Stripe design leader Katie Dill warns that in an era when AI has made building products extremely easy, quickly made output risks degenerating into uniform, soulless "zombie UI." Her core argument is simple: don't use AI only as a tool for building faster; use it as a tool that scales a clear point of view, strict quality standards, relentless editing, and creative experimentation.
Ultimately, a good product is judged not by "was it made with AI?" but by whether it carefully solves the user's problem and carries the brand's own personality.

1. Parallels Between the Postwar Building Boom and the AI Product Boom

Katie Dill begins with the enormous building boom after World War II. In the United States of that time, where returning soldiers, new construction technologies, and explosive demand for housing and buildings came together, buildings went up at astonishing speed. Architects were influenced by the modernism of the 1920s, a style characterized by simple geometric forms, clean unornamented surfaces, a limited color palette, and the removal of unnecessary decoration.

But she says she isn't trying to criticize modernism itself. Early modernism had a clear philosophy and purpose. Le Corbusier's Villa Savoye and the Bauhaus school embodied principles that went beyond mere appearance. The problem was that amid the urgent postwar construction demand, the style began to be consumed as a merely reproducible style.

"The people who first created this movement had a clear purpose. But after the war, under pressure to build fast, the style was copied over and over without intent."

The result was generic buildings that didn't fit their context. Bank buildings originally looked like they conveyed trust, stability, and security, but over time they turned into look-alike commercial buildings that could be anywhere. Katie calls these "zombie buildings": buildings that show no purpose, no personality, and no care for their users or surroundings.

"There are zombie buildings everywhere. Nothing distinguishes them from anything else, and nothing tells you 'this was made intentionally' or 'this design fits this purpose.'"

She believes we are now in the middle of another building boom: the AI building boom. A team of three can now do what used to take 30 people. This leap in productivity is exciting, but it also carries the risk of repeating the failures left behind by the postwar building boom.

2. Three Risks of the "Zombie UI" AI Can Produce

Katie points to three problems to be especially wary of when building products in the AI era.

First, large language models are good at producing the most plausible, highest-probability answer. That means they are good at reproducing what was popular in the past and is still common, but may be weak at producing original solutions that fit a specific brand and user context.

As an example, she shows a design that at first glance looks like an ordinary software company's website. In reality, it was the website of a Korean barbecue restaurant. The design's polish isn't bad, but it conveys almost nothing of the restaurant's atmosphere, the food experience, or what visitors expect.

"Guess what this brand sells. Software? No. It's a Korean barbecue restaurant."

Second is the temptation of output that looks "finished" far too quickly. Katie uses the microwave burrito she often eats as an analogy. Put a frozen burrito in the microwave and you have a meal in 90 seconds. Not because it is tasty or healthy enough, but because it solves the problem so quickly, people end up generously overlooking the quality issues.

"Honestly, sometimes it's barely edible. And yet I'm captivated by the speed."

Building UI with AI is similar. Type a sentence and press a button, and a plausible interface with rounded corners and drop shadows appears instantly. On the surface it looks like a finished product, but whether it actually solves the problem, differentiates the product, or means anything to users are separate questions.

"A glossy exterior often fools us. Does it really solve the problem? Does it really add differentiation?"

Third is the problem of output being treated as disposable as the cost and time of making things drop so low. Because it was easy to make, it is easy to throw away, and people don't think enough about who will maintain it or what long-term impact it will have. When these three combine, the world could be flooded with zombie UI that lacks context and personality and isn't maintained.

Katie notes that we spend half our waking hours looking at screens. That's why people want not just software that merely functions, but products that feel like someone made them with care, for them.

"We want software that feels like someone genuinely cared."

She gives examples: Grok's robot character with its small bits of liveliness and movement, the tiny detail of a calendar tab showing the actual date, and a wallet experience that anticipated the problems an agent would hit during an online purchase and built in tools to solve them. These elements aren't just pretty decoration; they are signals that the makers understood the user's context and built care and consideration into the product.

3. Principle One: Establish a Clear Point of View Before AI

The first piece of advice for building products with soul using AI is "have your own point of view." If you don't have one, AI will fill in a point of view for you, and the result will tend to be a generic, backward-looking answer.

"If you don't have a point of view, AI will provide one. And as we've already discussed, it's likely to be a generic, backward-looking one."

Point of view here means who the brand is, what it wants to be for its users, and what it values. This becomes the foundation of a product's quality bar. Especially in the AI era, when building is distributed across many teams, tools, and agents, without shared standards everyone may produce output heading in different directions.

Stripe names optimism as one of its important values. This value doesn't apply only to big brand campaigns. It is reflected in every decision, large and small: color choices, the tone of sentences, what topics they speak about, and what products they launch to help founders.

