The tools change. Taste and results don't.

“It feels kind of scary how it replaces taste.” A friend sent me that on X this week, and I keep coming back to it. It’s the worry under most of the AI talk in design right now, with Opus 5.5 out and studios rethinking how they work and what they sell.

I should say I’m biased. I’m a huge fan of AI and use it in almost everything I do. But my friend isn’t wrong to worry, and a fan owes a worried friend more than a thumbs up. So here’s a proper answer, in the first of a letter I’ll now write every week.

We’ve been here before

My first answer to my friend was about websites. Back in the day we hand coded them. When Framer components came along, you could have argued they killed taste, because nobody was writing the site by hand anymore. My friend reminded me that before that, everyone designed sites in Photoshop and sliced them up for a developer. Each of those changes came with people saying the craft was dying.

It goes back a lot further than websites. When photography arrived around 1839, the painter Paul Delaroche supposedly declared painting dead, though nobody has found a record of him saying it. Twenty years later Baudelaire called photography art’s “most mortal enemy”. In 1906 John Philip Sousa warned that recorded music would cause “a marked deterioration in American music and musical taste.” When designers started setting type on Macs, Massimo Vignelli called the results “an aberration of culture.” In 2018 Boris Müller wrote an essay asking why all websites look the same. And when a Midjourney image won the Colorado State Fair’s art competition in 2022, people online called it the death of artistry. Drag along the line below to see who said what.

Look at the list and the fear is always about the hand: the brush, the typesetter, the person writing every line of HTML. The tool changes how a thing gets made. It doesn’t change who decides what’s worth making, and that’s where taste has always lived.

Things change and that’s OK. Creativity will always exist, it just shows up in different forms.

That’s what I told my friend. Painters kept painting after 1839, and musicians kept playing after 1906. The work changed shape, some of it got better, some got worse, and people argued about which was which, the way we’re arguing now.

What taste actually is

People talk about taste like it’s a feeling you either have or don’t. In practice it’s a set of decisions. It’s choosing one direction out of ten. It’s cutting the section the client loves because the page reads better without it. It’s knowing that this founder’s customers will trust a plain, confident sentence more than a clever one. It’s noticing that a headline sits four pixels too close to the image and caring enough to fix it.

None of those are things you type. They’re things you judge, and you get better at judging by looking at a lot of work, making a lot of work, and being honest about which parts are good. AI changes the second part, because making is now cheap. I can see many versions of an idea in the time one used to take. Each one is a chance to train my eye, if I actually look and choose, and a chance to settle for something average if I don’t.

So I don’t think AI replaces taste. I think it makes taste the only part of the job that can’t be skipped.

Inside the machine

There’s real science behind both halves of this, starting with the machine. The models behind tools like ChatGPT and Midjourney learn from an enormous amount of existing work and get very good at producing what’s likely, and likely is another word for typical. Researchers keep finding the same pull toward the middle. In a 2024 study in Science Advances, writers who used AI ideas wrote stories that readers rated as more creative, but the stories came out more similar to each other. Tuning a chatbot on human feedback, the step that makes it pleasant to use, has been shown to make its answers less varied. And a 2024 paper in Nature found that when models learn from other models’ output, the unusual cases are the first thing to disappear.

Inside your head

Now the head. Psychologists have known since the 1960s that we tend to like things more the more we see them, but only up to a point. For things we look at, liking rises with exposure and then falls. In design studies, people prefer products that are new and still recognisable at the same time, and in one study of 88 chairs, the moderately novel ones were rated the most beautiful, ahead of both the most typical and the strangest. The industrial designer Raymond Loewy called it MAYA: most advanced, yet acceptable. Music shows something similar. Across 80,000 chords from pop songs, a surprise felt best when listeners were confident about what was coming.

Put the two together and you get the job. Left alone, the machine lands on the familiar side of that curve, because familiar is what it was built to produce. Taste is knowing where the sweet spot sits for this audience and this brief, and pushing the work there. The curve isn’t the same for everyone, either. What feels bold to a bank’s customers feels tame to gamers.

That knowing is trained, which is the hopeful part. In a well-known 1973 study, chess masters could rebuild a real game position from a five-second look, but with pieces placed at random they did hardly better than beginners. Their skill was a library of patterns built over years of play. A small eye-tracking study found something similar in art: untrained viewers looked at the people and objects in a picture, while artists spent more of their time on its structure. Designers build that same library, one project and one reference at a time, and no tool can hand it to you.

