You’re standing in line at a coffee shop, waiting on an oat milk latte, and your eyes drift to the corkboard by the door. There’s a flyer for an open mic night. The lettering is soft and glowy in a way lettering shouldn’t be. The guitar in the illustration has one string too many and a headstock that melts into the guitarist’s sleeve. The word “Thursday” is spelled “Thrusday.” Somebody printed this. Somebody pinned it up. Somebody thought it was fine.
You’ve seen a hundred of these by now. Restaurant windows, community centers, church bulletin boards, your neighborhood Facebook group. The look is unmistakable: over-lit, over-rendered, vaguely plastic, with text that falls apart if you read it twice. Readers have been sending these to journalists in bulk, and the collections are genuinely rough. One writer described the current moment as a “ChatGPT flyer pandemic.” The most interesting part of that critique wasn’t about AI at all. It was simpler than that: regardless of how these posters were made, they’re just ugly.
Why the flyers look like that
I explain AI agents for a living, and this is the example I keep coming back to, because it makes something abstract very concrete. An image model doesn’t know what a poster is for. It doesn’t know that a flyer’s entire job is to make a stranger, walking past at normal speed, absorb one piece of information. It knows what posters tend to look like statistically, which is a different thing entirely.
So you get the visual vocabulary of design without the decisions. Gradients where there should be contrast. A dozen focal points instead of one. Decorative flourishes crowding out the date and time, which are the only two things anyone actually needs. The hands are a punchline, but the hands aren’t the real problem. The real problem is that nothing on the page is prioritized, because nobody prioritized it.
This isn’t a hypothetical failure mode either. When a poster created with AI won the Ohio State Fair’s poster contest, some Ohioans pushed back, and a closer look at the winning entry turned up the same kinds of image errors that show up on the corkboard at your coffee shop. Those errors got past a judging process. They’re subtle enough to slide by a quick glance and obvious enough to bother anyone who stops.
The people cleaning up
The more hopeful development is the designers who’ve started taking these flyers and fixing them. It’s being framed as AI versus humans, with humans winning, and that framing is fun, but I think it undersells what’s actually happening. What the redesigns demonstrate is which part of the work the machine skipped.
Watch one of these before-and-afters and the fixes are rarely exotic. Pick one image and let it breathe. Set the headline in a typeface you can read from across a room. Put the date somewhere a person’s eye lands first. Cut three of the five decorative elements. Choose two colors. The output looks dramatically better, and almost none of the improvement came from rendering skill. It came from someone deciding what matters and removing everything else.
What this says about AI agents generally
Every AI tool has this same shape, and posters just happen to make it visible. A model can produce something that has the texture of finished work while missing the judgment that makes work good. Text that sounds like a report but doesn’t argue anything. Code that runs but solves a slightly different problem. A flyer that looks like a flyer but doesn’t tell you when the show starts.
Which is why the useful question isn’t whether to use these tools. It’s what you’re asking them to do. Generating a background texture, trying twenty layout variations quickly, mocking up a rough idea before you commit to it — fine, that’s real help. Handing over the entire decision of what this thing is supposed to communicate, then printing whatever comes back — that’s where the six-fingered guitarist comes from.
A short checklist before you print
- Read every word out loud. Typos in AI-generated text are common and easy to miss when you’re scanning.
- Look at hands, text on objects, and repeated patterns. That’s where the seams usually show.
- Squint at it from six feet away. If you can’t tell what the event is, the design failed regardless of how it was made.
- Count your focal points. More than one is a problem.
- Check that the date, time, and location are the easiest things to find.
Human creativity is still what’s being valued here, and the poster mess is a decent argument for why. AI-generated posters don’t have to be horrible. They’re horrible when nobody makes the calls that a poster requires. That job is still open, and it always was.
🕒 Published: