The Shrug
Ship something built with an AI agent and sooner or later someone looks at it, shrugs, and says "cool, but I could have done that in a weekend."
They're probably right. They also didn't.
That dismissal is worth sitting with, because it's familiar in a way that's easy to miss at first. Then it clicks: it's the exact reaction people have to modern art. "My kid could have painted that." "I could have taped a banana to a wall." "That's just a urinal."
True. None of them did, though.
The Banana and the Urinal
In 2019, Maurizio Cattelan duct-taped a banana to a wall at Art Basel and called it Comedian. It sold for $120,000. The internet did what the internet does — collective eye-roll, "I could have done that," the art world is a scam, etc.
The technical craft of taping a banana to a wall is approximately zero. A toddler can execute the piece. That was kind of the point.
A century earlier, in 1917, Marcel Duchamp bought a porcelain urinal, signed it "R. Mutt," and submitted it to an art exhibition under the title Fountain. He didn't make the urinal. He didn't modify it. He chose it, signed it, and pointed at it. That gesture is one of the most influential moments in 20th-century art.
Same critique then as now: "I could have done that."
True. Nobody did.
Why the Critique Misses
The "I could have done that" reflex assumes the only thing being valued is the craft — the technical skill of execution. If craft is what's rare, the critique lands. A 14th-century altarpiece with a thousand hours of gold leafing and tempera underpainting? Nobody's saying "I could have done that," because nobody could.
But Duchamp's whole point was that craft wasn't the only thing that made something art. The choice mattered. The framing mattered. The willingness to put it forward and say "this is the thing" mattered. He democratized execution and pushed the value to judgment.
That's exactly what AI is doing to software, writing, image-making, music. The execution layer is collapsing. What used to take a senior engineer two weeks now takes a coding agent and twenty minutes of good direction. What used to take a graphic designer four hours takes a prompt and three iterations.
When execution gets cheap, the rare thing becomes everything around the execution: taste, judgment, knowing what to make, knowing when to stop, knowing what's worth shipping.
"I Could Have Done That" Is a Compliment in Disguise
Here's the part that took me a while to internalize. When someone says "I could have done that" about my AI-assisted work, they're not insulting me. They're acknowledging that the output looks achievable. That's the goal.
If my work looked like impenetrable wizardry, that would mean I'd over-engineered it or chosen a problem nobody understands. If someone looks at my fifty-line bot proxy or my eleventy template or my agent eval grid and thinks "yeah, that's not that hard" — good. That means I picked a clean problem and shipped a clean solution.
The honest response to "I could have done that" is: "Cool. Show me yours."
This is where the conversation usually ends. Because it turns out picking the problem, making the choices, and actually finishing the thing is the entire game. Anybody could have. Almost nobody does.
Where the Defense Falls Apart
I want to be honest about this. The "modern art defense" of AI work has real limits.
Duchamp's urinal worked because there was a clear authorial choice behind it. A vision. A point being made. The piece wasn't valuable because anyone could have made it — it was valuable because the person who did make it was making a specific argument about what art is.
A lot of AI output is not that. It's slop. There's no authorial choice, no point being made, no taste in the curation. Someone typed "write me a blog post about productivity" into a chatbot, pasted the output, and hit publish. That's not the urinal. That's spam.
The distinction matters: the "I could have done that" defense only works if you actually made choices. Did you pick a problem worth solving? Did you decide what to keep and what to throw away? Did you ship something with a point of view? If yes, the dismissal is a compliment in disguise. If no, the dismissal is correct, and you should listen to it.
What Actually Deserves Critique
If we're going to evaluate AI-assisted work seriously — and we should — the questions are not "did a human type every character?" The questions are:
- Did you choose a real problem? Most AI slop solves nothing, because nobody asked whether the output was worth producing.
- Did you make the cuts? AI generates infinitely. The taste is in what you threw away, what you reworked, what you said "not yet" to.
- Did you ship the thing? Generation is cheap. Finishing is the work. Most AI projects die in a tab somewhere because nobody had the will to push through the boring last 20%.
- Does it have a point of view? The bar isn't novelty. The bar is whether there's a person behind it making a claim about the world.
None of those questions are about the technical craft of typing the code. They're about judgment, taste, and follow-through. That's what gets evaluated for any creative work, AI-assisted or not.
The Nix Parallel
I wrote a while back about how my coding agent writes better Nix than I do. The easy shrug at that is the same one: "well then you're not really a Nix developer."
Cool. I'm still the one who decided we're using Nix. I'm still the one who designed the system. I'm still the one who knows what reproducibility means for this project and what trade-offs we're willing to make.
The agent handles the syntax. I handle the choices. If you think the choices are easy, you've never had to make them at scale, on a deadline, with consequences. Go ahead. Show me your config.
The Real Question
Whenever I catch myself dismissing someone else's AI-assisted work with "I could have done that," I try to ask the follow-up out loud: did I, though?
Usually the answer is no. Usually I didn't even try. Usually the person who shipped the thing had the taste to choose the problem, the judgment to make the cuts, and the spine to put their name on it and ship.
That's the work. The typing was always the easy part — AI just made that obvious.
Try This Today ⚡
Next time you find yourself shrugging at someone's AI-assisted work — or worse, hesitating to ship your own because you're afraid of the shrug — run this three-question check:
- Did they (or you) choose a real problem? If yes, that's a judgment call AI didn't make. Credit the human.
- Did they (or you) make cuts? Look at what's missing, what's tight, what's been refined. Curation is invisible but it's the entire skill.
- Did they (or you) finish? A shipped thing beats an unshipped thing every time, regardless of who or what typed it. Finishing is the rarest skill of all.
If you're on the receiving end of the shrug, just smile and ask: "Yeah, totally. Could you show me yours?"
Modern art's answer to "I could have done that" was "you didn't." AI's answer is the same. Ship the thing. Sign your name. Let the work speak.
Header image by Mike Dorner on Unsplash
Content on this blog was created using human and AI-assisted workflows described here. Original ideas and editorial decisions by Justin Quaintance.