# The better AI gets, the smaller its share of the economy might get – Alex Imas and Phil Trammell

- Channel: [Dwarkesh Patel](https://www.youtube.com/channel/UCXl4i9dYBrFOabk0xGmbkRA)
- Published: 2026-06-04
- Source video: [Watch on YouTube](https://youtube.com/watch?v=Jj-kBHzUohs)
- Read on ReadTube: https://www.read.tube/p/videos/Jj-kBHzUohs
- Language: Original (source language)

## Summary

**AI Wealth May Not Stay Human**

The conversation argues that as AI and automation advance, the economy’s gains may flow increasingly to capital, compute, and automated production rather than wages or human services. The speakers stress that the key uncertainties are scarcity, demand elasticity, and whether humans remain intrinsically valuable in production or consumption.

The discussion explores how advanced AI could reshape wages, labor share, capital returns, and redistribution. The speakers argue that forecasting any one outcome is unreliable, so it is more useful to map scenarios: a world where human-in-the-loop services stay scarce, a world where labor share collapses, or a world where capital goods and compute absorb most of the gains. They repeatedly emphasize that economists currently lack the data needed to answer core questions about demand for human services, task-level automation, and which jobs are truly relational.

A major theme is that past automation often lowered the cost of goods and then increased demand for services, which preserved employment and kept labor share relatively stable. But they think AI may be different because it can automate entire supply chains, not just final tasks, and because some goods can become so abundant that their marginal value falls sharply. Even so, they caution that the overall capital share is ambiguous: if capital goods become cheaper while demand for them keeps expanding, capital income could rise a lot; if human-intrinsic goods remain special, labor could still retain value in some sectors.

They also debate political economy and redistribution. Faster AI may be easier to tax and redistribute than a slow “messy middle” where workers are displaced before the pie grows much, but they see many complications with UBI, negative income tax, universal basic capital, and wealth taxes. Their preferred practical advice is to think about indexing ordinary people and developing countries into AI-linked assets, since the gains may be concentrated in a few firms or, alternatively, broadly diffused like electricity. They end by stressing that the open-model question, the ease of indexing private AI firms, and the pace of commoditization will matter enormously for whether AI becomes a concentrated source of power or a broadly shared productivity shock.