Podcast Episode
To identify AI-generated code, the research team trained a specialized AI model to detect whether blocks of code were created with tools like ChatGPT or GitHub Copilot. By December 2024, AI was writing an estimated 30.1% of Python functions from US contributors, compared with 24.3% in Germany, 23.2% in France, 21.6% in India, 15.4% in Russia, and 11.7% in China.
The findings align closely with corporate disclosures from major technology companies. Microsoft CEO Satya Nadella stated in April 2025 that approximately 30% of the company's code is now AI-generated, while Google CEO Sundar Pichai previously reported that over 25% of new Google code comes from AI systems.
Frank Neffke, who leads the Transforming Economies group at the Complexity Science Hub, noted that the economic impact of generative AI in software development was already substantial at the end of 2024 and has likely increased further since the analysis concluded.
Simone Daniotti of the Complexity Science Hub and Utrecht University explained that beginners hardly benefit at all. The study found that experienced developers saw a 6.2% increase in output when using AI at average adoption rates, while early-career developers showed no statistically significant productivity gains. Other research has measured productivity gains for experienced developers at 3.6%.
This counterintuitive finding reveals a fundamental paradox. The developers who seemingly need the most assistance, those just starting their careers, gain the least from AI tools. Meanwhile, senior developers who already possess deep expertise are able to leverage AI to accelerate their work.
The explanation appears to lie in the ability to critically evaluate AI output. Experienced developers can quickly assess whether an AI suggestion is correct, efficient, and architecturally sound. They use AI to automate mechanical aspects of coding while focusing on higher-level design decisions. Beginners, lacking this evaluative capacity, may not recognize when AI is leading them down problematic paths.
The researchers also discovered that AI adoption helps experienced developers experiment more with new software libraries and venture into unfamiliar programming domains, a phenomenon not observed among newer programmers. For veterans, AI serves as a learning accelerator, not merely a code generator.
Johannes Wachs of Corvinus University explained that users in China and Russia have faced barriers to accessing these models, blocked by their own governments or by the providers themselves. He noted that recent Chinese developments like DeepSeek, released after the study's data collection ended, may help close this gap.
The geographic disparities reflect not just technological access but also regulatory environments and geopolitical tensions. Countries that restrict access to leading AI models may find their software developers at a competitive disadvantage in the global marketplace.
Rather than democratizing coding by making it accessible to beginners, current AI tools appear to be amplifying existing skill advantages. This could widen the gap between junior and senior developers, potentially creating barriers to career advancement for newcomers.
The research also highlights concerns about skill erosion. If beginners rely heavily on AI without developing fundamental understanding, they may struggle to build the expertise needed to eventually benefit from these tools. Senior developers express less anxiety about long-term impacts on their abilities, with 41% concerned about skill erosion compared to 50% of junior developers.
The study concludes that the key question for businesses and policymakers is not whether AI will be used but how to make its benefits accessible without reinforcing inequalities. As AI continues to reshape software development, ensuring that the next generation of programmers can build genuine expertise while leveraging AI assistance will be critical.
About 76% of professional developers either currently use or plan to use AI coding tools, and 41% of all code is now AI-generated. Yet 75% of developers still manually review every AI-generated code snippet before merging, demonstrating that AI functions as a powerful assistant rather than a replacement for human judgment.
As the technology continues to evolve, the challenge will be designing AI tools and educational approaches that help beginners build foundational skills while allowing experienced developers to maximize their productivity. The rapid diffusion of AI in coding is no longer a future possibility but a present reality reshaping one of the most critical skills in the modern economy.
AI Now Generates Nearly One-Third of New US Code, Landmark Study Reveals
January 23, 2026
Audio archived. Episodes older than 60 days are removed to save server storage. Story details remain below.
Artificial intelligence has quietly woven itself into the fabric of software development, and groundbreaking research published in the journal Science puts hard numbers on its reach. Nearly one-third of all new code written in the United States is now generated with AI assistance, marking what researchers describe as an extraordinarily rapid technological diffusion.
The Scale of AI Adoption
The study, led by researchers at the Complexity Science Hub in Vienna in collaboration with Utrecht University and Corvinus University of Budapest, analyzed more than 30 million Python contributions from roughly 160,000 developers on GitHub. The findings reveal that AI-assisted coding in the US jumped from around 5% in 2022 to 29% by the end of 2024.To identify AI-generated code, the research team trained a specialized AI model to detect whether blocks of code were created with tools like ChatGPT or GitHub Copilot. By December 2024, AI was writing an estimated 30.1% of Python functions from US contributors, compared with 24.3% in Germany, 23.2% in France, 21.6% in India, 15.4% in Russia, and 11.7% in China.
