The Ballet and the Beige Paint

I’ve been quietly chipping away at my non-fiction book for a while now. It’s a long project, but I didn’t find it particularly scary. I’m a Product Manager by trade, and lately I’ve been using Claude Code to build software projects from scratch. My brain is currently wired to look at big tasks and think: This is a system. I can build this.

So I decided to treat the book exactly like a software build. I wanted to see if I could be the “manager” and let the AI be the “writer.” I would provide the detailed outlines, the source material, the direction. The model would do the actual drafting.

For a while, it seemed to work.

Then I reached the chapter on “Complex vs. Complicated” systems. This is one of the core ideas in the book, the distinction between systems you can engineer and systems that emerge on their own. I needed to explain how a city street is a “complex” system, not a “complicated” one.

I fed my detailed outline into the model and asked it to explain the concept.

The result was grammatically flawless. Structurally perfect. Here is what it produced:

“Urban environments function effectively when there is a structured interaction between pedestrians and infrastructure. The street life is dynamic, with different people performing different tasks, creating a sense of organized diversity that keeps the city safe.”

Go back and read that sentence again.

It makes sense, right? It’s accurate. But did it spark a single image in your brain? Did you feel the street?

Now compare that to how Jane Jacobs described the exact same concept in The Death and Life of Great American Cities:

“The ballet of the good city sidewalk never repeats itself from place to place, and in any one place is always replete with new improvisations.”

“Structured interaction” vs. “The ballet of the good city sidewalk.”

One is a definition. The other is a picture you can see.

The AI gave me the average of all knowledge. It scanned everything ever written about cities, complexity, and urban design, then produced the statistical midpoint of that corpus. What it handed back was technically correct, defensible, and utterly lifeless.

Jacobs gave me the human insight. She watched people. She noticed something true about how they moved. Then she found a word, ballet, that made the invisible visible.

The AI gave me beige paint. Jacobs gave me the ghost in the machine.

This week, the 2026 State of the Game Industry report came out, and the data tells me I’m not alone in this feeling.

Negative sentiment toward generative AI in gaming has nearly tripled in two years, from 18% in 2024 to 52% today. But the more interesting finding is the split in who feels this way.

Among executives, only 41% view generative AI negatively. Among artists, the number is 64%. Among designers, it’s 63%.

Why the gap?

I think it’s because executives see the output. They see the “structured interaction,” the technically correct result, the efficiency gains. The makers, the people whose job is to produce the work, see something else. They see what’s missing. They feel the absence of the ballet.

One studio owner from Norway put it this way in the survey:

“We have no use for tools that create assets for us, because creating the game is what gives us joy.”

That quote stayed with me. It points at something the productivity framing completely misses. For the people doing creative work, the searching is the value. The struggle to find the right word, the right note, the right brushstroke. That’s not overhead to be optimized away. That’s the job itself.

That was the missing piece for my book experiment. Finding the word “ballet” wasn’t a logistical problem I needed to outsource. Finding that word was the work.

The model couldn’t find it because it doesn’t watch people. It doesn’t notice. It doesn’t feel the particular quality of a busy sidewalk on a Tuesday afternoon. It just averages what others have already noticed and serves back the consensus.

I’m not saying AI tools are useless. I use them every day. They’re extraordinary for certain tasks, for scaffolding, for first drafts of emails, for breaking through writer’s block at 2am. But there’s a line, and I think the gaming data is showing us where it is.

The people who make things can feel when the tool starts doing the making for them. And something in them resists.

Maybe that resistance is worth listening to.