AI Destroys the Old Learning Curve. Wright's Law Is Being Rewritten (www.wsj.com)

AI is rewriting a foundational industrial rule: Wright’s Law, which held that unit costs fall predictably as cumulative production doubles, is being upended because experience no longer needs to come from making things. Advanced simulation, model-based optimization, digital twins, reinforcement learning and large pre-trained models let teams run millions of virtual experiments before a single physical unit is built. That means the traditional learning curve doesn’t just steepen—it can collapse: design and reliability improvements occur at near-zero marginal cost, shifting the primary source of economy from physical scale to compute, data and model scale. For AI/ML practitioners this is profound. Techniques like synthetic data generation, transfer learning, and offline RL accelerate iteration and reduce dependence on costly prototyping and field trials, enabling rapid productization and lower barriers to entry. But it also reframes risk and measurement: gains realized in simulation can fail under distribution shift, so robust validation, domain adaptation, and safety/regulatory testing become critical. Economically, investment and competitive advantage move toward compute infrastructure, simulation fidelity and data quality rather than traditional manufacturing throughput—forcing engineers, managers and policymakers to rethink metrics for progress, validation and responsible deployment.