The findings align closely with corporate disclosures from major technology companies. Microsoft CEO Satya Nadella stated in April 2025 that approximately 30% of the company's code is now AI-generated, while Google CEO Sundar Pichai previously reported that over 25% of new Google code comes from AI systems.
Economic Impact Reaches Billions
The economic implications are substantial. The US spends an estimated $637 billion to $1.06 trillion annually in wages on coding-related work. The study estimates that AI coding assistants are already generating between $23 billion and $38 billion in additional value each year. Conservative estimates place the annual value at $9.6 to $14.4 billion, rising to $64 to $96 billion if higher productivity estimates from randomized control trials are applied.Frank Neffke, who leads the Transforming Economies group at the Complexity Science Hub, noted that the economic impact of generative AI in software development was already substantial at the end of 2024 and has likely increased further since the analysis concluded.
The Experience Paradox
Perhaps the most striking finding concerns who benefits from these tools. While less experienced programmers use AI more frequently, relying on it for 37% of their code compared to 27% for seasoned developers, the productivity gains flow almost exclusively to experienced programmers.Simone Daniotti of the Complexity Science Hub and Utrecht University explained that beginners hardly benefit at all. The study found that experienced developers saw a 6.2% increase in output when using AI at average adoption rates, while early-career developers showed no statistically significant productivity gains. Other research has measured productivity gains for experienced developers at 3.6%.
This counterintuitive finding reveals a fundamental paradox. The developers who seemingly need the most assistance, those just starting their careers, gain the least from AI tools. Meanwhile, senior developers who already possess deep expertise are able to leverage AI to accelerate their work.
The explanation appears to lie in the ability to critically evaluate AI output. Experienced developers can quickly assess whether an AI suggestion is correct, efficient, and architecturally sound. They use AI to automate mechanical aspects of coding while focusing on higher-level design decisions. Beginners, lacking this evaluative capacity, may not recognize when AI is leading them down problematic paths.
The researchers also discovered that AI adoption helps experienced developers experiment more with new software libraries and venture into unfamiliar programming domains, a phenomenon not observed among newer programmers. For veterans, AI serves as a learning accelerator, not merely a code generator.
Global Disparities in Access
The study uncovered significant disparities in AI adoption across countries. The US leads at 29%, followed closely by France at 24% and Germany at 23%. India has been catching up rapidly, reaching 20%, while Russia at 15% and China at 12% remain further behind.Johannes Wachs of Corvinus University explained that users in China and Russia have faced barriers to accessing these models, blocked by their own governments or by the providers themselves. He noted that recent Chinese developments like DeepSeek, released after the study's data collection ended, may help close this gap.
The geographic disparities reflect not just technological access but also regulatory environments and geopolitical tensions. Countries that restrict access to leading AI models may find their software developers at a competitive disadvantage in the global marketplace.
No Gender Gap in Usage
The researchers found no differences in AI usage between men and women, suggesting that adoption patterns are driven more by experience level and access than by demographic factors.Implications for the Future
The study's findings raise important questions about the future of software development as a career path, particularly for those entering the field. If AI tools primarily benefit those who already possess expertise, the pathway from novice to expert developer may become more challenging.Rather than democratizing coding by making it accessible to beginners, current AI tools appear to be amplifying existing skill advantages. This could widen the gap between junior and senior developers, potentially creating barriers to career advancement for newcomers.
The research also highlights concerns about skill erosion. If beginners rely heavily on AI without developing fundamental understanding, they may struggle to build the expertise needed to eventually benefit from these tools. Senior developers express less anxiety about long-term impacts on their abilities, with 41% concerned about skill erosion compared to 50% of junior developers.
The study concludes that the key question for businesses and policymakers is not whether AI will be used but how to make its benefits accessible without reinforcing inequalities. As AI continues to reshape software development, ensuring that the next generation of programmers can build genuine expertise while leveraging AI assistance will be critical.
About 76% of professional developers either currently use or plan to use AI coding tools, and 41% of all code is now AI-generated. Yet 75% of developers still manually review every AI-generated code snippet before merging, demonstrating that AI functions as a powerful assistant rather than a replacement for human judgment.
As the technology continues to evolve, the challenge will be designing AI tools and educational approaches that help beginners build foundational skills while allowing experienced developers to maximize their productivity. The rapid diffusion of AI in coding is no longer a future possibility but a present reality reshaping one of the most critical skills in the modern economy.
Published January 23, 2026 at 11:36am
