# Why hasn't AI increased unemployment? > [!quote] > "Economic growth may be constrained not by what we do well but rather by what is essential and yet hard to improve" — Aghion, Jones & Jones (2017), quoted by McCrory Peter McCrory (Anthropic economic research) argues that US AI adoption is now large enough to produce measurable macro effects, that those effects are showing up in *productivity* statistics, and that they are conspicuously **not** showing up in unemployment — and offers a framework for why. ## Author's thesis AI so far behaves as a **skill-biased, labor-augmenting** technology. It automates tasks, not jobs; the residual non-automated tasks constrain the productivity lift and simultaneously raise the returns to the humans who handle them. So adoption raises output per hour without displacing workers. ## Key arguments 1. **The labor market is stable, so absence of effect is informative.** June unemployment 4.2%, openings/unemployed just over 1, prime-age employment-to-population near multi-decade highs, layoffs at or below pre-pandemic rates. Meanwhile ~20% of firms use AI in at least one function (40% in the information sector) and quality-adjusted AI output grew >2,000%/year in 2024 and 2025 — big enough that a displacement effect should be visible. 2. **Task automation ≠ job automation.** No occupation in the O\*NET taxonomy has *all* its tasks systematically handled by Claude. Anthropic's own usage research finds complex outputs correlate with sophisticated user inputs, that the model struggles most on the hardest work, and that Claude Code users with more domain expertise succeed more often and recover better from model errors. 3. **Weak links cap the aggregate gain.** Even under rapid, widespread but *incomplete* automation, the essential-and-unautomatable tasks bind the system. Evidence: a 10–20x increase in lines of code generated after coding agents produced only a ~30% increase in software releases, and no increase in app usage. 4. **The young-worker signal is real but confounded.** Weakening hiring for young workers in exposed roles is consistent with Stanford's *Canaries in the Coal Mine*, but AI arrived during the largest non-recessionary labor slowdown on record, plus post-pandemic remote-work shifts — hard to isolate. ## Caveats US-only scope, short observation windows, and a notable conflict of interest: this is Anthropic's economics team publishing on whether Anthropic's product displaces labor. Exposure measures are proxies for adoption, not adoption itself. McCrory is explicit that the framework may not survive further capability gains — if the jagged frontier smooths, the returns to human expertise can fall away. ## Links - [[3. Resources/Permanent Notes/Weak Links Cap Aggregate Gains Even Under Widespread Automation]] — the claim this note promotes: the binding constraint is what's essential and hard to improve, not what automates well - [[3. Resources/Permanent Notes/AI Amplifies Existing Bottlenecks Rather Than Creating Capability]] — the same shape one level down; the 10–20x code / 30% releases gap is a bottleneck being amplified, not relieved - [[3. Resources/Permanent Notes/AI Compresses the Knowledge-Worker Pipeline From Both Ends]] — the young-worker hiring signal is the entry-level end of that compression, which McCrory reads as macro-confounded rather than AI-caused - [[3. Resources/Permanent Notes/AI Tools Don't Eliminate the Apprenticeship Layer They Relocate Its Cost]] — where the returns to expertise persist, someone still has to become an expert; McCrory's framework doesn't say who pays for that - [[3. Resources/Permanent Notes/Perm · The middle is disappearing as AI commodifies previously-expert work]] — contrasts: this source finds *rising* returns to complementary expertise where that note finds a hollowed middle; the difference is which skill is priced - [[3. Resources/Permanent Notes/AI-Driven Reset Is the Acceptable Public Frame for Margin-Driven Layoffs]] — explains why headline "AI layoffs" and stable aggregate unemployment can both be true - [[3. Resources/Literature Notes/Lit · AI solipsists and AI cynics]] — direct counterweight from the same day: Doctorow argues the displacement is driven by investor incentives regardless of whether the tools work - [[3. Resources/MOCs/AI & the Future of Knowledge Work MOC]] — applies-in: the MOC lists this note under Labor market & economics. ## Source - [Reader: Why hasn't AI increased unemployment? | Peter McCrory](https://read.readwise.io/read/01ky8rcpvrc9kds4q1n04emz23) — originally [x.com/p_mccrory](https://x.com/p_mccrory/status/2014281227663524038)