Papers

AI agents can perform tasks, but unlike workers they can be copied at the price of compute. This paper asks what that does to the economics of production. In a task-based model, a firm can let agents perform a task only after implementing it: integrating the agent, redesigning the workflow and preparing the data. Implementation uses a distinct kind of labor and is a fixed cost per task, independent of scale. Four results follow. Automation depends on market size relative to organizational frictions, so AI-native entrants can out-automate larger incumbents. Cheaper agents raise the demand for implementation labor: the technology substitutes for the workers who perform tasks and complements those who deploy it. When implementation capacity is inelastic, a fall in the price of AI raises the implementation wage rather than the number of automated tasks, and its effect on unit costs is bounded by the current automation share, the Hulten term that existing macroeconomic estimates take as given. Finally, the same bottleneck shields production workers: displacement is triggered by the expansion of implementation capacity, not by the falling price of AI. Agents are abundant; what is scarce is the organization able to deploy them.

Does generative AI widen or compress productivity differences between workers? This paper develops a microeconomic model in which an AI tool has two separate margins: a frontier, the problems it can solve for anyone, and leverage, the extra work it lets a person get through. The gain from adoption is V-shaped in ability, with its minimum at the worker whose knowledge equals the AI’s frontier. Frontier advances are equalising and leverage is skill-biased. The average AI×skill complementarity therefore changes sign as the frontier moves through the ability distribution, and the conflicting experimental findings describe different points on the same curve. Relative inequality never rises with ability-neutral leverage, but inequality in levels can. When the frontier is jagged, the ability to judge AI output becomes the skill-biased margin, and over-trusting workers can lose. Adoption is two-tailed, so the sign of selection bias depends on the vintage of the technology. Workers capture a share of the gain that rises with diffusion across employers. The return to ability falls to zero below the frontier, which removes the incentive for juniors to climb rungs the AI already occupies. The paper closes with seven testable predictions and the designs needed to test them.

Generative AI changes the production function of software: every answer, document or completed task consumes inference compute, so software acquires a positive marginal cost that scales with how intensively each customer uses it. I derive the consequences in three steps. First, flat access breaks. Heavy users become the least profitable, flat pricing’s share of attainable profit falls monotonically with inference cost, and it collapses once a unit of inference costs half the value of the first unit of use. Second, and centrally, the unit of sale follows control of compute. When the customer controls usage, as with copilots, usage pricing disciplines consumption and pass-through of inference cost exceeds Borch’s risk-sharing benchmark. When the provider controls the process, as with agents, usage pricing is cost-plus, and the efficient contract moves towards paying for outcomes. Outcome pricing is worth its verification cost when the compute bill per task is large. Third, with constant inference cost scale economies vanish; they survive only through firm-specific learning in inference. Common falls in compute prices raise margins without raising gross profit, and because AI also lowers the cost of building software, free entry pushes each firm’s gross profit towards its entry cost. Financial statements of 307 US-listed software firms are consistent with these predictions. Early generative-AI adopters saw gross margins fall four to five points relative to other firms after 2023 while growing faster, and market value tracks gross profit more closely than revenue. The learning rate in inference cannot be identified from public accounts.

When a local housing market receives a population shock, the shock is absorbed through new construction, conversion of the existing stock, households consuming less space and the relocation of existing residents, and prices move until these margins exhaust it. I develop a spatial-equilibrium framework in which these shares have a closed form and sum to one, and in which housing supply elasticity governs both the price incidence of a shock and the share of it a place can absorb as population. I estimate the framework for Spain’s two immigration waves using a 2003–2024 panel of 50 provinces and a shift-share instrument based on settlement patterns by country of birth. In the current wave, each 1% of population arriving raises rents by about 1% and house prices by about 4% cumulatively. Prices outrun rents, which points to capitalised expectations. The construction response, which absorbed about 0.7 of each unit of inflow in 2003–2008, has fallen to zero, and the implied short-run supply elasticity is indistinguishable from zero. The current wave has instead been absorbed by more residents per dwelling. Inflows do not displace natives: Spain-born adults move towards receiving provinces, which amplifies rather than offsets the shock. The results identify supply elasticity as the state variable that separates the two waves. They also show that, when supply does not respond, population growth is accommodated by crowding rather than by building.