AI makes the appearance of expertise cheap, but it makes earned judgment more valuable.
AI slop isn't proof that AI has nothing to say.
It's proof that many of the people using it didn't have much to say in the first place.
AI amplifies what's already there.
Give it someone's experience, judgment and an actual point of view built over years, and it'll help them process more, test more and say what they mean with greater precision.
Give it nothing underneath, and it'll do the same job on the nothing. It'll generate a framework, name it, explain it and wrap it in a polished carousel that reads like expertise to anyone who doesn't already know better.
That's the mechanism, and it isn't picky about what it amplifies.
A large-sample analysis of Common Crawl found that AI now writes as many online articles as humans do. The same researchers measured publication, not readership, and they suspect the AI-heavy share doesn't travel nearly as far as the human-written half. Volume isn't the same as reach, and this study can't tell us what any individual feed contains. But it does show how cheap publishing has become. We're producing as many AI-written articles as human-written ones, whether or not anyone finds them worth reading.
That distinction actually strengthens the argument. Generation has become abundant, but attention still has standards.
I should say plainly where I stand, because this isn't an argument for using AI less. I use it constantly, for research, for pressure-testing an argument before I commit to it, for finding the hole in my own logic faster than a person could find it for me. I think experienced people should be some of the most aggressive adopters of it, not the most cautious, because AI gives real experience more leverage than it's ever had. It can replace tasks outright, and it already automates work that used to require a person, letting smaller teams do what used to take a much bigger one. None of that is in dispute, but what's actually happening is narrower and more interesting than "AI will replace people" or "AI is just a tool." AI's output is shaped by what you bring into the exchange, and it can amplify judgment just as easily as it can amplify confusion and hand it back to you dressed as certainty.
prompts are downstream of thought
A good prompt helps, but prompts are downstream of thought. Before someone can ask a useful question, they need to already know enough to recognize what's missing from an answer, and enough judgment to push back on something that sounds plausible but isn't right for their situation. Prompting isn't a substitute for thinking, just a trace of the thinking that already happened, visible only after the fact.
There's a study that gets at this more precisely than intuition does, a five month experiment with 640 entrepreneurs in Kenya where half were given access to a GPT-4 business mentor. On average nothing happened, but underneath that null result was a real split. Entrepreneurs who were already performing well before the experiment saw real gains from the AI mentor, while the ones performing poorly got slightly worse. Here's the detail that matters, when the researchers checked what people actually asked the AI, the high performers and low performers asked similar questions, at similar length, and got similar advice back. The difference showed up later, in what each group did with the answer. Strong performers implemented in more specific, less generic ways, while weaker performers gravitated toward the same handful of obvious moves, like cutting prices or running more ads, regardless of whether that advice fit their situation. It wasn't the prompt that gave them away, it was the follow-through.
That's the honest version of "prompts are downstream of thought." It isn't that sharp people ask cleverer questions, it's that sharp judgment shows up in what you do with a plausible-sounding answer, which is exactly the moment most people skip.
what experience actually compresses
Over fifteen years building commercial systems in healthcare, I've written down a handful of ideas that hold up across companies. The Operator's Trap is one of them, the pattern where a leader gets promoted for skills the new role doesn't actually require, and nobody notices until the gap has already cost something. GTM Is Not a Strategy, It's a Claim is another, the idea that a go-to-market plan is only as good as the assumption underneath it that nobody bothered to test out loud.
AI could probably produce something that looks like those frameworks, given a decent prompt and enough iteration, but what it can't cheaply produce is the years behind them. Those two ideas came from watching a strategy work in one company and fail in another for reasons that took months to actually understand, from sitting in the room when a plan that looked airtight on a slide started to fall apart the moment it met a real customer. AI can reproduce the shape of that kind of thinking, and it can help me sharpen it, argue with it, and say it more clearly than I would on my own. What it can't do cheaply is manufacture the exposure to being wrong that made the idea worth having in the first place.
The illusion of explanatory depth is our tendency to believe we understand a system more deeply than we do until we're asked to explain how it works.
A tidy framework almost seems designed to exploit that weakness. It gives us a visible structure while hiding the mechanism underneath, and AI is extremely good at generating the visible structure. It isn't a reliable judge of whether there's anything real underneath it, and neither, it turns out, are most of us.
when experience becomes baggage
None of this lets experience off the hook, and I don't want it to. A widely cited review of health care research found that in more than half of 62 evaluations, quality of care actually declined the longer a clinician had been in practice, and only a small fraction showed improvement. Years on the job aren't the same thing as calibrated judgment, and someone can repeat the same mistake for fifteen years and still call the repetition experience.
AI complicates this further in a direction that should make anyone confident in their own judgment a little uneasy. In one large study, consultants using AI on a task outside the tool's real capability got the answer wrong more often than a control group working without it, yet independent reviewers rated their write-ups as higher quality even when the underlying answer was incorrect. The polish went up while the accuracy went down, and it isn't only novices who miss this kind of thing. In a separate study of experienced software engineers working in codebases they knew well, engineers using AI assistance were measurably slower, and afterward still believed the tool had sped them up. Fifteen years of pattern recognition didn't protect them from the illusion, and if anything, their confidence in their own read of the situation made them less likely to check it.
So I want to be careful here, because it would be easy to turn "slop" into a word that just means content I don't respect, and that's not an argument, it's a mood. The line isn't experienced versus inexperienced, or human versus AI-assisted. The line is examined versus unexamined. Experience only becomes judgment once it's been tested against outcomes, updated when it's wrong, and held loosely enough to notice when a comfortable pattern doesn't fit the situation in front of you. Absent that, fifteen years of reps and zero years of reps can produce the same failure, which is confident language wrapped around something nobody actually checked.
That's the real shift underneath all of this. When a capability gets cheap and available to everyone, the advantage stops living in the capability and moves to whatever that capability amplifies. For a company, that might be distribution, trust, or a proprietary workflow nobody else can copy. For a person, it's the judgment, curiosity, and track record they already have, sharpened rather than replaced. When someone has none of that underneath, access to AI doesn't magically supply it. It just gives the absence better production values.
The problem isn't that AI is replacing human thought.
It's that AI is revealing how often there wasn't much thought there to replace.
AI didn't create the emptiness in our feeds. It just gave it a content strategy.