AI gave seven models a law firm brief and all seven picked navy and gold

Let me catch you up. The industry fell in love with a phrase last year: "taste is the new moat." The pitch was simple. AI made good-looking screens cheap, so the one thing left to sell is the eye to pick the good one. Sounds right. It also falls apart the second you try to run a team on it. The sources in this cluster poke holes in that story from four directions, and together they point at something more useful than taste. Here's where we are.

The deep cut

• Name the fact behind the call or it's taste. Patrick Neeman's rule holds: if you can't cite the research or the user, you're trading opinions.
• The preset is the enemy, not bad quality. Mathieu Poli's seven AI models all picked navy and gold for a law firm, finish without a decision.
• Show the dirt, not the trophy. Alex Harper says recruiters now read a flawless portfolio as "did a human make this?"

Taste can't lose, which is the problem

Taste feels like perception, but a lot of it is expectation. A 2001 wine study dyed a white wine red and 54 trained tasters described it with red-wine words. Nothing changed but the color. Two skilled designers can look at one screen, disagree, and neither can prove the other wrong.

That's why "does it have taste?" is a weak review question. It has no failure condition. Ruiqi Zhou called taste unfalsifiable: a standard you can never fail works as a status marker, and status markers reward whoever already holds status. Usually the most senior person in the room.

So Patrick Neeman draws a line you can use. Taste is a preference you can't settle. Judgment is a claim you can check, because it rests on tested principles plus evidence about your users. Fitts's Law and Nielsen's heuristics turn "this feels off" into "this button is too small for a thumb." A junior who knows the law can argue with a VP. That's the real value.

The machine makes everyone's work average

Here's the part the moat crowd missed. Polish used to signal care, because careful work looked different from a template. Not anymore. Mathieu Poli ran twelve app briefs through seven AI models and got the same answers back. For a law firm, all seven picked navy with gold. A boxing gym in a working-class Marseille neighborhood came back red-on-black with condensed type, mood words like "raw, gritty," and zero trace of Marseille.

He likens it to the Yamaha DX7 synth in 1983. It shipped with 128 presets two people built in four days. One sound, "E.PIANO 1," landed on 61% of the 1986 R&B number ones. A quality sound handed to everyone became the sound that dated every record that kept it.

The lesson for your reviews: the floor is now polished by default, and 34 of those AI colors still failed contrast for white text. Chasing better finish is a race you lose every model release. The thing that sets work apart is a decision someone made and can defend.

Sameness isn't always a bug

Before you blame AI for everything looking alike, Luke Wroblewski lined up TVs, phones, and cars across decades. The pattern is obvious, and the causes are mostly boring. Once a product's purpose is clear, design gets out of the way. A TV is for watching a picture, so the ideal TV is mostly picture. Mature categories compete on price, and standing out gets risky.

Dense Discovery's Kai pushes the same point into the built world. Modern buildings feel drab partly because we learned to make them energy efficient and earthquake-safe, and those rules narrow the forms. "The intentionality of the market is to sell product," he writes. Convergence is the operating system working as designed, not a failure of nerve.

This matters for your team because it separates two things. Some sameness is physics and economics. Some is laziness. Only the second kind is yours to fix.

The portfolio stopped proving anything

The static portfolio is the clearest casualty. Alex Harper puts it flat: a flawless gallery now reads as "I wonder if they prompted Claude for three hours." Perfection went cheap, so pristine pixels stopped signaling skill. A single remote posting pulls thousands of applications, and nobody clicks your heavy Webflow link to read 3,000 words on why you chose a shade of blue.

What reads as real now is the messy middle. Show the bug you fixed, the bad first draft, the exact Figma components you shipped and who you worked with. Harper's rule: don't show the trophy, show the dirt on your shoes. AI can't counterfeit a hyper-specific troubleshooting story.

The fix for both hiring and craft is the same move. Write the reason next to the choice. Poli's example: a design system entry that reads "primary color: #1B2A4A" says nothing. "The blue of the firm's original sign, darkened to hold contrast" shows a decision, and gives the next person a reason to keep it or change it.

Three questions for your team

• In our last review, how many calls could we back with a fact or a user session, and how many came down to the most senior person's preference?
• Where in our product did we accept the AI or template default without writing down why, and what would it take to add the reason next to the value in our design system?
• When we hire, are we still asking for a polished portfolio, or do we ask candidates to walk us through a real decision and the mess behind it?