I read the New Yorker piece on "Claudian sameness" and had one clear reaction: everyone's mad at the wrong thing.
If you missed it, Kyle Chayka spent a column documenting a pattern that keeps showing up across digital products. Beige and cream backgrounds, paired with rust-orange accents. Big italic serif headlines, with subheadings underlined in a way that starts to feel excessive. A ticker bar running across the top, the kind that makes a SaaS landing page look like it's reporting breaking news. It shows up on websites and in pitch decks alike, basically anything built with the tool. He named names too, five of them: Newsletrix, Wesley Wang Media, GrassDX, Haute Living, DeployGraph. Five completely different businesses, all built with Claude Design, and somehow all wearing the same outfit.
Designer Matt Ström-Awn told The New Yorker that two separate startup founders showed him pitch decks that looked "generated by the same company." Celine Nguyen, another designer quoted in the piece, said she now feels "instinctively repulsed by the warm tones" she used to love. That's a rough thing to say about a color palette. It's a rougher thing to say about your own brand.
So here's my problem with how this story is being told. The headline everywhere is some version of "AI is making the internet look the same." And technically, sure, that's what's happening. But say it that way, and you let everyone off the hook. The founders. The marketing teams. The agencies. All of us who opened a tool, typed a prompt, and shipped whatever came out first.
The tool didn't skip a step; we did
Anthropic has actually been upfront about this. Their own guidance admits Claude has "a consistent default house style" and that the default is "persistent." They even warn that telling the model something vague like "don't use cream" usually just swaps in a different fixed palette instead of producing something original. That's not a bug report. That's a tool telling you exactly how it behaves when nobody gives it direction.
And that's the part that gets lost. A default isn't the tool failing you. It's the tool doing precisely what a default is supposed to do when nothing overrides it. The sameness isn't proof that AI can't produce distinctive work. It's proof that most of the people using it aren't giving it the right guidelines, design tokens, and governed content to work from.
I don't think this comes down to laziness. Some teams don't have a designer in the room to catch it. Some are moving fast without the time or the budget to find a better solution. And some just don't have much design knowledge of their own, so if it looks good and works, they ship it. Underneath all of that is the same gap: a brand that never got past a logo and a color pick, or one that was documented but not specific enough to survive contact with an AI that has opinions of its own. Most companies I've seen fall into that second, vaguer category, not the first.
That distinction matters because it changes what you're supposed to do about it. If the problem is that AI flattens everything, the fix sounds hopeless: avoid the tools or accept generic output as the cost of speed. If the problem is that your brand wasn't specific enough to hold its ground, the fix is concrete and already within reach. Write down your actual colors, not vibes. Name your typefaces. Decide what your brand refuses to look like. Make it yours, and make the AI follow it, not the other way around.
The developers clearly already figured this out. According to OneMinuteBranding, search interest for "design tokens generator" has grown over 900% in the last two years, mostly because AI coding tools started shipping code fast enough that nobody could keep three different codebases visually consistent by hand anymore. A tokens file, or a CLAUDE.md that spells out your palette and spacing and type choices, does for design work exactly what a style guide used to do for a junior designer on day one: it tells the tool what "yours" looks like before it reaches for what's easy. Content teams are catching up to the same idea: treating governance as the thing that makes AI output trustworthy, not a compliance step bolted on after the fact.
Taste didn't get less important; it got more expensive to skip
Here's the thing I keep coming back to. A passable design used to cost real money and real time. Now it costs a Claude subscription, plus someone who tells it what to do. That second part isn't optional, and it's not free either, but it's a different kind of cost: less "pay a designer to produce something," more "pay someone to direct the tool." That's genuinely good, but it also means passable is everywhere, which makes the thing that isn't automatic, direction, judgment, an actual point of view, the only part of the job that still separates one brand from another. Loredana Crisan, Figma's chief design officer, put it well: "The job of the designer is to stay with uncertainty long enough to discover something new." AI didn't remove that job. It just took away the excuse for skipping it. You can no longer blame a rushed, generic result on not having the time or the budget, so what's left on the page is a much more honest read of whether anyone actually had a point of view.
So no, I'm not going to pile onto "AI slop" as a phrase, or write the version of this article where the villain is a language model. The villain, if there is one, is skipping the step where someone decides what a brand actually is before asking a tool to build it. That step was always the hard part. AI didn't create it. AI just made it obvious the moment you skip it.
Where this leaves the rest of us
If you're handing brand or content work to an AI agent, human or otherwise, ask it what it's building from. Your tokens, your voice, your governed content, or its own defaults. That's not a nice-to-have anymore; it's the whole difference between output that looks like you and output that looks like everyone else who also didn't answer the question.
Marketing teams: does this sound like us, or like anyone?
Marketing teams feel this one as a plain fear: does the content we're putting out actually sound like us, or does it sound like whoever wrote it? No amount of clever prompting fixes that, because the problem isn't the prompt. It's what the model had to work with. This is the whole idea behind dotAI: it generates from your real content types and taxonomy instead of guessing at placeholder structure, so what comes out has your shape, not a generic one.
Brand, comms, and legal: what if it ships wrong?
For the people who own the narrative, comms, brand, and legal, the worry looks different but comes from the same hole. Their question is simpler and scarier: does anything ever go live off-narrative, badly timed, or just generic, before a customer or a regulator sees it? A style guide doesn't catch that, because nobody opens a style guide at 11pm before something ships. What catches it is a required sign-off step before anything publishes, a record of who approved it and when, and permissions that decide who's allowed to hit publish in the first place. That's the difference between hoping something reads off-brand before it goes live and actually knowing, with a trail behind it, that someone checked. And underneath all of that is the same structural choice: content modeled around what it actually is, not where it happens to sit on a page, so it holds its shape no matter what or who is building from it.
Replatforming: the one moment this stops being theoretical
There's one moment where all of this stops being theoretical: replatforming. Every team facing a CMS migration ends up asking the same question, lift and shift, or full redesign? Lift and shift feels like the safe answer, move what exists, touch as little as possible. But it just carries your undocumented brand, your inconsistent tokens, your never-quite-written-down guidelines into the new system, and hands all of it to whatever AI agent touches that platform next, with no AI governance in place to catch it. A redesign is the one point in a company's lifecycle where someone is already being asked to write down the colors, name the typefaces, and decide what the brand refuses to look like, the exact groundwork this whole piece has been arguing for, and the natural moment to build AI governance in alongside it, not bolt it on later. If you're replatforming anyway, that's the moment to actually do the part that prevents the sameness problem, not after.