generates the interface, we skip over a simple fact. The model learned to do that from somewhere. It learned from data, and most of the data, most of the design patterns, most of the “this is what a good interface looks like” examples, come from a particular part of the world. They carry that part of the world’s mind inside them.

And interface preferences are not universal. This is the thing people who haven’t worked across regions tend to not believe until they see it.

There’s a deep body of research on this, and it’s worth knowing because it turns a hunch into something much closer to fact. Decades ago the psychologist Richard Nisbett and his colleagues began documenting what he later called, in a book by that name, the geography of thought. The short version is that cognition is not the same everywhere. Broadly, thinking shaped by the ancient Greek inheritance tends to be analytic. It isolates the object, the single focal thing, and reasons about it on its own terms. Thinking shaped by East Asian traditions tends to be more holistic. It attends to the field, the context, the web of relationships a thing sits inside.

And this is not a soft cultural stereotype you can wave away. It shows up in the body, below the level of choice. In eye tracking studies of people looking at the same scene, Western viewers tend to lock onto the focal object quickly and dwell there, while East Asian viewers move their eyes more across the background and the relations between elements. In change blindness experiments, where something in an image quietly shifts, Westerners are faster to catch a change in the main object, and East Asian viewers faster to catch a change in the setting around it. The attention itself, the raw path an eye takes across a screen, is culturally shaped. We are, quite literally, not all looking the same way.

Now hold that next to interface design, and a lot of things that looked like mere taste turn out to be cognition. It is why Weibo can look so much denser than Facebook and still work beautifully for the people it was built for, while a Western designer’s gut screams “too much.” Density is not clutter if your attention was trained to read the whole field at once. It is why so much East Asian navigation runs vertical and grid shaped while Western navigation runs horizontal. It is why white space, which in the West signals confidence and premium, can read in other markets as emptiness, as a page that forgot to bring its content. Same pixels. Different nervous systems reading them.

None of that lives in the words. You can translate every string on Weibo into flawless English and it will still feel like a fundamentally different kind of object than Facebook, because the difference was never linguistic. It was in the system. The density, the hierarchy, the relationships between elements, the built in assumptions about how an eye moves and what it expects to find.

This, I understood only much later, was the research sitting underneath my confusion in China. What felt like a maze to me was not badly made. It was a system built for a kind of attention that was not mine, and my Western eye, trained to hunt for one focal object and a lot of breathing room, simply did not know how to move through it. The maze was in me, not the app.

That is the thing the industry keeps forgetting to translate. For years it mostly got away with it, because there was usually a human team in each region quietly fixing the layout even when the brief only said “translate the copy.”

But a generative interface does not have that human team in the loop. It has a model. And the model generates from its defaults.

This was never a hunch. It’s a research tradition.

The strange part is that design already knew all of this, and then mostly filed it away under “nice to have.”

Back in the 1990s and 2000s, people like Aaron Marcus took Geert Hofstede’s work on cultural dimensions, things like how much a culture defers to hierarchy, how individualist or collectivist it is, how much it needs certainty and structure, and mapped them directly onto interface design. Marcus and his colleagues showed, concretely, how power distance changes the way hierarchy and access should be expressed on a screen, how individualism versus collectivism changes what content belongs at the center, how a culture’s tolerance for ambiguity changes how much structure an interface needs before it feels safe rather than sloppy. Cross cultural human computer interaction has been a real, serious field for decades. The knowledge exists. It has existed for a long time.

It just rarely survived contact with a deadline. In practice, “make it work for that market” collapsed back into “translate the copy,” because that was the cheap, legible, schedulable version of the task. The deep version, the one that treats the interface as a cultural system, was almost always the first thing cut.

Which is exactly why this moment unsettles me. We are handing interface generation over to machines at the precise instant when the industry has spent thirty years proving that it will skip the cultural layer whenever it is allowed to.

The quiet export of one region’s mind

So here is what I think is coming, and why it worries me more than the usual “AI is biased” conversation.

Start with the models themselves, because this is the part that genuinely stopped me. In 2025 researchers looked at how vision language models, the systems that read and describe images, actually attend to a scene, and they found something quietly astonishing. The same kind of model behaved more holistically when it worked in Japanese, attending to context and to the relationships between things, and more analytically when it worked in English, locking onto the single focal object. In other words, the exact cultural difference Nisbett measured in human eyes reappeared inside the machine, switched on and off by the language it happened to be using. These systems have not just absorbed our words. They have absorbed our ways of seeing, and they carry them.

Sit with that for a second, because it dismantles the comfortable assumption underneath most localization work. We tend to imagine AI as neutral, culture free, a blank engine that we then dress in local clothes. It is not. It already has cultural cognition baked into it, inherited from whoever made its data. The question was never whether an AI has a cultural default. It is which one, and what happens when that default is handed the job of generating the screens for everybody.

And there are already fingerprints at the interface level too. A 2025 comparison of AI products found that Western ones, like ChatGPT, lean toward a transparency focused design, showing you the reasoning, the sources, the mechanics. East Asian ones, like DeepSeek, lean more relational. Same category of product. Different implicit idea of what the user is and what they want. The values are baked into the interaction, not printed in the text.

Now imagine that baked in signature, running through a generative UI engine, producing interfaces for the entire world, at a scale and speed no human design team could ever review.

If that engine learned mostly from Western patterns, it will generate Western interaction logic everywhere. Not because anyone decided that. Because that was the default in the training data, and defaults are invisible, and invisible things don’t get argued about. A user in Jakarta or Seoul or São Paulo gets a screen assembled from a mind that isn’t theirs, and it will feel slightly off, and they won’t be able to say why, and there will be no local designer in the loop to fix it, because the whole point of generative UI was to remove that step.

