Productivity Empowerment or Bounded Gains? How Work Nature Shapes the Productivity of Generative AI in Software Development
Type: article
Abstract: Generative AI (GAI) coding assistants can increase developer productivity, but their value in real-world workflows remains uncertain. This study investigates whether generative AI enhances productivity in software development and how workflow characteristics moderate this effect. Leveraging multi-source data from a large manufacturing enterprise, we combine a weighted difference-in-differences design with an instrumental variable analysis to estimate the productivity effects of GAI launch. We then use developer-day fixed-effects models to examine whether workflow characteristics moderate this relationship. The results show a robust productivity increase following GAI launch. However, these gains depend on workflow context. Project complexity significantly weakens the positive association between GAI usage and productivity, whereas current peer GAI usage strengthens it. Concurrent project volume and last peer GAI usage do not significantly moderate this relationship. This study shows how software-development workflows bound GAI productivity gains and offers guidance for work allocation, project design, and team coordination.
Language: English
Subfield: Information Systems
Field: Computer Science
Domain: Physical Sciences
Open Access status: green
OpenAlex ID: W7222907389