Dwarkesh Patel hosts economists Alex Imas and Phil Trammell for a rigorous, mechanism-heavy discussion on what actually becomes scarce after artificial general intelligence. The conversation rejects vague futurism in favor of specific economic reasoning. The central question: if AI can produce everything, what retains value? Their answer turns on the distinction between absolute and positional goods. Material abundance doesn't eliminate scarcity—it shifts it to goods whose value is inherently relative, like status, attention, and access. This has immediate implications for tax policy. Trammell argues that digital capital's frictionless mobility constrains governments far more than most realize: high capital taxes become unenforceable when capital is code that can reside anywhere. The optimal response is not a global wealth tax (which requires coordination that won't happen) but competitive institutional design to attract mobile tax bases. On labor, they push back against the 'capital eats everything' narrative. Even if AI substitutes for all productive labor, humans remain the consumers of positional goods—and someone must be the performer, the artist, the person whose attention is scarce. This preserves some labor bargaining power, though concentrated in winner-take-all markets. For non-AI-producing nations, the prescription is counterintuitive: don't rush to build domestic AI capacity. Instead, double down on rule of law, property rights, and regulatory predictability to become the jurisdiction where global digital capital wants to be domiciled. The episode closes with a sober assessment of transition dynamics—the real risk isn't the steady state but the speed of displacement, and economics has underinvested in modeling this gradient. Listeners will leave with a framework for thinking about post-AGI value capture that is grounded in mechanism design, public finance, and the economics of status competition.
Key Insights
- Labor's share of income may not fall to zero even under full automation because human attention and status-seeking create infinite demand for positional goods that AI cannot produce for you—someone still has to be the performer in a zero-sum status game.
- The optimal marginal tax rate in a post-AGI world could be lower than current top rates, not higher, because capital mobility constraints disappear when capital is digital, forcing governments into tax competition that makes wealth taxes and high capital taxes self-defeating.
- Non-AI-supply-chain countries should not panic-industrialize AI capacity but instead focus on institutional quality and property rights that make them attractive destinations for globally mobile digital capital, which becomes the primary tax base.
- Demand collapse from mass unemployment is unlikely because redistribution mechanisms already exist and can be scaled; the binding constraint is political will, not technical feasibility—and the relevant margin is consumption by the median voter, not aggregate demand.
- The transition period matters more than the steady state: the speed at which AI substitutes for different types of labor determines whether we get a smooth adjustment or distributional chaos, and current models of this transition are underdeveloped.
- Positional goods—things whose value depends on relative rather than absolute consumption—become the binding scarcity in a post-scarcity economy, which means inequality in status, not material deprivation, is the welfare-relevant problem to solve.
Who should listen: Macro investors and policy strategists modeling long-term capital/labor dynamics, tax base mobility, and country-level competitiveness under transformative AI scenarios.
Why This Matters
This episode directly informs how we model long-term asset allocation and country-level risk in a world where the capital/labor split is upended. The tax competition argument in particular challenges the default assumption that post-AGI states will easily fund generous UBI through capital taxation—if capital can flee frictionlessly, the institutional and jurisdictional layer becomes the binding constraint on value capture.
