AI Writes Nearly One-Third of New Software Code
The software industry is immense. In the US alone, companies spend an estimated hundreds of billions of dollars annually on programming labor. Every day, billions of lines of code keep the global economy running. How is Artificial Intelligence (AI) reshaping this backbone of modern life?
AI’s Global Uptake in Software Development
Generative AI is transforming software development, and it’s doing so fast, though not evenly across the world. In a study published in Science, Frank Neffke, Professor of Economic Transformation and Complexity at IT:U and Group Leader at the Complexity Science Hub (CSH), with colleagues at CSH examines the uptake of generative AI in software development.
The research team analyzed more than 30 million Python code contributions from roughly 160,000 developers on GitHub, the world’s largest platform for collaborative coding. They found that, by late 2024, AI had assisted nearly one-third of newly written software functions in the United States. The pace and pattern of adoption vary sharply by country, and the productivity gains are concentrated among experienced developers.
Mapping the Uptake: Rapid Growth with a Narrowing Global Gap?
To identify AI-generated segments, such as those produced with tools like ChatGPT or GitHub Copilot, the team trained a specialized model. The headline result: extremely rapid diffusion. In the US, the share of AI-assisted programming rose from about 5% in 2022 to nearly 30% in the last quarter of 2024.
However, adoption diverges meaningfully by country:
- United States: 29%
- Germany: 23%
- France: 24%
- India: 20% (with strong acceleration since 2023)
- Russia: 15%
- China: 12%
It isn’t surprising that the US leads, given that many of the most capable large language models (LLMs) originate there. Notably, new Chinese breakthroughs such as DeepSeek, released after the dataset closed in early 2025, suggest this gap could narrow quickly.
Who Benefits from AI in Coding: Experience Over Usage
Across the board, generative AI increased programmer productivity by an average of 3.6% by the end of 2024. That may sound modest, but given the scale of the global software economy, it translates into substantial value: With 29% of code AI-assisted and average productivity rising 3.6%, the implied annual value lift for the US economy would be worth tens of billions of dollars.
But who is driving these gains? The study found no gender differences in AI usage – experience, however, is pivotal:
- Less experienced programmers used AI more often: 37% of their code vs. 27% for experienced developers.
- Yet the measurable productivity gains were driven almost entirely by experienced developers.
- Novices saw negligible benefits from AI in performance terms.
This suggests AI does not automatically level the playing field. Instead, it may amplify existing advantages, especially for those who have sufficient experience to evaluate and integrate code proposed by AI coding assistants.
Interestingly, experienced AI users also experimented more with new libraries – collections of coding resources that help implement specific types of functionalities in a computer program – and new library combinations. That pattern hints at AI accelerating not just routine tasks but also learning, helping seasoned developers widen their capabilities and venture into new domains faster.
AI’s Impact on Business and Policy
As AI becomes embedded in core digital workflows, the stakes for business and policy are rising. Organizations that empower experienced developers to use AI well are likely to gain a competitive edge in both productivity and innovation.
“In a time when even a car is, at its core, a software product, we need to understand the barriers to AI adoption as quickly as possible at the company, regional, and national levels.”
Frank Neffke, Professor of Economic Transformation and Complexity
At the same time, companies – and educational institutes – need to rethink how they grow new talent and train junior programmers in the presence of AI. Given the speed with which this new technology improves, this is likely to be an ongoing challenge.