I was wrong about AI
On the aesthetic backlash, and why it reads as costume
Dan Yosefy · August 5, 2026
For three months in 2023 I used AI badly. Then I quit. Then I spent seven months learning to use it properly. This is what I got wrong.
The pitch we were sold: language models would take the jobs, dull the thinking, kill the ambition, make everyone dependent on the machine. Some of the fear tracks. IP protection is under attack. Voice models and deepfakes replicate actual people, distort them, steal them. Real problem, real stakes.
The response I want to talk about is the aesthetic one. Fonts engineered to confuse image recognition. Processes that foreground the handmade, the scan, the painted, the visibly human. In individual work: interesting. As a movement: trend. Designers joining the momentum, dressing fear as taste. The work carries a political argument, or it's costume.
Meanwhile: illustrators, art directors, designers I respect are working. Heads down. Passion intact. Using the tools. They learned to prompt. They learned what the model needs. They treat the tool like a tool. That was the secret nobody was gatekeeping; it was just work.
Here is where I was wrong. Early on my standards collapsed. When the tool delivered 85 percent of the brief, I stopped rewriting the last 15. That is the trap. You drop the vision, or you adapt yours down to what the machine produced. I adapted down. For three months.
Seven months of practice later, the tool works for me. The revolution is real. It deserves acknowledgment. It deserves a fight on the terms that matter, no exploitation of the communities whose labor trained the models, respect for scarce compute and water, containment to the industries where the use is legitimate.
The fear was correct about some things. The aesthetic backlash mostly isn't.