My feed has become an art gallery. Threads, X, LinkedIn: every third post is an interface. A glassmorphic banking dashboard. A travel app with a hero image the size of a sail. A crypto wallet for a coin that doesn’t exist, lit like a perfume ad. And underneath, always, the same two captions: “What do you think?!” and “Should I drop the Figma?”
Sure. Drop it in the trash.
(I know. And to be clear, this isn’t aimed at the kids posting work to learn in public. They’re doing the right thing, and I’ll come back to them.)
Here’s the thing I notice every time: these interfaces are genuinely pretty. That’s not the problem. The problem starts about four seconds in, when you try to actually USE one in your head. Tabs that couldn’t survive a fifth item. A chart with no axes, plotting nothing against nothing. White type on a pale gradient that would fail contrast so hard the lawyers get involved. A search bar in an app that contains six things. A balance with no currency. Lorem ipsum sitting exactly where the hard decisions live.
It looks like an app the way a movie set looks like a town. The front is immaculate. Walk through the door and there’s no room behind it… really just lumber holding up the pretty.
That distinction is the entire job. A design is a stack of decisions made under constraint: real data, worst cases, fat thumbs, slow networks, angry humans, at least one law. A picture of a design has no constraints, so it optimizes for the only judge it’ll ever face, which is the scroll. Call it screenshot Darwinism. The feed selects for whatever stops a thumb. It has no mechanism, none at all, for selecting what survives a user, because in this ecosystem no user ever arrives.
Design made to be looked at instead of used isn’t a new disease. Dribbble ran that economy for a decade: unusable weather apps, calendars with no Tuesdays, an entire genre of beautiful software nobody would ever touch. But the old fakery at least cost something. You had to be able to draw the picture, and drawing it taught you type and color and composition on the way through. Now any model will hand you the picture in forty seconds. The disease had a price of admission, and AI waived it.
There’s a reason the models are so unreasonably good at this exact thing, and it’s almost too on the nose: they learned design from screenshots. Millions of them. Fifteen years of the internet posting its prettiest fronts. The model has seen what design looks like at a volume no human eye can match, and it’s never once seen what design survives. It never watched the support tickets come in. It never saw the layout meet a long German surname. It was trained on a museum of facades, so facades are what it builds.
And the loop is closing. Today’s generated dashboard gets posted, the post gets scraped, the scrape gets trained on, the training generates tomorrow’s dashboard, slightly more confident than the last one.
Look at what’s missing from that diagram. At no point on that circle does the work pass through a person trying to accomplish anything. The loop is airtight. That’s what people mean, or should mean, when they call this stuff a polished turd. The term has nothing to do with ugly; this work is usually gorgeous. A polished turd is work with an answer to exactly one question, and the question is “how does it look?”
If you want to tell a design from a picture of one, ask it boring questions. What happens when the name is forty characters. What does the empty state say. Where do errors go. What, exactly, does tapping the pretty button do. A real design has boring answers everywhere, because somebody had to live those questions. A picture has no answers at all, because nobody ever asked, because nobody was ever going to use it.
Now, the kids. Posting work in public is how you build an eye, and half of building an eye is doing the reps, so post away… I mean that. The rot isn’t the posting. It’s when the feed becomes the client. When saves are the metric and the scroll is the user, you can now go years, an entire career, without your work ever touching a person with a task. It looks like ten thousand hours of practice. It is the same four seconds, ten thousand times.
And none of this is anti-beauty, for the record. I’ve spent a whole piece on corrections so small they amount to nudging a triangle two pixels, so I’m not about to argue that pretty doesn’t matter. Beauty earns trust. But the order of operations is fixed: beautiful and functional is design. Beautiful instead of functional is a costume. And the machine, left alone, can only make costumes, because the thing that turns a picture into a design is a human on the other end, imagined so specifically that the work bends around them. The model doesn’t have one. Neither does the feed. That’s the part that’s missing, and no amount of gradient can make up for it.
So, honestly: yes. Drop the Figma. Drop it on one real person with one real task, and watch what breaks. Then post the wreckage and what you did about it. That stops my thumb every single time.
Until then… I stand by the trash.