Reimers and Waldfogel's 2026 study examined book publishing between 2022 and 2025. The finding: new book releases tripled. Average quality declined. More books, worse books. The pattern is a microcosm of what's happening across every domain where AI generates output.
This isn't an anti-AI article. AI is one of the most valuable tools ever created. But there's growing evidence that the way most people use it — outsourcing thinking rather than augmenting it — is making the humans less capable, not more.
Key Takeaway
AI doesn't make you dumber. Outsourcing your thinking to AI does. The difference: using AI to draft something you then edit with judgment is augmenting. Pasting AI output into a document without reading it critically is outsourcing. One makes you more capable. The other atrophies the skills that make you valuable.
What Does the Evidence Show?
| Study | Finding | Interpretation |
|---|---|---|
| Book publishing (Reimers/Waldfogel) | Releases tripled, quality declined | AI lowered the creation barrier but not the quality floor |
| Stanford "workslop" | AI content looks polished but lacks substance | Outsourced writing produces surface-level output |
| UC Berkeley | AI users multitask more, focus less | AI expands scope but reduces depth of thinking |
| ManpowerGroup | AI use up 13%, confidence in AI down 18% | People use AI more but think it helps less |
The Cognitive Atrophy Risk
When you outsource a cognitive task repeatedly, the underlying skill weakens. This is well-established in cognitive science — use it or lose it. If AI writes all your emails, your writing skill atrophies. If AI does all your analysis, your analytical thinking weakens. If AI generates all your ideas, your creative muscle shrinks.
The irony: the people who benefit most from AI are those who need it least. An experienced writer uses AI to draft faster, then edits with sharp judgment. The output is better because the judgment is informed. A novice writer uses AI to generate content they couldn't write themselves, then can't evaluate whether it's good because they lack the skill the AI substituted for.
Karpathy's warning at Sequoia AI Ascent: "When AI gets better, the temptation is to learn less. Understanding becomes the bottleneck." The people who stop learning because AI handles the output will discover too late that their judgment — the skill AI can't provide — has deteriorated from disuse.
---📬 Getting value from this? We write about the uncomfortable AI truths, weekly. Get it in your inbox →
---How Do You Use AI Without Losing Your Edge?
1. Do the hard thinking first. Form your opinion, make your plan, draft your outline BEFORE asking AI. Then use AI to refine, expand, and polish. Starting with AI means starting with average — starting with your own thinking means starting with something original.
2. Edit everything AI produces. Not light editing. Deep editing. Ask "is this actually what I mean?" for every paragraph. Change sentences. Remove sections. Add your own examples. The editing IS the thinking — it's where judgment develops and stays sharp.
3. Do some tasks manually on purpose. Write one email per day without AI. Analyze one dataset by hand per week. Draft one proposal from scratch per month. Deliberate practice of core skills prevents atrophy. You don't need to avoid AI — just don't let it replace all practice.
4. Use AI for the mundane, not the meaningful. Formatting, conversion, data cleanup, standard templates — these are perfect for AI because no judgment is required. Free tools handle these instantly. Save your thinking for the tasks where judgment matters — strategy, decisions, creative direction, problem framing.
5. When you use AI, use it well. Vague prompts produce vague output that you barely read. Specific prompts produce focused output that engages your critical thinking. The Prompt Optimizer helps write prompts that demand substantive output — the kind that makes you think about whether the AI's answer is actually right.
---📬 Want more like this? Honest AI analysis, not cheerleading. Subscribe free →
---Frequently Asked Questions
Is this anti-AI scaremongering?
No. Every study cited confirms AI creates real productivity gains. The concern is about HOW people use those gains — whether they invest freed-up time in higher-quality thinking or fill it with more low-quality output. AI is the most powerful cognitive tool ever created. Like any tool, it can be used well or poorly.
Does this apply to AI coding too?
Yes. Developers who accept all AI-generated code without understanding it build codebases they can't maintain. Developers who use AI to generate code they then review, understand, and improve get the productivity benefit without the skill atrophy.
Should students avoid AI?
Students should use AI after attempting work themselves, not before. The learning happens in the struggle. Using AI to skip the struggle skips the learning. Use AI to review, refine, and explore after the core thinking is done.
Disclosure: Some links in this article are affiliate links. We only recommend tools we've personally tested and use regularly. See our full disclosure policy.