AI Keeps Stubbornly Refusing to Take Our Jobs
AI Keeps Stubbornly Refusing to Take Our Jobs
Noahpinion, September 7, 2026, by Noah Smith.
The argument
Published on Labor Day, Smith’s thesis is blunt: AI is not displacing workers, and the data bear this out. Prime-age employment (ages 25–54) sits near all-time highs even as AI adoption has accelerated. The fear of mass technological unemployment hasn’t materialized.
Smith applies the Acemoglu & Restrepo framework — a standard lens for thinking about automation’s labor effects — which splits the story into three effects:
- Displacement effect — automation substitutes for labor in specific tasks, reducing wages and employment.
- Productivity effect — automation makes production cheaper, raising demand and output, pulling employment back up.
- Reinstatement effect — new tasks and industries created by the technology absorb the displaced workers.
For AI so far, the productivity and reinstatement effects appear to be dominating. One key finding he cites: companies that adopt more AI tend to hire more humans, not fewer. The mechanism is that AI expands what a business can do, raising demand for the complementary human labor needed to capitalize on that expansion.
Key data points
- Census Bureau AI survey — one of the few large-scale surveys on actual AI deployment, not just sentiment. Shows AI adoption rates across industries.
- Kharazian et al. (2026) — academic paper with firm-level evidence on AI adoption and employment outcomes.
- The Economist AI jobs data — large cross-country dataset on AI-exposed occupations and measured employment changes.
- Hartley et al. (2026) — paper on the gap between job loss fears and actual job loss.
The key distinction
“Replacing tasks isn’t the same as replacing jobs.”
A job is a bundle of tasks. When AI automates one task within a job, the worker shifts time to other tasks — often tasks where human judgment, relationship-building, or physical presence matters more. The job doesn’t disappear; it reshapes.
Why this matters for the labor debate
The Acemoglu school (his 2022 book Power and Progress, written with Simon Johnson) has been the most influential articulation of AI-pessimism on labor. Smith is a sympathetic reader but pushes back on the empirical claims. The fact that we are three-plus years into mass AI deployment with employment still strong is evidence that needs to be taken seriously before further legislative or regulatory action is taken.