Generative AI gives seasoned developers the edge
Based on an analysis of around 170,000 developers, the research finds that despite rapid uptake of generative AI tools, the productivity and innovation gains are concentrated almost entirely among experienced programmers. That pattern is likely to widen skills and income gaps in the tech labour market.
“A common expectation is that AI will help less experienced workers become as productive as senior ones,” said Johannes Wachs, a Corvinus University researcher and one of the paper’s authors. “We find the opposite: the gap is widening, not closing.”
The authors used a machine-learning method to identify AI-assisted functions on GitHub across more than 30 million Python code contributions. The analysis covered six countries: the United States, China, France, Germany, India and Russia.
Juniors use it more, seniors use it better
The results suggest that by the end of 2024, about 29% of Python functions in the United States were written by generative AI. The US remains in the lead, though its advantage is narrowing. Germany and France follow closely at 23–24%, while India is approaching 20% after initially lagging behind. China and Russia, by contrast, have adopted later and more slowly, partly due to access restrictions and censorship.
Researchers estimate that the rise of AI use in the United States increased software development productivity by an average of 3.6%. But that gain is driven almost entirely by experienced developers, who were able to raise the volume of code they produced (commits) by 6.2%. Among less experienced programmers, the study finds no meaningful productivity effect.
One notable detail is that newcomers to GitHub appear to use AI tools more often in relative terms, with AI present in roughly 37% of their code compared with 27% for veteran developers. Even so, it is the more seasoned programmers who are better able to convert AI use into tangible gains. The study reports no significant differences by gender in adoption.
AI helps developers push into new areas
Generative AI is changing not only how much code is written, but also what kind of code developers produce. Those using AI are more likely to combine software libraries they have not previously used together, including tools for data visualisation, natural language processing and web technologies. The models suggest that average AI use among US developers could increase the number of new library combinations by 2.7%, pointing to easier entry into new technical areas. This “exploration advantage”, however, also appears mainly among experienced developers.
The US adoption estimates are strikingly close to figures reported internally by Microsoft and Amazon. A key contribution of the study, the authors add, is that it allows cross-country comparisons rather than focusing only on individual companies or case studies.
Tens of billions of dollars in value, but unevenly distributed
Software development – and open-source code in particular – offers a window into how AI changes work: every contribution is logged, timestamped, and traceable. The authors estimate that AI coding tools are already generating tens of billions of dollars in value for the US economy annually, though the precise figure remains uncertain.
Adoption is moving quickly, but it varies sharply across countries and individuals. The findings suggest generative AI is not automatically a force that narrows gaps; it can also amplify existing inequalities.
The analysis was published in Science on 22 January. Authors: Simone Daniotti (Utrecht University, Complexity Science Hub Vienna), Johannes Wachs (Corvinus University of Budapest, Centre for Economic and Regional Studies, Complexity Science Hub Vienna), Xiangnan Feng (Complexity Science Hub Vienna), and Frank Neffke (Complexity Science Hub Vienna; IT:U Interdisciplinary Transformation University).