"The Simple Macroeconomics of AI" is a working paper by Daron Acemoglu (MIT), issued as NBER Working Paper No. 32487 in May 2024 and published in Economic Policy 40(121), 13–58 (2025). It uses a task-based macroeconomic model to estimate an upper bound on AI's aggregate productivity gains, arguing they are smaller than prominent bullish forecasts and that AI is unlikely to reduce labor-income inequality.
Summary of argument
Acemoglu builds a task-based macroeconomic model to evaluate claims about AI's aggregate productivity and welfare implications. Taking estimates of which tasks AI automates and the task-level cost savings involved, and applying Hulten's theorem, he derives an upper bound on AI-driven total factor productivity (TFP) growth.
The paper frames its contribution as a methodological correction. Acemoglu argues that most bullish forecasts ignore task-level cost savings and complementarities and mechanically translate "jobs affected" into aggregate gains, which he contends is not how GDP actually moves.
Key findings
The paper's central estimate is a TFP gain of no more than 0.66% over 10 years, which Acemoglu characterizes as "nontrivial but modest" and well below bullish estimates from Goldman Sachs, McKinsey, and Erik Brynjolfsson. Adjusting for the observation that early AI evidence comes from easy-to-learn tasks while future gains would come from hard-to-learn tasks, the estimated TFP gain falls below 0.53% over 10 years.
On distribution, Acemoglu argues that AI is "unlikely to increase inequality as much as previous automation technologies" because its impact is more evenly distributed demographically, but adds that there is "no evidence that AI will reduce labor income inequality" either. He estimates that AI will widen the gap between capital and labor income. The paper also flags that some AI-created tasks may carry negative social value, citing algorithms for online manipulation as an example; this is presented as a category rather than a quantified effect.
Reception and relation to other sources
Within the macroeconomic debate on AI's effects, the paper takes the lower end of productivity estimates and serves as a counterweight to Autor's "Applying AI to Rebuild Middle-Class Jobs" (2024). Acemoglu and Autor have co-authored elsewhere but adopt different framings on this question. Acemoglu's claim is unconditional and macroeconomic: the upper bound on aggregate TFP gains is small, inequality is unlikely to be reduced, and the capital-labor split widens. Autor's claim is conditional and micro-structural: AI could augment middle-skill workers and restore the hollowed-out middle of the labor market if used well, on the argument that AI can extend expertise. The two are not strictly contradictory, since Autor's mechanism could operate even under Acemoglu's TFP ceiling, but they differ on the dominant narrative, with Acemoglu cautioning against overclaiming and Autor urging policy to shape AI toward augmentation.
The paper's estimates also sit below the more bullish task-level productivity gains reported in Generative AI at Work and the firm-level results described in real-world agent deployments (Source: oneusefulthing.org).
Relationships
- supports: AI and Productivity, AI Labor Disruption
- contradicts: Applying AI to Rebuild Middle-Class Jobs (framing), bullish macro forecasts referenced in Stanford HAI AI Index Report 2026
- related: Generative AI at Work, Generative AI and the Nature of Work, (Source: epoch.ai), Labor Disruption Timelines: Who Predicts What and Why