Automation

Estimates the cognitive frontier from commuting-zone wage responses under design-based inference, net of a college-wage trend that predates AI, and calibrates the physical frontier from firm-level evidence carried across its full range. Robots deliver about twice the productivity per unit of advance that AI does, and the ranking holds in every bootstrap draw. An AI advance lowers both wages and erodes the labor share more than twice as fast as a robot advance. A planner averse to inequality supports robot research about 70 percent more.

Develops a novel task-based framework showing why aggregate automation hides opposite skill-premium effects of robots and AI.

A two-frontier task model writes the reinstated mass as a reinstatement rate times the distance the frontier has travelled, so in logs the robot-to-AI gap splits into a rate term and a frontier-width term. Measured on two decades of occupational task data, frontier width carries most or all of it: across nine measures the implied relative rate straddles one, against the near five that a calibration of the same model imports. The reinstatement gap is mostly an exposure gap.