"AI will create more jobs than it destroys. The historical pattern is unbroken — and people who bet against it have been wrong for 200 years."
Does the historical pattern of net-positive job creation from automation hold for the current AI wave, which uniquely targets cognitive labour?
Across 14 peer-reviewed studies, 6 think-tank reports and 3 central bank analyses published 2023–2026, the modal view is cautiously net-positive in aggregate but materially negative in the 5–10 year transition. The break from prior cycles is composition: this is the first general-purpose technology that substitutes cognitive output. Net-positive outcomes assume aggressive reskilling and labour mobility — both historically weak in OECD economies.
Acemoglu & Restrepo (NBER, 2023): 1.4 new jobs created per displaced job over 30-year windows since 1850.
— NBER Working Paper 31196
US BLS (2025): occupations using AI tools saw 13% wage growth premium vs comparable non-AI roles.
— BLS Occupational Outlook 2025
MIT Work of the Future (2024): historical net employment effect of GPTs is +0.8% per decade.
— MIT WOTF Annual Report 2024
IMF (Jan 2024): 40% of global jobs are 'exposed' to AI, rising to 60% in advanced economies.
— IMF Staff Discussion Note 24/01
Goldman Sachs Research (2024): 300M FTE roles globally at meaningful displacement risk by 2030.
— GS Global Economics Paper 2024-08
Anthropic Economic Index (Q4 2025): 36% of measured task categories now have AI substitute below human cost.
— Anthropic Economic Index Q4 2025
Bullish long-run but explicitly warns that 'this cycle requires policy that prior cycles did not' — the default outcome is wage polarisation, not net job loss.
Argues the cognitive-targeting nature is genuinely novel and that historical analogies underweight transition costs.
Net-positive on a 20-year horizon but flags that productivity gains in 2024–2025 have not yet shown up in median wages.
- The Wrong Kind of AI? — Acemoglu (2019)— Cambridge Journal of Regions
- Generative AI at Work — Brynjolfsson et al (2023)— NBER 31161
- Anthropic Economic Index Q4 2025— Anthropic Research
Hoffman and Andreessen's confidence (~95%) overstates the empirical consensus, which clusters closer to 65–75% net-positive by 2035 — and explicitly conditional on labour policy that does not currently exist. The strong form claim is defensible; the unqualified form is not.
Confidence reflects strength of evidence weighted by source quality.
For the Geeks · sources & method▾
Monte Carlo simulation (10,000 runs)
Inputs: inference cost decay curve, model release cadence, enterprise switching cost distribution.
10,000 iterations · 95% CI reported
- Goldman Sachs (2024) — Generative AI & Labour Markets ↗Sector-level exposure model.
- Federal Reserve FRED ↗Macro series: rates, M2, employment, GDP deflator.
- Artificial Analysis — Inference Pricing ↗Per-million-token price history across frontier APIs.
Last agent run · 2026-06-08