Website CV Paper
Fields
Labor, Industrial Organization, Innovation
Job market paper
Artificial Intelligence in the Firm: Bottlenecks in Software Production
We examine the impacts of generative AI on productivity, output, and employment in software engineering. Firm production requires combining complementary inputs, across which AI has differential productivity effects. Inputs that are difficult to automate can become “bottlenecks,” constraining output and shaping labor demand. We study these dynamics using a novel proprietary dataset from an engineering analytics platform, covering 445M work events — GitHub coding activity, Jira issues, and Google Calendar events — for 720,542 workers at 655 firms. We estimate the effects of two generations of AI coding tools — assistants and agents — by exploiting variation in timing of firm adoption. AI substantially increases code output but has smaller effects on final software output and no detectable effect on employment. Code review becomes the bottleneck. Code waits longer to reach a reviewer, spends longer in active review, and receives more requested changes. We estimate a model of firm production and AI adoption, then use it to simulate AI’s long-run impacts under continued firm adjustment and technological progress. The model predicts moderate increases in output and decreases in total employment, with effects of smaller magnitude if the bottleneck persists.