this working paper from Harvard PhD students Fiona Chen and James Stratton is making the rounds, perhaps because it confirms many people's suspicions about the tradeoffs in vibe coding: LLMs mean lots more code written, but also lots more problems created, and relatively little productivity benefit
View full-size imageImage description from the author
Artificial Intelligence in the Firm: Bottlenecks in Software Production Fiona Chen Harvard University Job Market Paper James Stratton Harvard University
Read more of image description from the author
Current version: August 4, 2026 First version: January 7, 2026 Click here for current version Abstract We study the impacts of AI coding assistants and agents on software engineering work, using a novel proprietary dataset from an engineering analytics platform, covering 300 million work events — including GitHub coding activity, Jira issues, and Google Calendar events — across 718 firms. We use a staggered difference-in-differences design, exploiting variation in firmlevel adoption timing of AI coding assistants and agents. Both technologies increase coding productivity. However, productivity gains do not fully pass through to changes in software output or employment. For AI agents, this incomplete pass-through reflects a bottleneck from code review: review times increase, a larger share of code updates require revisions, and reviews involve more comments. We develop a model of software production to structure these results, in which AI affects both productivity and quality of intermediate outputs, and in turn can generate a review bottleneck and limit pass-through.
