Coding Assistants and the Open-Source Labor Market

Abstract: Generative AI tools like ChatGPT have the potential to significantly reshape how individuals work, primarily through automating tasks and increasing worker productivity. Based on research on automation technologies like robots and computers, understanding both which tasks are most likely to be automated and which workers are most likely to see productivity gains is crucial for understanding these technologies’ potential impacts on income inequality. I focus on the latter of these two questions and study the introduction of generative AI coding tools on the productivity of open-source software developers. Leveraging these AI coding tools’ heterogeneous accuracy across programming languages, I measure these tools’ impact on productivity for high and low-skill developers. I find that high-skill developers see the largest changes in productivity in this real-world setting, in contrast to recent findings from laboratory settings and research that suggests that generative AI tools might be “inverse skill-biased”. While these tools do increase productivity generally, low-skill programmers are approximately 12.3% less productive than their high-skill counterparts after their introduction.

Link to the full paper here.