Design Taste AI Can't Copy: The 5 Calls I Still Make Myself

The design taste AI can't copy is the judgment that tells you which of ten good-looking options is right for this product, these users and this moment, and what to remove. It is the part of the job that did not get faster. Claude can build a screen from my Figma frame in minutes, and a video model can animate it, but neither one knows when the result is generic, when the motion lands two percent off, or when a "swipe up" feels cheap in the hand. I wrote it this way in a post on X after shipping my experiments: "The taste is still the slow part. Good. That's the part that's ours." This post is about what that slow part looks like in practice, the five calls I still make myself, and how a designer can train them.

What design taste means when AI builds the screens

"What is design taste" is the most searched version of this question, and most answers stay abstract: a sense of beauty, a refined eye. I think that definition is too soft to be useful once AI is in the room.

Here is the version I work with. Design taste is a series of small decisions you can defend:

• Which reference is the right feeling, and which one is just pretty.
• Which single moment in the product deserves the most care.
• What the interaction should feel like in the hand, not only how it looks in a frame.
• When the output is "fine" but generic, and you should keep going.
• When to stop polishing and ship.

AI changed the cost of everything around those decisions. It did not make the decisions. In my AI design workflow from idea to App Store, the first build takes minutes. Making it feel right takes the evening. That evening is taste.

Design taste AI cannot copy: the five calls I still make

These come straight from how I build the projects on my AI design experiments page. Each one is a moment where the model gave me something reasonable, and the right answer was something else.

1. Choosing the feeling before the layout

I do not start in Figma. I start by collecting: Dribbble and Behance for the feeling, real products and Mobbin for how things should work, and AI images for the world the product lives in. For the Hywith forest landing page, I generated the forest first, with light rays, moss and little flowers, and only then designed the product on top of it. As I put it on X: "When the world exists first, the UI has somewhere to live." An AI can generate a thousand forests. Picking the one that makes the product feel calm instead of decorative is the call.

2. Designing the wow moment first

Every project I shipped this year has one moment people remember: water pouring into a glass in Pulse Health, a glass ball that fills under your thumb in Rare Box, a character whose eyes follow your cursor in Jane Noah. I design that moment first, and the rest of the screens support it. A model will happily give equal weight to every screen. Deciding where the weight goes is taste.

3. Feeling it in the hand, then changing the design

In Figma, the Rare Box said "swipe up to open." Then I had it on my phone, and swiping felt cheap. So I asked for hold instead. Now a glass ball appears under your finger, fills while you hold, dives into the box, and the card bursts out. No frame in the file told me that. Only using it did. In the old handoff days, that change would have died in a ticket. Now I can feel it the same evening and change it.

4. Catching what the model added on its own

On one landing page, the video model made a beautiful sky, and then quietly added raindrop rings to the lake. On a sunny day. On another project, a model re-rendered every clip about 2% zoomed, so the whole design jumped the moment the video started over the still frame. You only catch that if you stare at it like a designer. My fix both times was a rule, not a prompt: keep the clip where it is good, and build the rest in code.

5. Refusing the generic page

"Skip the thinking and you get the same generic page everyone else gets." That line from my first X post on shipping instead of handing off still sums it up. The model's default is the average of what it has seen: a centered hero, a gradient, three cards. Taste is noticing that the result is average even when nothing is technically wrong, and pushing past it.

Where AI is fast and where taste decides

Step	What AI does well now	Where taste decides
References	Generates endless moods and worlds	Picking the one that fits the product and its users
Layout	Builds screens from Figma frames through the Figma MCP	Which screen carries the wow moment, what to remove
Motion	Video models animate between a first and last frame	Spotting drift, odd pacing, things nobody asked for
Interaction	Writes the gesture, the physics, the states	Whether it feels right on a real device
Sound	Generates SFX and music on demand	When sound helps, and when silence is better
Shipping	Deploys a live link in minutes	Deciding it is ready, and what not to ship

Can AI have design taste?

Models are getting better at the surface of taste. They pick nicer type pairings than they did a year ago, and they follow a strong reference well. But taste is not a style. It is a judgment tied to a context: this user, this product, this constraint.

At ViewStats, the creator platform built by the team behind MrBeast, the biggest insight was that creators were not asking for more charts, they wanted confidence about what to make next. "People rarely want more information. They want clarity." No reference board would have told you that. It came from interviews, onboarding analytics and the Discord community. The full story is in the ViewStats creator tools case study. That is taste too: knowing which problem the interface should solve before deciding how it looks.

So my honest answer: AI can imitate the results of taste when you show it good references. It cannot yet decide what matters for your users. That is still the designer's job, and I think it is the most valuable part of it. I made a similar argument in will AI replace product designers.

How to develop design taste with AI, not against it

The good news: AI makes taste easier to train, because you can see more versions faster. A few habits that work for me:

• Collect before you prompt. Build a small board of references and write one sentence on why each one is there. If you cannot say why, it is decoration.
• Design two moments, not every screen. The start and the end. Describe the in-between in words. It forces you to decide what the moment is about.
• Test on a real device. Previews lie. One of my landing pages felt laggy even though every tool said it was fine. The cause was blur effects on buttons over a moving video. Swapping them for static frosted fills fixed it.
• Write down what bugs you. Use the build like a user, list every small thing that feels off, then fix them one message at a time.
• Keep the Figma frame as the reference. Check every AI output against it. If the model moved your design, you will see it.

None of this is a magic prompt. People ask me for "the prompt" all the time. There is no such thing. There is a proper design first, a clear idea of how each piece should behave, and a lot of small, boring fixes.

FAQ about design taste and AI

What is design taste in simple words?

It is knowing which option is right for a specific product and user, and what to leave out. Taste shows in the choices you can explain, not in a style you repeat.

Can AI replace design taste?

Not today. AI can produce many good-looking options and follow strong references. It still needs someone to decide which option fits the context, and to notice when the result is generic or subtly wrong.

How do I develop design taste as a junior designer?

Look at a lot of real products, collect references with a reason attached, build things and use them on a real device. AI helps here: you can try ten versions of a moment in an evening and learn why one works.

Will AI make every app look the same?

Apps built from default prompts already look the same. Apps built from a clear design, a chosen mood and one strong moment do not. The gap between those two groups is getting wider, not smaller.

I am curious about your side of this. Which part of your work got faster with AI, and which part still feels slow in a good way? If you want that slow part applied to your product, you can work with me as your product designer.