Green on Purpose

The dashboard had everything. Stat cards, charts, tables, filters, buttons. It had been built quickly, with AI doing most of the heavy lifting, and on screen it looked finished.

Then the PM walked a merchant through it on a client call, and somewhere along the way, the pitch started to slip. Every screen told a different story. Nothing connected one number to the next. There were buttons everywhere, but nothing told the merchant what to do next.

Nothing was broken. Something was missing.

When the dashboard came back for a redesign, it was clear the fix had nothing to do with better charts. Someone had to decide what the merchant should see first, what it meant for them, and what they should do next. AI could build every screen. It couldn’t make that call.

Earlier this year, OpenAI president Greg Brockman posted five words on X:

Taste is a new core skill.

Greg Brockman

After that call, it made complete sense to me.

And it wasn’t just the dashboard. At Juspay, I work across very different products: an AI customer engagement platform, a checkout that’s been live for three years and is now being rebuilt, the cart experience, and this analytics product. This year, all of them ran into the same wall. Building got easy. Deciding what to build, and how it should feel, didn’t.

The day building got cheap

A couple of years ago, even a working prototype could take weeks of engineering time. Today, one person can build one over a weekend. When building gets this cheap, more people do it, and products start to look the same.

Clayton Christensen described what happens next in The Innovator’s Solution (2003) through the Law of Conservation of Attractive Profits. When one part of a business becomes commoditized, the money doesn’t simply vanish. It moves to whatever remains difficult to do well. Today, that’s choosing: what to make, what to leave out, and how it should feel.

I see this most in checkout. In India, almost every serious player has a one-click checkout, so the feature itself isn’t a moat. A clever technical edge lasts until someone ships something similar next quarter. When we started rethinking ours, the question wasn’t “What can we add?” It was “Why would anyone pick ours?”

This isn’t a new problem. In a 1995 interview, Steve Jobs criticized Microsoft for lacking taste: not just visual polish, but original ideas and a distinct point of view in its products. The lesson isn’t that taste guarantees success. It’s that when everyone can build, having a reason for building something differently matters.

Zombie UI

Katie Dill, Head of Design at Stripe, gave a talk that put a name to what I’d been seeing: zombie UI. She draws a parallel with the post-war building boom, when builders copied modernism’s visual language without its underlying thinking. Cities filled with what she calls zombie buildings. They stand, they function, but they seem to care about no one.

AI, she warns, can do the same to software: screens that look finished but feel empty.

AI is very good at producing plausible answers from familiar patterns. Without a clear point of view, those patterns become the default. And when a screen appears in seconds with rounded corners and soft shadows, it feels done long before it is. Dill calls this the microwave burrito problem: speed makes something feel ready, even when the quality says otherwise.

That dashboard was zombie UI. Nobody did anything wrong. The tools made “done” so easy that it became easy to mistake for “good.”

Someone has to be the editor

Quality control used to happen before the build. You had twenty ideas and the budget to pursue one, so you argued, challenged assumptions, and pruned early. Now you can build all twenty in a week. That’s a gift, but the filter has moved to after the build, where saying no is much harder because the thing already works.

So every team needs an editor: someone who experiences the product like a real person and asks whether it solves the problem end to end.

That’s my role on the dashboard now. The work isn’t simply making better charts. It’s connecting the information, giving it a story, and making sure the merchant can move from understanding a problem to deciding what to do about it.

Dill draws on an idea from Nabeel Qureshi to explain why AI-generated slop bothers us: we can sense when the details didn’t matter. The cup is green, but it could just as easily have been blue. Nothing about the choice was deliberate.

The editor makes sure it’s green on purpose.

Taste starts before the first pixel

Taste isn’t just polish. A few years ago, pixel-perfect execution was a moat. Now AI can tighten spacing, clean up layouts, and generate polished interfaces in seconds.

So I think of taste in two orders.

First-order / before

Comes before any pixel: which problem is worth solving, what the user should feel, what matters most, and what to leave out.

Second-order / after

The craft that follows: hierarchy, interaction, typography, motion, and all the details that make an experience feel considered.

