Generative AI is rapidly reshaping software development, and the pace of change is striking. Recent research shows that AI-assisted coding has spread quickly over the past few years, although adoption varies widely across countries. In the United States, the share of newly written code created with AI support climbed from about 5 per cent in 2022 to nearly 30 per cent by early 2025. Elsewhere, the uptake has been slower, with China, for example, reaching closer to 12 per cent over the same period. These figures underline both the speed of diffusion and the uneven geography of AI use in programming.

The implications matter because software sits at the heart of the modern economy. In the United States alone, firms spend hundreds of billions of dollars each year on wages for coding and related tasks. Globally, billions of lines of software code are written and maintained every day to keep digital services, industrial systems, and consumer products running. As AI tools become embedded in everyday programming work, they are beginning to alter this critical backbone of economic activity.

The findings are based on a large-scale analysis of real-world programming behaviour. Researchers examined more than 30 million contributions written in Python by roughly 160,000 developers on GitHub, the world’s largest collaborative coding platform. Because GitHub logs every addition and modification to a codebase, it offers an unusually detailed view of how software is produced. By focusing on Python, one of the most widely used programming languages, the study captures trends that are likely representative of broader developments in the industry.

To identify AI involvement, the researchers used a specially trained model designed to detect whether segments of code were generated with the help of tools such as ChatGPT or GitHub Copilot. The results point to swift adoption, particularly in the United States. By the end of 2024, around one-third of newly written software functions will be created with AI support. At the same time, the analysis reveals substantial regional gaps. Several European countries fall behind the US but still show relatively high usage. In contrast, countries such as Russia and China lag further back, partly due to limits on access to leading AI models.

Differences also emerge between programmers themselves. The study finds no meaningful gap in AI usage between women and men. Experience, however, plays a significant role. Less experienced developers rely more heavily on generative AI, using it in over a third of their code, compared with just over a quarter for more seasoned programmers. Despite this, the measured productivity gains, averaging about 3.6 per cent by the end of 2024, are driven almost entirely by experienced developers. Beginners see little benefit, suggesting that AI does not automatically level the playing field and may even widen existing skill gaps.

Economically, even these modest productivity gains add up. Given the scale of spending on programming work in the United States, AI-assisted coding could already be generating tens of billions of dollars in additional value each year, and this is likely a conservative estimate. Looking ahead, software development is set to change even more deeply as AI becomes a core part of digital infrastructure. The key challenge for companies, policymakers, and educators will be to ensure that the advantages of AI are broadly shared, rather than reinforcing inequalities in an economy that increasingly runs on code.

More information: Simone Daniotti et al, Who is using AI to code? Global diffusion and impact of generative AI, Science. DOI: 10.1126/science.adz9311

Journal information: Science Provided by Complexity Science Hub