There's a genre of online writing you've seen a hundred times by now. It's the AI think-piece. It shows up on Medium, LinkedIn, corporate blogs, and newsletters with titles like "Why AI Won't Replace You (But Someone Using AI Will)". You scroll through it, nod along, and close the tab having learned nothing you didn't already know.
We need to talk about why that keeps happening.
Why AI Won't Replace Programmers — But Will Redefine What They Do
If you work in software, you've felt the ground shifting. Code generation, automated testing, AI-assisted debugging — tools that were research demos two years ago are now part of daily workflows. The transformation is no longer hypothetical. It's here, and it's accelerating.
Naturally, this raises the question everyone is quietly asking: are we automating ourselves out of a job? When an AI can scaffold an entire service from a paragraph of English, what exactly is left for us to do? The anxiety is understandable. Some make it sound like the profession has a five-year expiration date.
But the picture is more nuanced than the doomsayers suggest. AI is remarkably good at producing plausible code — and remarkably bad at knowing whether that code is right. It can't sit in a meeting and figure out that what the PM is describing and what the client actually needs are two different things. It doesn't make the judgment call that the spec is wrong and the feature shouldn't be built at all. These are human problems that remain stubbornly resistant to automation.
That said, it would be naive to pretend the role stays the same. The developer of 2030 probably spends less time writing and more time reviewing, directing, and integrating AI-generated output. The key skills shift: system-level thinking, clear communication with both humans and machines, and the ability to evaluate AI output critically rather than accept it at face value.
So what should you actually do about it? Start by getting comfortable with AI-augmented workflows. Learn prompt engineering as a genuine skill in specifying intent precisely. Integrate AI tools into your development and pay attention to where they help and where they hallucinate. The developers who build this fluency early will have a real advantage.
Because here's the bottom line: the developers who thrive in the next decade won't be the ones who wrote the most lines of code. They'll be the ones who understood what the code was for, and used every tool available — AI included — to build it well. The industry is changing. The ones who change with it will do just fine.
Sounds familiar, doesn't it?
The text above is AI-generated. It took one prompt and about four seconds. But here's the thing you should actually pay attention to: you didn't need AI to write it. A human could produce that article just as easily, because it follows a template so rigid that authorship is almost beside the point:
1. Hook
"AI is changing everything."
2. Fear
"Will programmers disappear?"
3. Reassurance
"Not exactly."
4. Pivot
"But the role will change dramatically."
5. Advice
"Learn prompt engineering / AI tools."
6. Inspirational ending
"The future belongs to those who adapt."
This is a fill-in-the-blanks structure. Swap "programmers" for "designers," "marketers," or "lawyers" and the article still works — which is exactly the problem. It "works" the way a stock photo of a handshake works. It fills a space. It communicates nothing.
The discourse around AI-generated content — "AI slop" — focuses on the machine side of the equation: low-effort synthetic text flooding search results, spam blogs, and social media. That's a real problem. But it has an older, human-made counterpart that nobody names.
Call it human slop.
Human slop is content that is technically written by a person, but follows such predictable formulas that it carries no more informational value than its AI-generated equivalent. It existed long before LLMs. The LinkedIn thought-leader post. The corporate keynote that says "innovation" fourteen times and commits to nothing. The trend article that restates conventional wisdom with enough confidence to pass for insight.
The template above didn't come from nowhere. AI learned it from us. LLMs are, among other things, a mirror trained on the largest corpus of human writing ever assembled, and what that mirror reflects back is how much of that writing was already slop. The formulaic AI think-piece is a copy of a human original that was already a copy of nothing in particular.
This is worth sitting with, because the usual framing of the AI-content problem is backwards. The fear is that AI will degrade the quality of online writing. But a significant share of online writing — particularly in the professional/thought-leadership space — had already degraded into template-filling before a single LLM was deployed. AI didn't create the problem. It made the problem visible by doing the same thing faster and at scale, which forced a question that was always there: if a machine can produce your article in four seconds, what was the article actually contributing?
Human slop persists for the same reason junk food persists: it satisfies a craving without providing nutrition. The craving, in this case, is for legibility — the feeling that you understand what's going on in a fast-moving field. A six-paragraph article that walks you from fear to reassurance to actionable advice feels like understanding, even when the advice is generic enough to apply to any profession in any decade ("adapt or be left behind").
It also persists because it's safe. Taking an actual position — "prompt engineering is a transitional skill that will be automated away within five years," or "most business applications of LLMs are currently net-negative on productivity when you account for error-correction overhead" — invites disagreement, demands evidence, and risks being wrong. The template avoids all of this. It gestures at change, hedges everything, and lands on a motivational platitude. Nobody will argue with "the future belongs to those who adapt." Nobody will learn from it either.
The antidote to slop — human or AI — isn't a matter of who writes it. It's whether the writing contains something that couldn't be predicted from the headline alone. That means specificity: actual numbers, named tradeoffs, documented failures, unfashionable conclusions. It means a willingness to say things that might be wrong rather than things that are unfalsifiable. It means writing that could not be generated from a six-line template, because the ideas in it are particular enough to resist compression into a formula.
None of this is new advice. Good writers have always known it. But the arrival of LLMs has made the distinction between substantive writing and template-filling into a practical, almost economic question. If your article is indistinguishable from what a model produces in four seconds, the market will eventually price it accordingly — regardless of whether a human wrote it.
The uncomfortable takeaway is that "AI slop" and "human slop" are not two separate problems. They are the same problem. The machine version is just cheaper.