What is AI slop?
Seems like every product team has coding agents building at least some of their UI these days. Agents are great at getting a working screen up quickly, but left alone they produce a very recognizable kind of interface. People call it AI slop (or agent slop, or UI slop, depending on who you ask). It's common enough that LinkedIn now has a "Seems Like AI Slop" button.
So what does it look like? You get a dashboard shell, a card grid, filler metrics, decorative gradients, and missing states. Add familiar sans-serif type and rounded cards nested inside more rounded cards and you get the idea. Each screen works fine by itself. It just looks like every other agent-built screen out there.
Why does this happen? A language model predicts the most likely next thing. Ask it for a settings page and you get something close to the average of every settings page it's seen. Nobody made decisions about this particular product or its audience, so the defaults win. One designer described slop as design produced from defaults rather than intent, which I think nails it.
There's a second flavor that's more about content than visual style. Agents are comfortable with text, so their screens fill up with explanatory paragraphs, navigation to everywhere, and labels that read like notes. They also like showing everything at once, which skips decades of thinking about progressive disclosure (revealing options only when people need them). That's what was going on in the share sheet comparison I mentioned: most of the UI was stuff the agent added.
Of course, slop didn't start with AI. When David Holz asked where AI learned slop from, my answer was Sturgeon's Law: "90% of everything is crap." A lot of code is not great period, and the fundamental job of making something good doesn't change with the tools.
Design guidance files and design systems that agents read before writing UI help. But you can follow every rule and still ship something generic, because someone has to decide what this specific product should be. In meetings where everyone shows up with their own AI prototype, a big part of the work is figuring out what was actually intentional versus what the model decided to put there.
So when you see slop, ask what someone actually intended, then cut everything else.