There is a website that the last two years of AI have produced a million times. You have seen it. Purple-to-blue gradient. A headline that promises to ten-times something. Three feature cards. A wall of logos. A photograph of a team that was generated this morning. It is competent, it is empty, and it is everywhere.
We are calling the end of it. Not because we hope — because the reason it existed is going away.
Why every AI site looked the same
The sameness was never a model problem. It was a taste problem. Early AI builders were trained to produce something safe — the statistical centre of every website that came before, which is to say the most average possible page. Average is safe. Average never embarrasses anyone. And average, repeated a million times, is exactly the grey sludge we have been swimming in.
The tools optimised for not being wrong. Nobody optimised for being good. Those are different targets, and the gap between them is the whole problem. A site that is never wrong is also never memorable. It is the architectural equivalent of beige.
The opposite of generic is not complicated. It is decided. Someone has to be willing to choose.
What changed
Two things ended the era at once. First, the imagery crossed a line — the new generation of image and video models stopped looking generated, which removed the single biggest tell of an assembled site. Second, and more importantly, builders stopped aiming for average. The work moved from how do we make any site to how do we make a site with a point of view, built in.
That second shift is the real one. It means encoding actual design opinions — real type at real scale, asymmetry on purpose, restraint as a default, motion with manners — into the tool itself, so the easy path produces something good instead of something safe. The taste is not left to the user to supply. It is built into the floor.
What comes next
The next era of the AI-built web is not faster. It is better. The speed is already solved — a site in a minute is no longer remarkable. The frontier now is taste: which tool has an actual point of view, which one makes work you would be proud to sign, which one raises the floor instead of flooding the zone with more beige.
That is the only race we are interested in. Anyone can make a million average sites. The question that matters is whether the next million are worth looking at — and we are betting the whole company on the answer being yes. The generic AI website had a good run. It is over. What replaces it is the part we are here for.
