August 22, 2026
Scroll anywhere lately and you'll see the backlash. "No AI slop" in bios. Job posts specifying "no ChatGPT written cover letters." A whole vocabulary built around spotting it: the em dash, the "it's not just X, it's Y," the LinkedIn post that starts with "I used to think..." People are tired, and they've started saying so.
Here's the part that gets missed in most of that discourse. The problem was never the tool. Plenty of people use AI constantly and never produce anything that reads as slop. The difference between those people and the ones flooding your feed with generic listicles isn't the model they used. It's taste.
Slop is what happens when someone accepts the first draft. It's not that AI writing is inherently bad, it's that AI writing is inherently average. It's trained to produce the statistically likely next sentence, which means left completely alone, it drifts toward the most common way anyone has ever said a thing. That's fine as a starting point. It's a disaster as a finished product.
The people putting out slop skipped a step. They asked, got something plausible looking back, and shipped it. No read through with a critical eye, no "would I actually say it this way," no editing pass. The tool did exactly what it's built to do. The person just never applied judgment to the output.
Taste is that missing step. It's the thing that looks at a draft and knows it's flat, generic, or just wrong for the audience, even when nothing about it is technically broken.
This shows up constantly in writing, but it's just as real, and more expensive, when the thing you're building is software.
AI can generate a working feature in minutes now. It can scaffold a whole app. And a huge amount of what gets shipped by non technical builders using AI tools has the exact same problem as the LinkedIn slop post: it works, technically, but it's the most generic possible version of the thing. The default button style. The placeholder empty state nobody rewrote. The onboarding flow that reads like every other onboarding flow because the model pulled from every other onboarding flow.
Working code and good code aren't the same bar. A slop app passes the smoke test and fails the moment a real person tries to actually enjoy using it. Taste is what catches that gap, because taste is the only part of the process that knows what "good" looks like beyond "did it run without an error."
Taste isn't a mysterious gift some people have and others don't. It's closer to a set of habits.
You know what you're building before you ask for it. If you can't describe what "good" looks like in your own words, you have no way to recognize it when the AI hands something back. Vague brief in, vague judgment out.
You read output like an editor, not a customer. A customer asks "does this work." An editor asks "is this the best version of this," "does this sound like anything I've actually seen before," "would I be embarrassed if someone I respect saw this." That second pass is where most slop gets caught, if anyone bothers to make it.
You have references, and you use them. Good taste is rarely invented from nothing. It's built by paying attention to things that are actually well made, whether that's a competitor's product, a designer you admire, or a piece of writing that didn't sound like everyone else's. When you can point at something specific and say "closer to that," you give the AI, and yourself, something real to aim at instead of a vague adjective like "clean" or "modern."
You're willing to throw out the first answer. The first draft from any model is a floor, not a ceiling. If you stop there because it's "good enough," you've made the same choice as the person posting slop, just in code instead of words.
You know your own voice well enough to notice when it's missing. This is the fastest tell. Read a paragraph or look at a screen and ask: does this sound like me, or does it sound like the average of everyone. If it's the second one, that's the taste gap showing.
There's an argument going around that AI levels the playing field, that it makes everyone equally capable. It doesn't, not really. It makes execution cheap. It doesn't make judgment cheap. So the gap that used to be about who could execute has shifted almost entirely to who has the taste to know what's worth executing and when to say "not this."
That's actually good news if you're paying attention to your own taste instead of resenting the backlash. The anti slop sentiment isn't a sign that AI made things worse. It's a sign that the bar for using AI well just got higher, and the people willing to bring judgment to the process are the ones who'll stand out because of it, not despite it.
The tool got faster. It didn't get better taste. That part's still on you.