Burak Reis

Does AI kill creativity?

With agents I can produce any idea that crosses my mind within a few hours. The research says something similar: each person makes more creative work, but what comes out starts to look alike.

Burak Reis5 min read

In short

  • Stories by writers who got ideas from AI were rated more creative, but the stories resembled each other. A 2026 meta-analysis combining 19 studies finds this similarity small but consistent.
  • In images, average novelty dropped while the novelty of the most novel work went up. Sliding toward the average is not inevitable; it depends on how you use the tool.
  • People want to see effort. The same text scores lower once it is labeled "written by AI."

One night I couldn't sleep. We had been making something new with the agents, again and again. The next day another idea came to me, and the day after that, another.

Ideas used to come and go. I'd jot one down on the side, and most of them stayed there. Now I split every idea into branches right away: research, trial, plan and production. A few hours later I have something that works. The same happens when I spot a bug or a gap somewhere online: I can build the fix immediately. I want to enter every market, to be able to make everything I see.

As my capabilities grew, my sense of things started to slip. When I noticed that, I got curious: what is AI doing to my creativity? Am I more creative, or am I just producing more?

Each person more creative, the whole more alike

One of the clearest experiments on this question was published in Science Advances in 2024. Anil Doshi and Oliver Hauser had 293 people write short stories. One group wrote alone, one group got one idea from AI, and one group got five. The stories of those who got five ideas were rated 8.1 percent more novel and 9 percent more useful. The writers who gained the most were the ones who were less creative on their own.1

But there is another side to it. The stories of writers who got ideas from AI resembled each other more. Looked at one by one, each is better; put side by side, they are more monotonous.

And it isn't limited to one experiment. This year Alwin de Rooij and Michael Mose Biskjaer ran a meta-analysis combining 19 studies and found that work produced with AI becomes more alike. The effect is small (d = 0.334) but consistent. The authors read it not as a loss of creativity but as a reshuffling of creative diversity: it goes unnoticed in a single person and starts to matter when many people work with the same systems.2

I think this result describes my night exactly. Looking at my own work, I produce faster and better. But look at it together with thousands of people using the same tools, and all of us may be getting pulled toward the same average.

Is sliding toward the average inevitable?

Images show a similar picture. Eric Zhou and Dokyun Lee examined more than 4 million artworks and more than 50 thousand users on an art-sharing platform. People who started using text-to-image tools produced more, and their work was liked more. Average novelty fell over time. The novelty of the most novel work, though, went up.3

I think this distinction matters. AI pulls the average down, but it can raise the ceiling of the best work. Which side you end up on depends less on the tool than on what you do with it.

I lived through this distinction when I was writing down what I don't want into my CLAUDE.md files. On the first try, AI always puts the safest choice in front of me. The average is right there, in plain sight. My job is to recognize that average and decide where to break from it.

Making creates abundance, not attention

Music is a good example of what happens when making gets easy. According to figures Deezer shared in July, more than half of the new music uploaded to the platform in June was made with AI, around 90 thousand tracks a day. But those tracks get only 1 to 3 percent of total listening.4

So: half of the production, a few percent of the attention. My reading is this: as making gets cheaper, what is valuable stops being the making itself and becomes attention. And so the question "how much did I make" loses its meaning.

People want to see effort

There is also the side of the reader and the listener. Manav Raj, Justin Berg and Rob Seamans had more than 27 thousand people read the same texts across 16 experiments. When a text was labeled "written by AI," its score dropped. What explained the drop most was that the text was found less original.5

There is a cost on the making side too. In January, Anthropic published an experiment with 52 programmers: those who learned a new library with AI help scored clearly lower on the comprehension quiz afterwards than those who learned it by hand, 50 percent against 67 percent. The difference in time to finish the task was not significant.6 It is a small experiment, a single task. Still, it made me think: where I believe I am gaining speed, I may be losing understanding.

What changed for me

So AI doesn't kill creativity, I think. But left on its own it takes everyone to the same place, and does it very fast. The real question changed for me. I no longer ask "can I make this," because the answer is usually yes. I ask "is this worth making."

On November 15 I gathered all my projects in a single list and set myself a rule: at most three production jobs at a time. The rest waits on the list.

While writing this I noticed one more thing. My wanting to produce every idea immediately seems to have a chemical side as well. In the next post I'll look at dopamine and the thing called "dopamine addiction."

Notes

  1. Anil R. Doshi, Oliver P. Hauser, "Generative AI enhances individual creativity but reduces the collective diversity of novel content", Science Advances, July 12, 2024. Source ↩
  2. Alwin de Rooij, Michael Mose Biskjaer, "Generative AI makes creative output more homogeneous", ECCE 2026. Source ↩
  3. Eric Zhou, Dokyun Lee, "Generative artificial intelligence, human creativity, and art", PNAS Nexus, March 5, 2024. Source ↩
  4. Deezer, "AI music has surpassed 50% of new music uploads for the first time", July 21, 2026. Source ↩
  5. Manav Raj, Justin M. Berg, Rob Seamans, "The Artificial Intelligence Disclosure Penalty", Journal of Experimental Psychology: General, 2026. Via PsyPost ↩
  6. Anthropic, "How AI assistance impacts the formation of coding skills", January 29, 2026. Source ↩

Published on November 16, 2026.

Burak Reis designs brand identities and motion. About, portfolio.

Follow via RSS