"At Stripe we take optimism very seriously. So we put it into every detail, big or small."

Katie shares the example of a design review of a Stripe ad. The ad, which used the brand's signature parallelogram element and customer imagery, seemed good enough at first. But the team felt that a few details weren't quite right. A partner from another department who wanted to ship quickly asked an important question.

"What's the quality bar for AI-generated output?"

Katie sees an important insight in the question itself. Users look at whether the output is good, not whether AI or a human made it. So the standard must be the final experience, not the method of production.

"Users don't care how it was made. What matters is whether it's good. And that is the foundation of our bar."

At the end of this discussion, the team found a full 17 things to improve in the ad: subtle fixes like making corners a bit softer and making speech bubbles slightly smaller. Inside Stripe, this kind of fine polishing is reportedly referred to as a verb: "Pepsi popping."

"We're not chasing perfection. We're trying to go one level deeper than what the customer will see."

Even if users don't consciously notice every change, this kind of refined finishing clearly shows up in the trust and polish the product conveys. To get there, you shouldn't just listen to what users say, but observe what they actually need and want. You also need to cultivate your own taste and judgment by looking across many fields—the products around you, art, science—at what is ordinary and what is exceptional.

4. Principle Two: Encode "Intent," Not "Consistency," in Your Design System

The second piece of advice is to encode your standards into the machine. Going forward, people can't personally take part in every decision. An environment is already arriving where UI is created without a designer in the room, AI agents find and fix problems while people sleep, and generative interfaces are created in real time for each user.

That's why the role of the design system becomes important again. But Katie says the design systems of the past differ from the systems we need now. In the past, designers were always on the team, so people could fill in the gaps that weren't written in the docs. Now, because distributed building and automated generation must be possible, decision-making criteria need to be far more explicit.

"The target of design is no longer the screen. It's the system itself."

She uses Gutenberg's printing press as an analogy. Many people know Gutenberg invented movable type, but few know about the detailed typesetting system behind it. Gutenberg didn't just make uppercase and lowercase letters, but designed about 290 unique pieces of type, including letters of varying widths, abbreviations, and ligatures connecting letters. The goal was to avoid ugly gaps when lines of text were justified.

"Gutenberg built a system scalable enough, yet specific enough, that something made by machine would feel hand-made in quality."

This is what design systems in the AI era should aim for. If past systems aimed at scaling consistency, new systems must aim at scaling intent.

Stripe says it first built an MCP model that understood its design documentation, but it wasn't as accurate as hoped. When three people entered the same prompt, they got three different results. So afterward they built a CLI (command-line interface) based on their design system. It delivers the needed documentation at the right moment, where developers are working, and makes AI better at following the established frameworks and principles.

The important change is that this system doesn't deal only with individual components like buttons or colors. It includes templates and entire workflows, so that it understands how Stripe's products work and how their elements connect together.

"If earlier design systems scaled consistency, new systems need to scale intent."

But a system is only a starting point. Katie quotes the architect Christopher Alexander.

"A system can satisfy all the rules and still be dead."

In other words, following the rules well doesn't automatically make a product alive.

5. Principle Three: Don't Confuse "Built" with "Good"—Edit

The third piece of advice is don't confuse done with quality. In the past, because resources and time were scarce, there was a natural filter in the development process itself. Even with 20 ideas, you had to pick one, and you developed that one through constant testing and review.

Now, by contrast, you can build all 20 ideas in a week. This is great in that it lets you quickly get your hands on real output instead of endlessly debating hypothetical products in a meeting room. But the filter that used to operate before building now has to operate after building. And rejecting something already built is much harder.

"We can now build 20 ideas in a week. But the filter that existed throughout the process now has to operate after we build, and at that point saying 'no' is much harder."

The person who plays the decisive role here is the editor. What matters is who in the organization looks at the product from start to finish and judges whether all the elements form one cohesive experience. Katie stresses that organizations must abandon the idea that "built" means "finished," and that "finished" means "good."

"We have to let go of the idea that 'built' means 'done,' and that 'done' means 'good.'"

A good editor experiences the product like a user. They check not only whether each individual feature looks good, but whether it flows naturally within the user's whole journey.

• Does this product really solve the user's problem?
• Does it actually match the way users think?
• Even if individual elements are good, is the overall flow awkwardly disjointed?
• Rather than stopping at simply approving or rejecting the output, how can it be pushed toward a more complete state?

Quoting the writer Nabil Kureishi, Katie describes two characteristics of great art. One is small, unexpected details that make you marvel, "They even thought of that?"; the other is a deeper meaning and recurring themes that tie the whole work together.

"Good art has small surprises that make you say, 'Wow, they thought of that?', and a deeper meaning beneath the surface that ties the whole work together."

