Key Takeaways

  • AI defaults to generic past patterns if you do not bring a clear point of view.
  • Customers evaluate software quality on the final output, never on whether a human or machine built it.
  • Stripe uses the internal term "Pepsi bubbling" to describe tuning details one level deeper than what customers actively spot.
  • High taste requires active noticing of real-world details, user context, and cross-disciplinary art and science.

The Threat of Zombie UI

After World War II, a massive construction boom led to rows of identical, uninspired structures: zombie buildings. Katie Dill sees the exact same pattern threatening the modern AI software wave. When everyone uses the same foundational models without strong opinions, products blend into monotonous "zombie UI."

As Dill puts it: “First one, you have to have a point of view. If you don't, AI will give it for you. And as we've already talked about, that's very likely to be generic and looking at the past.”

Large language models predict the next most likely token. By definition, statistical averages produce average products. If you ask a tool to build a dashboard or write an onboarding flow without explicit taste constraints, you receive the mathematical mean of the internet.

The Quality Standard Never Changes

Some teams lower their expectations when software is generated by machines. Dill encountered this directly when colleagues asked what standard they should apply to machine-generated work.

Her reaction was immediate: “What a curious question. Why should that matter how it was made? It doesn't matter to the users how it was made. It matters to them if it's good or not.”

End users do not grade your product on a curve because you prompted a model instead of writing raw code. They care whether the interface feels crisp, fast, and reliable. Lowering the bar for machine output is an internal excuse that customers will penalize immediately.

Why Stripe Uses 'Pepsi Bubbling'

To keep software from feeling flat, Stripe practices an internal technique called "Pepsi bubbling." The term originated from adjusting visual styling down to the smallest detail: "the frost a little bit here and there, softer edges, a bit smaller bubblers." Stripe turned the phrase into a verb for extreme craft.

“Now, we're not shooting for perfection,” Dill explains. “We're shooting to go one level deeper than what your customer could see. That meticulous craft will show up for them.”

The goal is not vanity polish. When you adjust the subtle friction points, visual hierarchy, and edge cases past what an average user can consciously describe, they still feel the difference. They register the product as solid, trustworthy, and intentional.

Hone Your Taste by Noticing

AI can generate code, copy, and layouts in seconds, but it cannot decide what feels right. That requires human taste. Dill argues that taste is not an innate gift; it is a discipline of observation.

“Frankly, it's all about getting really good at noticing,” Dill notes. “Notice what your users need and what they want, not just what they say. Notice what about the world around you signals good and great and meh.”

Look at physical hardware, architecture, fine art, and daily human habits. The builders who create distinctive products are the ones who extract signals from the real world and encode those standards into their prompts, design tokens, and review bars.

What to Do With This

Audit your product's latest AI-generated feature against your best hand-crafted screen. Pick three specific elements: typography spacing, microcopy tone, or hover states. Refine them one level deeper than what users explicitly asked for before shipping.