Who is using AI to code? Global diffusion and impact of generative AI
Simone Daniotti presents research on the global spread of AI coding tools, analysing millions of GitHub commits to explore who benefits, how productivity changes, and what it means for skill gaps.
Abstract
Generative coding tools promise substantial productivity gains, but uneven uptake could widen skill and income gaps. The authors trained a neural classifier to identify AI-generated Python functions in more than 30 million GitHub commits by about 160,000 software developers and tracked how rapidly and where these tools are adopted.
They estimate that AI writes roughly 29% of Python functions in the United States, with this lead narrowing relative to other countries. Quarterly output, measured by online code contributions, increased by about 3.6%. Productivity gains primarily accrue to experienced, senior developers, who also expand into new areas of software development more readily.
In contrast, early-career developers show no significant productivity gains, suggesting that AI adoption may widen skill gaps and influence future career trajectories.
About the speaker
Simone Daniotti is a researcher at the Complexity Science Hub in Vienna and the Copernicus Institute of Sustainable Development at Utrecht University.
This event is part of the Financial Computing and Analytics Research Group seminar series at UCL Computer Science.
Further information
Related events
Perceived shocks and impulse responses
Economist Raffaella Giacomini presents new research on how people perceive economic shocks and how those beliefs shape expectations over time, using innovative methods and real-world inflation data.
Optimal Fees in Automated Market Makers
This seminar by Leandro Sánchez-Betancourt (Oxford) studies optimal dynamic fees in automated market makers, showing when higher or lower fees improve trading behaviour, liquidity and price formation.
