Compute-Anchored Wage Theory for AI Agents
A new paper on arXiv challenges the intuition that AI agents drive cognitive-labor wages to zero due to infinite replication at zero marginal cost. The authors argue that agents are not labor but a production technology converting compute capital into effective cognitive labor. The elastic-supply margin shifts from labor to compute capital markets. They derive a Compute-Anchored Wage (CAW) bound: on substitutable tasks, human wages are bounded above by λ · k · r_c, where λ is productivity, k is capital per agent, and r_c is compute rental rate. The paper, arXiv:2605.05558, is a theoretical contribution to AI economics and policy.
Key facts
- Paper title: Who Prices Cognitive Labor in the Age of Agents? A Position on Compute-Anchored Wages
- Published on arXiv with ID 2605.05558
- Authors argue agents are not labor but a production technology
- Agents convert compute capital K_c into effective cognitive labor L_A
- Elastic-supply margin shifts from labor to compute capital market
- Derives Compute-Anchored Wage (CAW) bound
- CAW bound: human wage ≤ λ · k · r_c for substitutable tasks
- Framework based on textbook factor-pricing (Mankiw 2020)
Entities
Institutions
- arXiv