What clients are really buying

Taste is half of it. The other half is that nobody hires a studio for files. A founder comes to us because the homepage isn’t turning visitors into calls, or a raise is coming and the brand doesn’t look like a company worth funding, or a launch gets one shot in the feed. What they’re buying is the result: more booked calls, a round that closes, a launch people actually watch. The iAgent launch video I wrote about in No. 003 matters to me because it passed 1.9 million views by their CEO’s count, not because of how many hours went into the edit.

That’s where AI changes the most, and I think it’s good news for designers. When making the thing gets cheap, it gets hard to sell the hours, so the conversation moves to what the work is supposed to do for the business. Taste is how you get there. It’s knowing which of ten directions will make this founder’s buyers trust them, and having the nerve to cut the other nine. Taste that never shows up in a result is decoration, and a tool that makes decoration free is only a threat to people selling decoration.

How I use it

Day to day, I’m loving the creative process more because of it. This site is one example. Until this month it was a Framer site, and I rebuilt it in code. A few years ago that would have sounded like going backwards. This time AI wrote the code, and I spent my hours on how it should look and what it should say. Then & Theory’s shop went from an empty folder to a live store in three days the same way.

The pattern is the same on client work. AI does the making, fast and in volume, and I do the directing and the choosing. It gives me ten directions in the time one used to take, and I still have to know which one is good, which one is a cliché, and which one will get the person paying for it the result they came for.

Try the figure above. With no brief, the directions slide toward the same safe, centred layout. With a strong one, they spread out and get interesting. That’s the honest picture of working with these tools: what comes out depends on what you bring.

Where it still falls short

I’m a fan, but it doesn’t do everything. Three gaps show up in my work every week. The first is that, left alone, it’s generic. Ask for a landing page with no direction and you get the average of every landing page it has seen: centred headline, two buttons, three cards. Nothing in it is wrong, and none of it belongs to anyone. The fix is a real brief, real references and someone willing to say no to the first answer.

The second is that it doesn’t know the client. It can’t sit on a call and hear what a founder is nervous about, or know that their buyers have been burned by a competitor and need proof before promises. That understanding is most of what makes design work for a business, and it still comes from people talking to people.

The third is the last ten percent: spacing, type details, how an animation feels when it eases in. AI gets close and then stops, because “close” is where the average lives. Getting from close to right still needs a trained eye, and a fair amount of patience.

What’s changing at my studio

My studio runs differently than it did a year ago. Turnarounds are faster. A store in three days or a full identity in a few weeks doesn’t feel like a rush job anymore. Speed used to cost quality, and now it mostly costs decisions, which have to be made quicker and with more confidence.

Clients see more directions, earlier. Instead of a handful of polished concepts late in a project, they see a wide spread in the first days, and we spend the time narrowing it down together. My job has moved toward curating and explaining why one direction is right for them.

And we sell outcomes more than hours. When making a page takes a fraction of the time it used to, billing by the hour stops describing the value. We’d rather price around what the result is worth to the client, a site that books calls or a launch that lands, than around how long the layout took. It’s a better conversation for both sides, because it starts with what they actually want.

What happens to agencies

The last thing my friend asked was what the future holds for people running design agencies. I run one, and I think it’s the same job with new tools, like the moves from Photoshop to Figma to Framer. Making things is getting cheaper and faster, so what clients pay for moves toward judgment and results. Studios can be smaller and still make more. Selling hours of production will get harder to defend. Things adapt, and some people and studios will get replaced, the way they always have when the tools change.

I’m not the only one seeing this. A designer I follow, who runs his studio alone, said this week that agency demand is already dropping. But he also said far more companies are being started, a lot of them AI startups, and that what they pay for is time to value: something real in 48 to 72 hours. He’s moved from handing off designs to shipping the code himself. That’s the shape I think studios are heading toward: fewer people, faster turnarounds, and the result as the product.

Don’t switch out of fear

That still isn’t a reason to change how you work out of fear of being replaced. If you love designing by hand, keep doing it. People still paint in oil and shoot on film, long after the tools that were supposed to end them. What you make things with is a creative choice, and it’s yours.

We only live for so long. Spend it on the work that makes you happy. For me that work has a lot of AI in it right now, and I’m enjoying it more than I expected. If hand work is what makes you happy, keep it. Just make sure you chose it, and fear didn’t.

If you run a studio, I’d like to know what’s changed for you this year, or what you’ve decided to keep the same. Reply and tell me. Next week, something else.