A few weeks ago, a company raised $7.9 million to teach AI what looks good. Not what works, not what passes a benchmark. What looks good. Design Arena, where 5.3 million people vote blind on which of two AI-generated designs they prefer, closed a seed round in early August on a simple premise: machines can produce almost anything now, so the only question left is which outputs deserve to exist.
I've been building an interior planning app, a tool that helps people figure out where the couch goes when they move somewhere new, and this is the question that keeps me up. My app can generate a dozen furniture layouts for a room in seconds. So can everyone else's. The generation is free. The choosing is not. And the choosing is the whole game.
The floor came up. The ceiling didn’t move.
Here’s the shift nobody quite prepared us for. AI didn’t make design worse; it made competent design cheap. As the studio Jacob Tyler put it recently, AI raised the floor while leaving the ceiling untouched. The polished hero image, the elegant grid, the moody photography that used to signal a brand had money and care behind it: all of that is now ten dollars a month and ships before lunch.
Which means the work that used to differentiate you is now the work AI is best at. Compete on production polish in 2026 and you’re competing on the thing that’s been commoditized. The value didn’t disappear; it moved up the stack, the way it did when Photoshop democratized retouching and Squarespace democratized small-business sites. Every time a tool democratizes a craft, the advantage relocates above execution. This time it has a name people keep reaching for: taste.
Taste is not a vibe. It’s a system of refusals.
The word gets thrown around now like it’s a personality trait, something you either have or you don’t, best signaled by a nice moodboard. That framing is comfortable and wrong.
The sharpest definition I’ve read frames taste as the cost of the refusal. AI is a yes-machine. Ask for ten options and it gives you ten. Ask which to keep and it hands back the one that looks most like the other nine, because it has no reputation to protect and nothing to lose by agreeing with itself. A person with years in the room has something at stake every time they say no. Taste lives there, not in the skill of the hand but in the discipline of rejection. A brand’s real identity is negative space: the choices it has refused so consistently the absence becomes recognizable.
There’s proof this is now measurable, and it’s my favorite research finding of the year. When Design Arena studied OpenAI’s GPT-5.6 Sol, which jumped 18 places to rank first on their web design leaderboard, they mapped 1,000 of its generated sites and found deliberate gaps in its design space, exactly where its predecessor produced purple gradients, bento-box layouts, oversized hero text, and offset compositions. GPT-5.6 had learned the clichés of AI-generated design and was actively suppressing them. Its edge wasn’t producing more. It was refusing more.
That’s taste, rendered as an engineering result. The best model won by knowing what not to make.
What this means if you build things
The trap is treating taste as aesthetic preference: hiring a strong portfolio and hoping vibes transfer. Taste is a system of decisions, and systems can be practiced. A few things I’m putting into my own workflow:
Write the refusal list before the feature list. Before designing a flow, I name what this product will never do: the patterns, tones, and shortcuts that are off-limits even when they test well. Constraints are the one asset AI can’t generate for you, because it has no stake in holding them.
Treat AI like a talented junior. The designers I trust run AI output through the same critique they’d give a junior teammate: never shipping the first generation, always asking not just “is this good” but “is this ours.”
Design in negative space. Specificity is now the signal that a human curated something. Audiences trained on years of AI saturation skim past the generic and stop at the particular. What you leave out does more identity work than what you put in.
Curate your inputs like they’re the product. Taste develops through deliberate exposure and honest reflection about what works and why. If your references are the same feeds everyone scrolls, your refusals will be everyone else’s too.
The open question
If a model can learn to suppress its own clichés, taste stops separating humans from machines. It starts separating people who decide from people who merely generate. The tools raised everyone to the same floor. The ceiling is still built one refusal at a time.
So the question I’d leave you with isn’t whether AI has taste. It’s whether you’ve done the harder work of knowing, precisely, what you’d refuse to ship, even when the machine tells you it’s fine.
I'm starting a series: Products With Good Taste. Each piece will take one product that earns the label, from Linear's radical restraint to Stripe's disciplined docs to the AI-native tools treating generation as a first-class surface. If there's a product whose taste you can't stop thinking about, tell me in the comments. It might be the next one I take apart!
References
Jacob & Tyler, “Brand Differentiation in the AI Era: Taste Is the Only Moat” - https://www.jacobtyler.com/blog/brand-differentiation-taste-moat/
TechCrunch, “Design Arena creators raise $7.9 million to bring taste to AI models” - https://techcrunch.com/2026/08/03/designarena-creators-raise-7-9-million-to-bring-taste-to-ai-models/
Design Arena Notes, “How OpenAI’s Sol Finally Learned Design Taste” - https://notes.designarena.ai/how-openais-sol-finally-learned-design-taste/
MindStudio, “What Is Taste as a Durable AI Asset?” - https://www.mindstudio.ai/blog/taste-as-durable-ai-asset
Notice The Elephant, “Taste As A Crucial Skill And Differentiator In The Age Of AI” - https://noticetheelephant.com/taste-as-crucial-skill-and-differentiator-in-the-age-of-ai/
The State of AI in UX & Product Design: 2026 (Designlab) - https://designlab.com/blog/ai-in-ux-product-design-trends-2026
Forbes Councils, “AI Is Cool But 2026’s Most Valuable Marketing Asset Is Still Human Taste” - https://councils.forbes.com/blog/ai-is-cool-but-2026s-most-valuable-marketing-asset-is-still-human-taste