The Aesthetics of Proof (Of Being Human)

//Set in late 2026, because that’s where we’re headed...

I was in a conference room on Hudson Street when I first understood what we had done to ourselves. Not a real conference room. A Zoom grid, which is what we’ve been calling conference rooms since COVID. Ten windows. Nine black squares with names. One woman, camera on, sharing her screen. She was showing us thirty images for a campaign. “Which ones feel most on-brand?” she asked. This was September. This was a Thursday. This was when I asked my teams to start adding three deliberate mistakes to everything they make.

Someone unmuted: “Can we see the prompts?”

Someone else: “What about the source training images?”

Then the CMO, who a month earlier had watched a brand-safety scandal vaporize forty percent of her company’s market cap, said: “I need to know which ones have human hours attached.”

The woman stopped sharing her screen. Started sharing a spreadsheet. Two columns. “AI-Generated.” “Human-Verified.” The second column had three entries.

“We’ll go with those three,” she said.

Later that day I opened Photoshop. A vintage version, pre-subscription, pre-AI. And spent 20 minutes pushing a logo a few pixels to the left. Then a few pixels to the right. I recorded my screen. I narrated my doubt. I saved seventeen versions. This is what Collins now calls “Certified Human.” It costs three times their standard rate. The logo ended up pretty much where the AI had placed it in six seconds. I billed for the hours. I billed for the doubt. Mostly, I billed for the ability to be sued.

I keep thinking about that endlessly memed scene from *The Devil Wears Prada*. The cerulean sweater. Miranda Priestly explaining to her assistant how that specific blue didn’t just happen. How it moved from Oscar de la Renta to department stores to the bargain bin where the assistant selected it, thinking she was above fashion. The assistant thought she was choosing. She wasn’t. The choice had been made for her by people in rooms she’d never enter. That was taste: invisible labor, invisible decisions, the accumulation of knowing why this blue and not that blue.

Now AI generates three hundred blues in the time it took Miranda Priestly to purse her lips. The system knows their Pantone numbers, their social media engagement rates, their psychological impact scores. But here’s what I’ve learned: we’ve become the assistant in the sweater, except worse. She at least wore the cerulean. We just document that someone, somewhere, once cared about it.

I also think about Yves Klein in 1960, suspending pure ultramarine pigment in synthetic resin, creating IKB. International Klein Blue. A color so specific he patented it. The AI can reproduce IKB perfectly: #002FA7. What it cannot reproduce is Klein’s conviction that this exact blue could make people levitate, his decision to sign his name to a color like it was a painting, the six years he spent developing it, the gallery show where he had naked women covered in IKB press themselves against canvas while a small orchestra played his “Monotone Symphony”: a single note held for twenty minutes. The AI workflow knows the hex code. It doesn’t know about the women, the symphony, the twenty minutes of one note.

But here’s what I didn’t tell the CMO: I prefer the AI’s blues. They’re cleaner. More focus-grouped. More defensible. When I choose them, I feel nothing, which is a relief. No cramping doubt. No 3 a.m. worry that the cerulean should have been azure. The machine chooses and I perform the theater of choosing. Miranda Priestly’s invisible labor has become my visible performance. I make the spreadsheet. I document the hours. I create what our teams now call “process porn.” Time-lapse videos of humans making decisions, slowly, incorrectly, correcting them, making them incorrectly again. Clients watch these videos the way people once watched craftsmen sweat as they huffed and puffed blowing glass. As proof that suffering occurred.

The CMO’s daughter isn’t studying art history at Brown like her older sister did. She’s studying a 6-syllable specialty that kind of rhymes with semiotics. Something about artificially accelerated evolution of prompt architecture engineering. She tells her parents it’s art history because that still sounds human. She can generate a perfect Tintoretto in twelve seconds. She can’t explain why Tintoretto spent four years on the Scuola Grande di San Rocco ceiling, painting in the dark, ruining his eyes. She doesn’t need to. The explanation is the twelve seconds.

There’s a designer I know who’s gotten rich adding what he calls “human signatures.” Deliberately sloppy kerning, colors that don’t quite match, the kinds of mistakes that prove a person was tired, was hungry, was thinking about not thinking about their divorce while adjusting the blue. Anthropic have started training models on his mistakes. The mistakes are now perfect. Too perfect. So he makes new mistakes, uglier ones. This is what we used to call an arms race. This is what we call taste.

Adobe, three months before Figma acquired them, released “Creative Chain” certificates. Every pixel comes with metadata: was it generated, edited, or chosen? Who chose it? When? Why? For how long did they consider other options? The certificates take longer to read than the image took to make. Spotify launched “Artist Hours Verification.” There’s a browser extension that shows you the “Human Labor Score” of every website. My own website scores a 3. I made it in an afternoon. The documentation of my making it took four weeks.

I’ve started lying on the forms. I say I agonized for hours over blues I chose in seconds. I create fake version histories. I perform struggle I didn’t feel. Other designers do this too. We trade techniques. How to make your doubt look authentic. How to make your process look tortured. How to make your choices look chosen. We’ve become method actors playing ourselves, but worse versions. Versions who care about cerulean the way Miranda Priestly claimed to.

Last week a nervous client asked me to prove that I personally selected every color in a supervised, shot-by-shot grade. Not that they were good colors. Not that they worked. Just that I, a human who could theoretically be fired, had selected them. I showed him an hour of screen recordings dialing in the art-directed look. My hand moving the cursor. My voice explaining each choice. What I didn’t tell him: I was watching K-dramas on Viki on my other monitor. I was choosing colors based on what happened in the show. Blue when someone cried. Red when someone lied. The client loved the documentation. He said he could feel my intentionality.

The test has become simple. When it takes longer to prove you made something than to make it, you’re not an artist. You’re a notary. You’re not creating. You’re certifying. You’re not choosing. You’re performing choice.

But there’s something else, something I understood in that Zoom conference room. The CMO who insisted on human verification. She couldn’t tell the difference between the AI images and the human ones. Neither could I. They’re pixel perfect copies. We chose the human ones because they came with receipts. We chose them because somewhere, someone could be blamed. We chose them the way people choose organic vegetables when ordering grocery delivery. Not for the taste but for the story, for the ritual of caring about something we can’t actually discern.

I think about Klein’s models, covered in blue, pressing themselves against canvas. I think about that orchestra, holding one note for twenty minutes. I think about the audience, watching, believing this meant something. I think about Miranda Priestly’s cerulean, how it trickled down, how it was never really a choice.

Now I watch my cursor move across the screen. I watch myself pretend to choose. I watch the documentation accumulate. The proof that I existed, that I doubted, that I picked this blue and not that blue for reasons I can no longer remember or never had. I watch myself become the receipt.

This is what we talk about when we talk about taste: not the thing itself but the performance of having chosen it. Not the blue but the metadata. Not the work but the proof of work.

The CMO’s daughter sent me her portfolio last week. It’s all prompt-architectured, every piece. But she’s added something. She’s created fake documentation for each piece. Fabricated screen recordings of her “making” them with Kling 5.1, Pika 6 and the latest open-source Tencent HunyuanVideo community model. Invented version histories. Written fictional artist statements about her agonizing over every choice. The documentation is more creatively visceral than the work. The lies are more human than the truth.

I hired her immediately.

That’s what we talk about when we talk about taste.​​​​​​​​​​​​​​​​