Both matter, but you can polish the wrong thing perfectly. No amount of pixel work would have saved that dashboard.

First-order taste isn’t only a designer’s job. PMs, engineers, and founders make these calls every day. What designers bring is the habit of connecting what users need, what the product intends, and how the experience should feel.

That’s why design belongs in the room before the build, not after it.

The cup has to be green on purpose, but first, someone has to decide whether you needed a cup at all.

Taste is distribution

Taste was never just about looking good. When every product does roughly the same thing, people notice the one that feels like somebody actually made deliberate choices. And when an experience is memorable, people talk about it.

Look at Apple’s iPhone Duo, announced in September. One of the details people noticed was the soft blur as the phone unfolds, preserving the sense of where your content sits while the display changes shape. It’s a small interaction, but it communicates something important: the software was designed for the fold, not simply adapted to it.

A recreation of the effect appeared on a Galaxy Fold, too. But copying an interaction is different from having the judgment to decide which details matter in the first place. One reproduces an output; the other can keep generating meaningful decisions.

We’ve felt a smaller version of this on our cart. Instead of offering dozens of themes and leaving merchants to figure everything out, we invested in good defaults. A merchant can install it, change nothing, and still get a cart that fits their store.

That means fewer decisions, a faster path to going live, and less tweaking afterwards. The simplicity isn’t an absence of design. It’s the result of decisions we’ve already made on the merchant’s behalf.

That opinionated simplicity is a big part of our traction.

Teaching taste to the machine

More and more products are being built without a designer in the room. Teammates are using AI tools, and agents are fixing things overnight. One person can’t review every output. Taste has to be built into the system.

Dill makes this point through Gutenberg. For the 42-line Bible, his workshop used around 290 distinct characters, not just the 26 lowercase and 26 uppercase letters of the alphabet. The set included different letter widths, abbreviations, and ligatures, helping machine-printed pages retain the visual rhythm of carefully handwritten manuscripts. (Gutenberg Museum · Katie Dill’s talk, at 10:08)

Design systems need the same shift: from component libraries that keep things consistent to opinionated systems that include full flows, meaningful defaults, and the reasoning behind them.

The old system scaled consistency. The new one has to scale intent.

That’s the takeaway for designers: your taste reaches furthest when it lives in the system, not just in your files. It’s why I’ve started building internal tools and going straight to coded prototypes, and why our team has taken more than seven products through design in three months.

The goal isn’t to review every pixel an AI produces. It’s to give the system enough direction that it can make better decisions before a designer ever sees the result.

Taste is a muscle

Taste isn’t something you’re simply born with. Rick Rubin told 60 Minutes that he can’t play an instrument or run a soundboard; what he’s paid for is “the confidence that I have in my taste.”

That confidence comes from years of noticing: what people need, not just what they say; what makes you smile in a product versus what makes you sigh; which details make an experience feel thoughtful and which ones are merely decorative.

Bring a point of view, or the model will bring you the average. And as Dill puts it, protect the strange.

The point isn’t that AI can never have taste. It’s that access to powerful tools doesn’t automatically give you the judgment to use them well. That part still takes practice.

For designers, this is good news. The job is moving upstream, and there’s more room for us to contribute, not less. For PMs and founders, it means bringing design into the conversation at the start. For engineers, it means “Should we?” and “How should this feel?” are becoming some of the most valuable questions in the room.

The bottleneck moved

For years, the team I work with shipped with very little design capacity, and design was usually the last step before launch. That has quietly flipped.

Designers now come in at the start of new products. On our newest AI product, nobody wrote code first; I worked out the story before we moved to execution.

And a few weeks ago, our head of product said something I’ll remember for a long time: taste is becoming our bottleneck, and we need more designers.

I’m finishing this on October 15, the day of Figma’s first-ever Config India, here in Bengaluru. Enjoy the new features. But tools change every year now. Some get replaced, and some you’ll build yourself.

What stays is the eye you bring to them.

So yes, learn the new tools. But train your eye and your judgment even harder, because a shortcut saves you minutes, and good judgment saves the product.