Your product doesn't have a quality problem. It has an authorship problem.

You've felt it. The demo goes well. The screens are clean, the copy reads fine, the flow works. The person across the table nods along. And somewhere underneath the nodding is a small, unnameable off. They can't put a finger on it. Neither can you.

That feeling has a name now. It was Merriam-Webster's word of the year: slop.

Almost everyone uses the word wrong, though. They think slop means bad. It doesn't. Bad is easy to see and easy to fix. Slop is quieter, and much more expensive.

Slop isn't low quality. It's unowned.

A cleaner definition, and one that will change how you see your own product: AI slop is anything you shipped that nobody decided.

The word that matters is decided, not made. A thing can be built in seconds, polished to a shine, and still be a stack of defaults that no human ever chose on purpose.

Fine is the most dangerous word in product.

One test cuts through it. If a competitor could publish it word-for-word without lying, it's slop. Their onboarding, their empty states, their landing page: swap in your logo and nothing breaks. That fails a different test than quality. Call it authorship. Researchers even have a name for the look of it, “superficial competence,” a veneer of polish with nothing decided underneath.

Why your product drifts there on its own

The machine's whole job is to hand you the average.

Ask a model for a landing page and it returns the average of every landing page it has ever seen. Ask for a value prop and you get the center of the category. You can't prompt your way out of that. It's how the tool works: large models give you the statistical middle of whatever you asked for.

Sit with that for a second. A startup's entire reason to exist is to not be the average. You're a bet that the middle is wrong about something. So the tool that makes you fast is, by default, dragging you toward the exact center you started the company to escape.

There used to be a tax that saved us from this. It was called friction. Writing was hard, and the difficulty forced a decision. You couldn't finish the sentence until you knew what you thought. Take the friction away and the artifact now exists before the decision does. The doc is written. The screen is built. Nobody ever stopped to decide what it should be.

The machine version of this even has a name: model collapse. Train a model on the output of models and it slowly forgets the world, drifting toward a blander average of itself. A 2024 Nature paper showed it's often irreversible. We're doing the human version to our own products, one “good enough” ship at a time.

What it actually costs you

Slop doesn't crash. That's what makes it so expensive.

Your error rate is flat. Your dashboard is green. Nothing pages you at 3am. The cost shows up somewhere no chart is watching: in the gap between “it works” and “I trust it.”

It leaks exactly where you need conviction: the pricing page, the moment of commitment in onboarding, the upgrade screen. Average doesn't close. Then the slower bill arrives. Your brand becomes a tab, one of nine open, and not the one they remember. The best people you're trying to hire rule you out, because from the outside a company that ships defaults looks like a place that runs on prompts instead of judgment.

Call it trust debt. It compounds faster than technical debt, and it's worse, because technical debt at least shows up in a stack trace. Trust debt shows up in a retention number six months late, and by then you can't tell which decision caused it.

The scariest part is what has happened on the other side of the screen. The audience has already turned.

Careful, polished, human work gets branded slop all the time now, because polish itself reads as suspicious. Being seen as slop is a tax you pay even when you're innocent. This isn't a fringe grumble. The curl project shut down its entire security bug-bounty program under a flood of empty AI reports, and Linus Torvalds called the Linux security list “almost entirely unmanageable.” The flood is real, and the people you're trying to win already have their guard up.

The fix isn't less AI. It's more decisions.

The trap most teams miss is simple. AI now generates faster than you decide. Every surface it touches has a blank where a choice should sit, and the model fills that blank with the average. Your taste, sitting in your head or a brand doc, does nothing to stop it. Only decisions written down where the machine can read them do.

So measure the thing that actually predicts drift. Pull twenty live surfaces (empty states, error copy, the pricing page, onboarding, that transactional email) and mark each one Decided or Defaulted. The share you defaulted is your Slop Rate. Most teams sit at 60 to 80% and have never once looked. Track it like churn.

Then close the gap with two moves. First, write five refuses: falsifiable “we will never” rules, each with a reason. “No exclamation marks; we're the calm one in a panicked category.” A rule without a reason gets overwritten by the next prompt. Second, put a gate on everything that ships: a named decider, the refuse it honored, and one alternative you deliberately killed. No receipt, no ship. A high kill rate isn't waste. It's the sound of authorship.

Do that, and the tools rearrange themselves around it. The checklist hardens into an SOP so the standard survives your next ten hires. The generation step becomes an agent that only proposes work inside your refuses, never the internet's average. Those refuses become guardrails compiled into the pipeline, so drift can't reach production. That's the whole job. It's also, plainly, the job we do.

AI is the material. A human is the author. Slop is what you get when you forget which is which.

The next time it feels off

You'll be able to name it. So will they.

Your product can look like everyone else's. Almost all of them do now, and that's the whole problem and the whole opening. Or it can feel like yours: a series of small, deliberate decisions a real person made on purpose, in a direction the average would never pick.

That difference is authorship. It's a choice, and the only catch is that you have to make it on purpose.