The Best AI Labour Piece You'll Read This Month Has a Problem.

Clara Shih's essay on AI displacement and the China Shock is making the rounds, and it deserves more than a share. It also deserves more than a standing ovation.

The core argument is the best thing written in this cycle of AI labour panic. Shih borrows from Autor, Dorn, and Hanson's China Shock research — peer-reviewed, heavily cited, deeply uncomfortable reading — and makes the move that most AI commentators refuse to: you don't need AGI for serious disruption. You need 14%. That's it. A 14% displacement across white-collar sectors, achievable with models that already exist running on hardware that already exists, produces a shock roughly four times larger and five times faster than the manufacturing collapse that hollowed out entire regions of the United States and, incidentally, reshaped American politics for a generation.

The entry-level freeze observation is the best part. The China Shock's early signal wasn't mass layoffs on a Friday afternoon. It was plants quietly stopping backfill. Attrition doing the work. No announcement, no drama, just — gradually, then suddenly — no pipeline. Shih makes a credible case that this is already happening in knowledge work: new grad hiring in tech down over 50% since 2019, graduate underemployment at 42.5%, and — crucially — no 18-month rebound this time. The people who'll carry this aren't the senior engineers in the headline. They're the 23-year-olds who never get the junior role that was supposed to teach them the actual job.

So: the diagnosis is rigorous, the historical analogy is honest, and the credential devaluation section, applying Spence's signalling theory to AI hiring, is original enough to deserve its own essay.

Then the policy section arrives and the piece becomes a different document.

UBI. Wage subsidies. Public service programmes. Superannuation-style capital accounts. K-12 reform. Baby investment accounts. Each one a serious multi-decade political project, presented as a coherent response to a compressed five-year timeline. The piece criticises economists for telling displaced workers that markets will dynamically reallocate, and then commits a structurally identical error: asserting that political systems will dynamically intervene. The China Shock research that underpins the whole argument is actually devastating evidence against this optimism. Trade Adjustment Assistance — the policy mechanism designed for exactly this problem — was chronically underfunded and worked for almost nobody. If the analogy holds, the historical precedent for effective policy response is not encouraging.

The Lincoln-FDR paragraph is a tell. You reach for 1862 and 1933 when the present-tense argument isn't quite closing. Both of those interventions came after catastrophic visible damage had already accumulated — not in the window Shih is describing, where we supposedly still have time to act. The New Deal didn't prevent the Depression. It responded to it.

None of this invalidates the diagnosis. The piece is worth reading carefully — especially if you've been nodding along to the "markets will sort it out" rebuttal without sitting with the actual Autor et al. findings. The China Shock comparison is the right frame. Displaced manufacturing workers could suddenly buy cheap televisions. They were still worse off. A laid-off financial analyst whose fixed costs are a mortgage, private school tuition, and a car lease gets no structural relief from AI making software cheaper to produce.

The argument Shih is actually making — quietly, under the rhetorical ambition — is narrower and more important than the one she's trying to make: we got the China Shock badly wrong by assuming smooth adaptation, and we are about to repeat the mistake at four times the scale. That case is made. It doesn't need the FDR paragraph.

Knowing better and acting better are not the same thing. The China Shock is the proof.