This is the same homogenization I wrote about before, the flattening of everything toward one average. But it has moved. It used to be about content, the words and images all starting to sound the same. Now it’s moving down into the structure of interaction itself. Into how the screen thinks. That’s deeper, and it’s harder to see, and it matters more, because the structure shapes the thought.

There’s even research now with a title I can’t stop repeating, about how AI interface design can engineer trust and create vulnerability at the same time. How the design of the thing decides whether you stay critical or hand over too much. If that design is generated from one culture’s assumptions and shipped to all the others, we are not just exporting a look. We are exporting a way of relating to the machine. And we’re doing it silently.

Why the designer’s eye is exactly the wrong thing to automate here

I keep coming back to something Riccardo Falcinelli writes about, in a couple of books that shaped how I see this. In Guardare, pensare, progettare, he ties the actual neurons of the retina to the act of designing, how much of what we call good design is your nervous system doing things you never consciously chose. And in Cromorama, his book on color, he shows that even the way we see, something that feels as natural and given as which colors go together, has been trained into us by centuries of images and culture. Our gaze is not neutral. It was built, slowly, by where we come from.

Which means a designer’s eye is a cultural instrument. Mine is European, shaped by the images I grew up inside. A designer in Shanghai has a different one, just as trained, just as sure of itself, pointing at different things. Neither is the correct one. That’s the whole point.

When you automate interface generation, you don’t remove the cultural eye. You just freeze one particular eye into the system and call it default. And then you scale it. The machine doesn’t have a gaze of its own. It has the averaged gaze of whoever made its training data, wearing the mask of neutrality.

This is why I don’t think “add more languages” fixes it. The words were never the problem. The problem is that a system built from one culture’s habits of seeing is being asked to design for all of them, and it will quietly assume everyone’s eye works like the eye it learned from.

So what do we actually do

I don’t want to just ring an alarm, because alarms are easy and useless. Let me try to say what I think the work is now.

Localization has to move down a layer. Stop treating it as a translation task and start treating it as a systems task. The question is no longer “what does this say in Japanese,” it’s “does the density, the hierarchy, the rhythm, the amount of guidance, the use of space, match how this market actually reads a screen.” That’s a harder question and it needs local people who can feel the answer, not just render the strings.

Someone has to stay in the loop on purpose. Generative UI wants to remove the human design step. In your home market you might get away with that. Across cultures you can’t, not yet, maybe not ever. The human who knows the local system has to be put back into the process deliberately, precisely at the point the technology wants to remove them. That’s a choice, and it costs money, and I think it’s the difference between a product that travels and one that quietly alienates half the planet.

And feed the machine something other than the default. If your generative system only ever saw one region’s patterns, it will only ever generate one region’s screens. The brands that get this right will be the ones that treat local interaction patterns as training material worth protecting and providing, not as a nuisance to smooth away. The variety is the asset. The average is the risk.

The thing under all of it

Here’s where I land, for now.

The industry spent years treating localization as a words problem, because words are the part you can see. The system underneath, the culture built into how a screen thinks, got left untranslated because it could be gotten away with, because there was always a human in each market quietly correcting for it.

That human is exactly who generative AI is designed to remove. And the moment they’re gone, the untranslated system underneath gets shipped to everyone, at scale, wearing a face of neutrality it does not have.

I don’t think the answer is to refuse the tools. The tools are extraordinary and I use them every day. I think the answer is to finally take seriously the thing we always half knew and never quite acted on. That an interface is not a container for words. It’s a way of seeing, made by people, for people who see the same way. And when a machine starts making those interfaces for the whole world, someone with a trained eye and a conscience had better still be in the room, asking whose way of seeing just got treated as the default.

Because it’s going to be somebody’s. It always is.

I’d really like to hear from people who’ve worked across very different markets on this. Where have you felt a system, not just a translation, fail to travel? Tell me in the comments. I’m still forming my view and I’d rather form it with you.

About the author

Lucrezia Spapperi Gestri is a Brand Communication and Marketing Digital Strategist working across Europe and Asia-Pacific. She writes about brand positioning, AIGC-driven innovation, cross-cultural design, and the messy places where meaning, culture, and machines meet. Portfolio: lucreziasgdesign.github.io/portfolio. LinkedIn: linkedin.com/in/lucrezia-gestri-spapperi.

A few things I read while thinking about this

• On the shift in interfaces: generative UI and agentic UX trends for 2026 and adaptive, agentic, ambient UI
• On culture in the screen: cross-cultural UX, East and West and global-local UX strategy
• On how differently we look: Richard Nisbett, The Geography of Thought, and cultural variation in eye movements during scene perception (PNAS), plus Masuda and Nisbett on culture and change blindness
• The one that stopped me: Contrasting Cognitive Styles in Vision-Language Models, holistic attention in Japanese versus analytical focus in English (2025)
• The older design tradition: Marcus and Gould, Crosscurrents: Cultural Dimensions and Global Web User-Interface Design, built on Geert Hofstede’s cultural dimensions
• On AI localization and bias: cross-cultural UX research in a global AI era
• On interface and trust: Engineering Trust, Creating Vulnerability
• Riccardo Falcinelli, *Guardare, pensare, progettare. Neuroscienze per il design and Cromorama. Come il colore ha cambiato il nostro sguardo* (Einaudi)

💡 Stay inspired every day with Muzli!

Follow us for a daily stream of design, creativity, and innovation. ***Linkedin | [Instagram](https://www.instagram.com/usemuzli/) | [Twitter](https://x.com/usemuzli)***