This is exactly where AI can be weak. AI output often looks as if every detail was placed by chance. For example, the cup is green, but it feels like it wouldn't have mattered at all if it were blue. An editor, by contrast, takes responsibility for every choice.

"What's especially grating about bad AI output is that the details don't seem to matter. The cup is green, but it feels like it could just as well have been blue."

The Liveliness Created by 56 Iterations

Katie cites the Stripe team that made the opening animation for this event. The team first built a 3D model of the scene and used AI to generate several animations. Even the first attempt was interesting and cute, but on closer inspection the motion was somewhat stiff and awkward. It also left you wishing you could see more of the scene.

So the designer made it again, and again. After repeating it 56 times, they arrived at a much more natural, living result.

"We did it one more time. And again. And again—56 times."

The final result may look like a very small difference. But viewers feel the care inside it. At the same time, it's important that without AI it would have been hard to build such a complex scene or experiment with so many perspectives. AI isn't a tool where you have to stop at the first output; it's a tool that helps people explore a broader range of possibilities.

"AI opened up the space of possibilities."

6. Principle Four: Unleash Creativity and Artistry, Not Just Efficiency

The last piece of advice, and Katie's favorite, is to unleash creativity and artistry. AI could be used to mass-produce predictable, repetitive output. But at the same time, it has the potential to be the most powerful creative catalyst in history.

The more easily everyone can make anything, the harder and more important differentiation becomes. Today's chat UIs, graphs, and command-based interfaces are not the pinnacle of modern interaction. Now that we can actually converse with computers, we should imagine interfaces that are more generative, vivid, responsive, and dynamic.

"It's time to invent new interfaces. It's time to invent new aesthetics. The best practices haven't been written yet. We are the ones who will write them."

Katie believes that just as multitouch technology opened up entirely new interactions and synthesizers made previously impossible sounds possible, AI also enables new forms of creation. Indeed, on the internet, more interesting data visualizations, restaurant websites with distinctive personality, and personal creative and artistic work using AI are emerging.

Stripe, too, uses AI not as a substitute for creativity but as a means of amplifying its creative team's capabilities. For example, on the cover of the magazine Built to Grow, building on excellent iterative work by a human colorist, they used AI to refine the fine placement of colors and lines more precisely. This is a case not of eliminating good human judgment, but of extending that judgment more broadly.

"It was like taking good human judgment and scaling it so we could make something truly amazing."

How to Make AI a Better Creative Partner

Katie suggests the following practical stance.

1. Make your inputs more specific.
   Don't just say "Make a website for a Korean barbecue restaurant"; provide what the brand believes, its definition of quality, the experiences it considers important, and reference material.

   "This is what I believe. This is our quality bar. This is what we care about."

2. Rigorously scrutinize the output.
   Don't fall into the "burrito dilemma" of being satisfied with the first fast result; go one step further. If needed, have AI agents with different perspectives critique and refine the results, but the responsibility for ultimately pushing toward something better lies with people.

3. Organizational culture must protect experimentation.
   Repeating familiar patterns is safe and comfortable, but distinctive attempts that make the status quo better are much harder. Leaders shouldn't just tell teams to use AI; they should provide space for strange, unfamiliar, not-yet-finished experiments.

   "Protect the weird and the unfamiliar."

If AI has lowered the cost of building, some of those savings should be reinvested in making more special output. If you use AI only to do what you were already doing faster, you miss the most interesting possibilities.

"If we use AI only to build what we were already building faster, we miss the most interesting part."

7. Closing: Choose a Creative Renaissance, Not Digital Zombie Buildings

As a contrast to the replication of postwar modernism, Katie brings up Gothic architecture. In 1850, John Ruskin praised the craftsmanship and quality of Gothic architecture. No two Gothic buildings were exactly alike, and even each individual column carried the touch and unique judgment of the person who made it.

"No two Gothic buildings were the same. Honestly, not even one column was exactly like another."

Users aren't impressed by the mere fact that you got something moving with 3D tools and AI in 30 minutes. What impresses them is how well the product solved their problem, how well it anticipated their needs in small details, and how distinctive the brand's voice is.

"Users aren't impressed that we got something moving in 30 minutes with Three.js and Blender. They're impressed by how we solved the problem and how we anticipated their needs in the details."

In the end, the choice in the AI era is clear. We can use AI to fill the digital world with yet more zombie UI, or, conversely, we can build stronger, more thoughtful products that bear the maker's touch. The standard Katie emphasizes is not the technology itself, but the point of view, systems, quality bar, and ambition of the people using it.

"AI shouldn't be used only to raise the floor of our ambition, but to raise its ceiling."