The videos
You've seen them. "I built a whole website with AI in five minutes." "This brand identity took sixty seconds." "Designers are finished." The thumbnail is a shocked face, the video is a screen recording, and the result is... fine. A hero section, three cards, a gradient. Something that looks like a website the way a stock photo looks like a memory.
What the videos leave out is the part that matters. Nobody asked who it was for. Nobody threw out the first ten versions. Nobody looked at it the next morning and felt that small, quiet no. The video ends where the work begins.
Before the screen, there was glue
I want to go back a bit, because we've been here before.
Before computers, graphic design was a physical trade. Type was set by a typesetter and came back on strips of paper. Designers cut it apart with an X-Acto knife, waxed or rubber-cemented it to a board and nudged it into place under a lamp. Headlines came from sheets of rub-down letters, one careful letter at a time. Color separations were cut by hand from a red film called rubylith. A single correction could mean an afternoon. If you wanted to try a layout three ways, you built it three times.
Then, in the mid-1980s, the desk changed. The Macintosh arrived in 1984, and a year later came Aldus PageMaker, Adobe's PostScript and Apple's LaserWriter: page, type and printer, all on one desk. People called it desktop publishing. Photoshop followed in 1990.
The fear was the same fear you hear now. If anyone can set type, what's a designer for? And for a while, the fear looked right. Every office suddenly had a newsletter with eleven fonts on it. Church flyers in drop shadows. Bevels on everything.
But something else happened too. A handful of designers picked up the new tool and made work that couldn't have existed before. April Greiman made a whole issue of Design Quarterly on a Mac in 1986, a life-size fold-out poster of pixels and type. Zuzana Licko and Rudy VanderLans built Emigre around fonts designed for the Mac's chunky screen, and leaned into the bitmap instead of hiding it. Older designers called some of it ugly. It was also the future.
The computer didn't replace designers. It replaced the glue. The people who learned it went further than the people who didn't, and the people who only had the computer made flyers with eleven fonts.
It's happened before that, too
When photography arrived in 1839, a French painter named Paul Delaroche supposedly looked at a daguerreotype and said that from that day, painting was dead. (Like most perfect quotes, it's probably not quite what he said.) Painting didn't die. It stopped needing to be a camera, and it went somewhere new: Impressionism, then everything after it. The machine took over the copying, and the people were freed to do the seeing.
Every one of these tools takes over a kind of labor. None of them has ever taken over the eye.
A brush, not a hand
Here's how I think about AI now: it's a brush. A remarkable one. It's the fastest sketchbook ever made, a pencil that can draw a hundred thumbnails before your coffee cools.
That speed changes how ideas feel. An idea used to have to survive the cost of making it. If a concept would take two days to prototype, it had better be worth two days, so a lot of strange, maybe-brilliant ideas died in the notebook. Now I can see a rough version in minutes. I can ask "what if the footer was a black hole?" and look at it before I've talked myself out of it. Ideas get to be tried instead of guessed.
And some things that used to be out of reach aren't anymore. This site has planets you can spin, a wormhole, a hidden boss fight with its own sound (the build is in How I hid a boss fight in a website). For a one-person studio, the hard part of something like that was never the idea. It was all the hours around it: filling in the CMS, tightening details for the tenth time, chasing a bug across three browsers at midnight. That's where AI earns its place on my desk. It takes the bulk work, so my hours go where they matter: the idea, the feel, the part nobody else would think to make. It didn't decide there should be a boss fight. It just helped make sure I had the time to build one.
That's the real gift. Not "it does it for you." It's "you get to find out."
The machine's favorite answer
There's something about how these models work that every creative should understand. They're trained on an enormous amount of what already exists, and they get very good at producing what's likely. Ask for a landing page and you'll get the most landing-page landing page there is. Ask for a logo and you'll get the center of every logo it's ever seen.
In other words, AI's natural instinct is the average. That's not a flaw, exactly; it's what it's built to do. But the average is exactly what makes work forgettable. (It's the whole problem in Less is a bore, and the trend cycle I wrote about in The Alchemist.)
Taste is the opposite move. Taste is choosing the unlikely thing on purpose, and knowing why it's right. It's the color nobody would pick, the pause before the headline, the joke in the corner of the 404 page. The model offers the probable. You choose the true.
Directing the machine is a skill
This is the part the videos really get wrong. Using AI well isn't pressing a button. It's a craft, and like any craft, there's a beginner version and a master version.
The beginner asks for "a modern website" and takes the first thing that comes back. The person who's put in the hours does something very different:
- They know what they want before they ask. A vague prompt gets the average. A clear point of view gets something with a spine.
- They can see what's wrong. The spacing that's a little off. The color that's almost right. The animation that's smooth but has no reason to exist. You can't fix what you can't see, and seeing takes years.
- They iterate like an editor. Keep this, lose that, push this further, start over. The first answer is a sketch, never a deliverable.
- They know where it lies. AI says wrong things with total confidence: code that looks right and isn't, facts that sound real and aren't. Checking is part of the job (that's Descartes' doubt all over again).
- They know when to take the pen back. Some moments need a hand, not a generator. The last 10% is almost always yours.
Photoshop didn't make anyone a designer. Knowing Photoshop made designers faster. AI is the same deal, just bigger.
What it can't hold
I want to be careful here, because nobody knows exactly where this technology goes. But I'd bet on this: the human eye, the human hand, the part that cares, isn't going anywhere.
Taste isn't a set of rules. It's built out of a life: the sites you loved at thirteen, the book on your shelf, the song that was playing the summer everything changed, the call that goes quiet when a client sees, for the first time, something that finally looks like them. Taste knows who it's for. It knows that this restaurant should feel like a grandmother's kitchen and not a tech startup, and why. It can feel when something is almost right, which is the hardest feeling in design, and it won't let go until it is.
A model can produce a thousand images of a sunset. It has never stood in one. It can't want anything. It can't be moved, and so it can't know what will move someone else. It can only guess what usually does.
That's why the work still needs a person. Not to push the button, but to care what happens after.
Faster than ever, still human
So: should creatives use AI? Yes. Learn it the way designers learned the Mac, with curiosity and a little suspicion. Let it take over the glue, the busywork, the hundred thumbnails. Let it show you ideas faster than you've ever seen them. Let it hand you tools that used to be out of reach.
Then do the part only you can do. Choose. Cut. Wait a day and look again. Say no to the obvious version. Put a little of your own life into it.
The brush has never been the artist. It's just never been this fast.
In practice
- Use it to see, not to finish. Sketch, explore, prototype. The final call is yours.
- Bring a point of view. The clearer your intent, the less average the output.
- Assume the first answer is the median. Push past it on purpose.
- Check everything. Code, facts, claims. Confident isn't the same as correct.
- Reach further. Try the thing that used to feel impossible. That's where AI earns its keep.
- Keep your hands in it. The last 10%, the detail nobody asked for, the feeling: that's the work.