Everyone's work is starting to look the same.

There's real research on how AI flattens creative work, and the effect is smaller than the panic suggests. The bigger risk is forgetting to think big enough.

Photo by Google DeepMind

If you look closely you can see it happening.

When you open someone else's product and it looks pretty similar to something you shipped. You notice the same onboarding, the same empty state with a very similar friendly little illustration, the same rounded, reasonable, slightly-too-polished everything. In some respects the industry has been converging for a long time, Design Systems made things looks harmonious and best practices spread. Users got a clear reference point across the products they use every day, and we all got to spend our time on harder things.

The downside has been that everything started to look the same.

Aurélie Radom, writing for UX Collective put her finger on the mechanism, and it's a great read: AI didn’t democratise design judgment, but rather it democratised the output. Models are all trained on the same things, so they keep converging on the same familiar shapes, the same type systems, the same compositional logic, repeated across products that have nothing to do with one another. Brand strategists are publishing data on what they’re calling the Great Flattening.

So where’s it’s coming from? It isn’t coming from the culture, or the trend cycle, or lazy designers. It’s coming from inside our files and there is now research that demonstrates this.

---

A few good papers

A team at USC, led by Morteza Dehghani, published a review in Trends in Cognitive Sciences on what happens to human expression when we all start thinking through the same handful of models. Their argument is that when your writing, your reasoning, your way of seeing a problem etc…all get filtered through the same system, your distinctness gets sanded down. We drift toward a shared middle of standardised expression and standardised thought.

On your own, using AI, you tend to produce more. More options, more variations, more volume, but each one a little less original than what you’d have made alone, unless you are actively doing something about that.

In groups, it gets worse: teams using llms generate fewer distinct ideas than teams who just talk to each other and bounce things around, although I’d argue once again that this can lead to group think unless you’re actively managing that sessison.

Then a systematic review and meta-analysis went looking for whether this holds up. They did 19 studies, 61 effect sizes and they found a measurable homogenisation effect when people co-create with AI. In other words, their outputs end up more similar to everyone else’s.

Having said that, the effect they measured is small. AI is not catastrophically flattening all of human creativity. It's doing about as much damage as a badly run brainstorming session, so let's not panic.

---

Why designers observe it most

Homogenisation is strongest in what the researchers call “semantically-constrained ideation”. You have a tight prompt with a narrow space of good answers and a very clear target you’re generating toward.

This is pretty much what happens when we prompt for a good onboarding concept, or empty states, or sign up flows for example. I think these have naturally converged over the years as we’ve become better and better at them. I personally don’t think this is a big problem because applying universal patterns that are proven to work well allows users to have a clear reference point across their products and allows us all to spend time on other things.

The constrained, targeted generation is where the flattening is most observed. The convergence isn’t happening to some abstract pile of content out on the internet. It’s happening in precise tasks.

---

Does it really matter?

The same tool that collapses variance can also expanded our possibilities: more people can make something competent now. Someone with no design team or experience can get to “good enough” on their own. That is genuinely, wonderfully good because clean, clear, sharp design across more surfaces is great.

AI is also lowering the ceiling: it made more people capable, and unless we’re intentional about it, it’s making all of us a little more alike and all the work is converging more (more about this in a different post).

The question isn’t really "is AI good or bad for design" and instead becomes something far more interesting: what can we do with all these new possibilities?

---

Go to the edges

There is in my opinion a huge opportunity for Designers to spend time on the things that are not obvious or easily thought about: how about articulating the culture you’re observing across a demographic? Designing at the edges where typical patterns don’t work? Exploring motion design or layouts that were never possible before AI? Turning alignment into a design problem, sitting with researchers and helping encode what we actually value into systems that will act on it at a scale we've never seen?

The greatest danger not thinking big enough in this era.

Look at what the people at the real edges are doing right now.

Project CETI is using machine learning to decode the language of sperm whales, and they intend to make it possible this decade. They’ve already found a sperm whale “alphabet,” and that the clicks shift with conversational context, and that whales from different parts of the ocean carry different dialects, as distinct as a Liverpudlian and a Londoner.

Earth Species Project is doing it across the animal kingdom, building models for birds, elephants, and dolphins, trying to understand intelligences that share our planet and that we’ve never once been able to hear.

The Vesuvius Challenge trained models to read a Roman scroll carbonised by Vesuvius without ever unrolling it, and recovered a lost work of philosophy on the nature of pleasure that no human had read in two thousand years.

AlphaFold predicted the structure of nearly every known protein and handed it, free, to three million researchers, which is part of why Demis Hassabis and John Jumper share a Nobel Prize.

This decade will drive more change and more possibility than any decade in human history. If the ceiling of our ambition is worrying about convergence in a sign-up flow, something has gone wrong with where we’re putting our energy.

Here’s the thing about those trailblazers, they’re pointing the same energy at a much bigger mountain. Thank you to the people out at the edges for elevating my thinking and my gaze onto the vast horizon in front of me.

---

The question to sit with

Speed was always the easy, obvious thing to measure. Are you faster? Sure. We're all faster now.

A year from now, when everyone on your team has the same tools and the same prompts and skill files, can anyone tell your work apart from theirs? Is there anything in it that could only have come from you, and only you?

Go look at the last three things you shipped. Did you decide this? Or did the model? How original is it?

I don’t think we’re doomed to sameness. I think we just have to choose, on purpose, to keep our edges. The model won’t do it for us, it’s clearly pulling the other way. You have far more power here than you know, and using it is the most fun part of the whole job.

Creativity is limitless right now. So let the model run the well-trodden paths for you, and go and venture somewhere new.