AI Made Taste and Experience the Premium.

I'm in the middle of a move right now. Another chapter, another life adventure, and rather than sit twiddling my thumbs in the gap between the old life and the next one, I took on a small design project to fill it. A quick one, a week or three, just enough to keep my hands busy while everything I own sits in cardboard boxes. And it was somewhere in the research for that client, knee-deep in competitor sites and the usual inspiration trawl, that I started noticing the same thing over and over. The purple.

A very specific blue-to-violet gradient, the one that used to whisper "futuristic" and now just announces "nobody made a decision here." Once you've seen it, you can't unsee it. The same gradient, the same Inter font, the same three little boxes with icons in a row, the same rounded corners on absolutely everything. I can spot it in about a second now. It's the visual equivalent of a handshake that goes on slightly too long.

There's been a word for this for a while now: AI slop. What's new isn't the name, it's the scale. Just last week the New Yorker ran a piece about how it's quietly taking over the internet. The example everyone seized on was two different startups proudly showing off two different sales decks, then slowly realising they were basically the same deck. Same bright mission slide. Same four-box "here's our market." Same centred line announcing the disruptive masterstroke. Different logos, identical everything else, both built in the same tool: Claude Design, Anthropic's web builder.

The New Yorker pinned the whole thing on Claude, which is a touch unfair, because this was never really a Claude problem. Point v0 at a blank page, or Lovable, or ChatGPT, and you get the same defaults: the same purple, the same three boxes, the same tidy nothing. It's baked into how these models work, not into one company's product. They all reach for the middle, because finding the middle is the entire job.

So that's where we are. The obvious response is to panic, or to write yet another mournful think-piece about the death of creativity. I'd rather do the opposite. Because once you understand why the slop happens, it stops looking like a threat and starts looking like the best thing to happen to anyone with real judgement in years.

The look isn't a style. It's an average. This is the part people get wrong. They assume the model picked purple because purple is on-trend, like it has taste and made a call. It didn't. An AI image or layout tool works by predicting the most likely next thing based on everything it was trained on, which means it reaches, every single time, for the statistical middle of the road. Ask it for "a modern landing page" and it doesn't design one. It calculates the average of every landing page it has ever seen and hands you that. The purple has a wonderfully boring origin story too: Tailwind, one of the most popular CSS frameworks of the last few years, ships with an indigo-and-violet default palette. So the training data was already soaked in purple. Tailwind makes purple easy, developers reach for purple, the model learns purple is "correct," and round it goes. Nobody chose this. It's just what falls out when you average the modern web and hand it back.

Which is exactly why it has nothing to say. A real design era, even a naff one, had a point of view. Skeuomorphism, all those leather-stitched calendars and felt notepads, was a daft idea executed with total conviction. You could stand in front of it and tell what it was reaching for. Slop has no conviction, because you can't average your way to one. It takes a little from everything and commits to none of it. It is the design equivalent of a press release that's been through legal four times: technically present, saying nothing, offending no one, remembered by no one.

And let's be honest with ourselves. None of this is new. The internet has been full of average design since the day it learned to load. Clumsy layouts, weak typography, UX held together with hope and a contact form that goes nowhere. We didn't need a single line of AI to produce thirty-odd years of forgettable websites. We managed that one entirely on our own. What's changed isn't the quality of the bad stuff. It's the speed. AI didn't invent mediocre design, poor instincts, or lazy UX. It just gave us a quicker, easier, cheaper way to mass-produce all three. The slop was always here. We've just industrialised the factory.

Here's the uncomfortable bit, though. A lot of the time, it's fine. This is where I lose some of you, but stay with me. If you're spinning up a landing page for a two-week conference, or an internal tool that six people will ever touch, or a quick microsite that exists to collect email addresses and then die quietly, the purple gradient is completely acceptable. Nobody's brand was ever made or broken on a disposable page. For the boring 80% of what gets built, "looks like everything else" is a perfectly reasonable place to land, and pretending otherwise is just designer vanity. Slop got that work done in an afternoon for free. Good. Genuinely. That work was never where the value was anyway.

But "free" and "worthless" sit a lot closer together than people like to admit. The moment anyone can generate the median in ten seconds, the median is worth precisely nothing. It has no scarcity, so it has no price. Which means all the value, all of it, has quietly migrated to the one thing the average can't give you: the deliberate decision to be different from it. The bit where someone looks at the perfectly competent purple thing and says, no, not for this. That judgement used to be one ingredient among many. Now it's the whole meal.

And that judgement has a name. It's taste. Not taste as in "nice things," the way people use it when they mean expensive. Taste as in knowing what to keep and what to throw away. Knowing that this client should not look like a tech startup even though the tool desperately wants them to. Knowing that the headline "Build the future" says absolutely nothing, and that "Financial infrastructure for the internet," which is how Stripe actually describes itself, says everything in five words. Knowing why a 10% drop shadow on every card makes a page feel like a spreadsheet wearing a costume. Taste is a thousand small decisions, most of which the user will never consciously notice, all of which they'll feel. The machine can make every one of those decisions. It just makes them all towards the middle.

This is the part I find quietly reassuring after twenty years of doing this. The skill that's suddenly scarce isn't the one I worried AI would take. It was never really about who could push the pixels around fastest, or who knew the keyboard shortcuts, or who could produce a clean grid. The model does all of that now, instantly, and frankly it does the grid better than a tired junior at 5pm on a Friday. What it cannot do is sit in a room, understand a real business and a real human problem, and have an actual opinion about what this specific thing should be. That's not a tool. That's a person with judgement, and judgement is built the slow way, out of years of getting it wrong and noticing.

So if you want to climb out of the sea of sameness, here's where the work actually lives. Not in better prompts, which give you diminishing returns fast. In the decisions around them:

• Be specific where the machine is vague. Real copy beats "Build the future" every time. If your headline could sit on a fintech, a CRM and a dog-walking app without changing a word, it's slop. Say the actual thing.

• Use negative constraints. "No purple, no gradients, no dark mode" genuinely works, because you're forcing the model off the well-worn path. Then point it somewhere real: "warm and muted, more worn leather and dark wood than neon."

• Give it your brand physics, not just your logo. Real fonts, real spacing rules, real colour tokens, real photography of real people. Without them, it fills every gap with the average. With them, it has something to deviate towards.

• Know what to delete. The most undervalued skill in design right now is looking at a technically fine layout and removing the three things that make it generic. That's not prompting. That's taste, applied with a delete key.

So what does all of this actually tell us? That AI didn't devalue design. It devalued the cheap, repeatable, averageable part of design, the part that was always closer to assembly than to thinking. And in doing so it has, almost by accident, made the genuinely human part more valuable than it's ever been. The point of view. The specific call. The decision to ignore what's statistically most likely because you understand the thing in front of you well enough to know it deserves better. That was always the actual job. It just got a lot harder to fake.

Further reading

A few pieces worth your time if you want to go deeper on any of this:

• The generic style of AI web design — Kyle Chayka's New Yorker column, the one that kicked off the current round of hand-wringing. The two-identical-decks story lives here.

• Why Your AI Keeps Building the Same Purple Gradient Website — the best plain-English explanation of where the purple actually comes from, Tailwind and all.

• The New Skeuomorphism: How AI Makes Bad Design Look Good Enough — a sharp take on why averaged design ends up with no point of view.

• Aesthetics in the AI era: visual and web design trends for 2026 — Ioana Teleanu on the human counter-movement: friction, texture, and design that's trying to feel made again.

• Canva's 2026 Design Trends — the data behind the backlash, drawn from a community of millions, if you like your trend-spotting with numbers attached.