A massive global analysis of GitHub activity shows generative AI is already writing close to one-third of new software functions in the United States, transforming how experienced developers work and raising big questions for education, business and policy.
Generative artificial intelligence is no longer just a coding assistant on the side. It is already helping write close to one-third of new software in the United States, and it is quietly reshaping who benefits most from the technology.
A new study published in the journal Science finds that by the end of 2024, about one in three newly written software functions in the United States were created with the help of generative AI tools such as ChatGPT and GitHub Copilot. The research offers one of the clearest pictures yet of how quickly AI is spreading through software development worldwide — and who is gaining from it.
The team, led by researchers at the Complexity Science Hub (CSH) in Vienna, analyzed real-world coding activity on GitHub, the world’s largest collaborative programming platform. They focused on Python, one of the most widely used programming languages in industry, research and education.
Simone Daniotti of CSH and Utrecht University, the corresponding author of the study, noted that the scale of the analysis was unprecedented.
“We analyzed more than 30 million Python contributions from roughly 160,000 developers on GitHub, the world’s largest collaborative programming platform,” he said in a news release.
To detect AI-written code, the researchers trained a specialized AI model to spot patterns typical of code generated by systems like ChatGPT or GitHub Copilot. They then tracked how the share of AI-assisted code changed over time and across countries, and how it affected productivity for different kinds of programmers.
Their conclusion: AI-assisted coding is spreading at remarkable speed.
“The results show extremely rapid diffusion,” added Frank Neffke, who leads the Transforming Economies group at CSH and is a professor at Interdisciplinary Transformation University Austria. “In the U.S., AI-assisted coding jumped from around 5% in 2022 to nearly 30% in the last quarter of 2024.”
By early 2025, the share of new code relying on AI had climbed to 29% in the United States, compared with 12% in China. European countries fell in between, with Germany at 23% and France at 24%. India, at 20%, has been catching up quickly.
“While the share of AI-supported code is highest in the U.S. at 29%, Germany reaches 23% and France 24%, followed by India at 20%, which has been catching up fast,” Neffke added.
The study highlights a clear global gap. The United States leads AI adoption in coding, while China and Russia lag behind, in part because of limited access to leading large language models (LLMs). In those countries, government restrictions and corporate policies have blocked or constrained access to popular Western AI tools, though some users rely on virtual private networks (VPNs) to get around those barriers.
CSH faculty member Johannes Wachs, also an associate professor at Corvinus University of Budapest, noted that the gap may not last.
“It’s no surprise the U.S. leads – that’s where the leading LLMs come from. Users in China and Russia have faced barriers to accessing these models, blocked by their own governments or by the providers themselves, though VPN workarounds exist. Recent domestic Chinese breakthroughs like DeepSeek, released after our data ends in early 2025, suggest this gap may close quickly,” he said in the news release.
Beyond geography, the study looked at who is actually using AI to code. Interestingly, less experienced programmers rely on AI more often than seasoned developers. Early-career coders used generative AI in about 37% of their code, compared with 27% for more experienced programmers.
Yet the productivity payoff runs in the opposite direction.
By the end of 2024, generative AI use was associated with a 3.6% increase in productivity — measured in part by how often developers commit code and how broadly they use software libraries. Neffke acknowledged that 3.6% might sound small in isolation, but emphasized its scale.
“That may sound modest, but at the scale of the global software industry it represents a sizeable gain,” he said.
Crucially, those gains were concentrated among experienced developers. For them, AI tools were linked to more frequent coding, broader use of existing software libraries and more exploration of new libraries and combinations of tools. In other words, AI seemed to help senior programmers move faster and venture into new technical territory.
For beginners, the story was different. Despite using AI more often, they did not see statistically significant productivity improvements.
“Beginners hardly benefit at all,” Daniotti added.
The findings suggest that generative AI may amplify existing skill differences rather than flatten them. Experienced developers can use AI to automate routine tasks, quickly learn unfamiliar tools and expand into new domains. Novices, by contrast, may lack the foundational knowledge needed to judge, debug or adapt AI-generated code effectively.
Wachs noted the pattern points to a deeper role for AI in learning and innovation.
“This suggests that AI does not only accelerate routine tasks, but also speeds up learning, helping experienced programmers widen their capabilities and more easily venture into new domains of software development,” he said.
The economic implications are enormous. Software is now embedded in nearly every sector, from finance and health care to manufacturing and transportation. In the United States alone, the researchers estimate that companies spend between $637 billion and $1.06 trillion each year on wages for programming-related tasks, based on an analysis of about 900 different occupations.
“The U.S. spends an estimated $637 billion to $1.06 trillion annually in wages on programming tasks, according to an analysis of about 900 different occupations,” added co-author Xiangnan Feng of CSH.
If 29% of that work is AI-assisted and productivity rises by 3.6%, the study estimates that generative AI is already adding between $23 billion and $38 billion in value each year to the U.S. economy.
Neffke stressed that these figures are probably on the low side.
“This is likely a conservative estimate,” he said, adding that “the economic impact of generative AI in software development was already substantial at the end of 2024 and is likely to have increased further since our analysis.”
The results arrive as universities, coding boot camps and employers race to adapt curricula and workflows to an AI-first era. For students and early-career developers, the study underscores the importance of building strong fundamentals in programming and problem-solving, rather than relying too heavily on AI-generated snippets.
For companies and policymakers, the research raises urgent questions about access, training and inequality. If AI tools primarily boost the productivity of already skilled workers, they could widen gaps between top performers and everyone else — within firms, across regions and between countries.
Wachs framed the challenge bluntly.
“For businesses, policymakers, and educational institutes, the key question is not whether AI will be used, but how to make its benefits accessible without reinforcing inequalities,” he said.
Neffke added that understanding barriers to AI adoption is now a strategic priority, not a technical detail.
“When even a car has essentially become a software product, we need to understand the hurdles to AI adoption – at the company, regional, and national levels – as quickly as possible,” he said.
The study offers a first global map of how generative AI is changing software work. But the authors emphasize that this transformation is still in its early stages.
As AI tools grow more powerful and more widely available — including new domestic models in countries that currently lag behind — the share of AI-written code is likely to keep rising. The next challenge will be ensuring that students, workers and institutions are prepared not just to use these tools, but to thrive in a world where AI is woven into the very fabric of software.
Source: Complexity Science Hub