*Result*: Who is using AI to code? Global diffusion and impact of generative AI.

*Generative coding tools promise big productivity gains, but uneven uptake could widen skill and income gaps. We train a neural classifier to spot artificial intelligence (AI)–generated Python functions in more than 30 million GitHub commits by 160,097 software developers, tracking how fast, and where, these tools take hold. Currently, AI writes an estimated 29% of Python functions in the US—a shrinking lead over other countries. We estimate that quarterly output, measured in online code contributions, consequently increased by 3.6%. AI seems to benefit experienced, senior-level developers: They increased productivity and more readily expanded into new domains of software development. By contrast, early-career developers showed no significant benefits from AI adoption. This may widen skill gaps and reshape future career ladders in software development. Editor's summary: As generative artificial intelligence (genAI) advances reshape careers, one profession remains distinctly affected: software development. Recent research found that genAI job cuts eliminated many entry-level positions, whereas senior-level developer jobs, which require critical thinking and uncodified, tacit skills, were spared. However, a fuller picture was needed to understand genAI's impact on human innovation and inequities. Daniotti et al. trained a machine learning classifier to detect genAI-generated code among software developers in the US, China, France, Germany, India, and Russia (see the Perspective by Wu and Vasilescu). Although entry-level developers used genAI the most, it did not appear to benefit them. Experienced, senior-level developers, by contrast, leveraged genAI to increase their productivity and software innovations. The growing skills gap may be a bellwether for changes in genAI-exposed industries. —Ekeoma Uzogara INTRODUCTION: Generative artificial intelligence (genAI) has the potential to radically transform the nature and productivity of human labor, yet we lack systematic measurement of how fast genAI is diffusing and who benefits from it. In this work, we develop a method to detect AI-generated code at the individual level and apply it to more than 30 million code contributions by 160,097 software developers across six countries. We find that genAI adoption is rapid and global and that it boosts productivity and accelerates expansion into new technical domains—but only for senior-level developers. Early-career developers, by contrast, despite being the most frequent users of genAI, show no measurable benefits. Therefore, rather than closing skill gaps, genAI appears to be widening them. RATIONALE: We measure the uptake of genAI coding assistants in Python projects on GitHub, the world's largest collaboration platform for software development. We developed a neural network classifier that accurately determines whether a function—a self-contained unit of code—was generated by AI or written by a human. We trained this classifier on human-written functions expanded with synthetic equivalents created by chaining two large language models (LLMs): one model describing human code in natural language, which a second model uses to reimplement the function. We applied this classifier to more than 30 million commits from 160,097 developers across six countries (US, France, Germany, India, China, and Russia) between 2019 and 2024, charting genAI diffusion globally. For 100,097 US developers, we use fixed-effects models—comparing how the same developer's output changes with changes in their genAI usage—to estimate effects on the nature and volume of code produced. RESULTS: GenAI tools have diffused rapidly, with sharp upticks coinciding with new generations of coding assistants. The US had an early yet narrowing lead: By the end of 2024, an estimated 29% of Python functions were produced with substantial AI support. France and Germany draw close (23 and 24%, respectively), and India caught up considerably, although China and Russia still lag. In the US sample, we find no gender difference in usage rates, but early-career developers use genAI more often compared with senior-level developers. GenAI increases productivity—point estimates suggest a 3.6% boost in quarterly commit rates at current adoption levels. At the level of the US economy, this additional output translates conservatively into a yearly value of 23 billion to 38 billion USD, but the true value of genAI may be substantially larger. GenAI is also associated with increased experimentation with new software libraries, which suggests that it helps programmers expand into unfamiliar technical domains. Notably, however, these productivity and exploration gains accrue exclusively to senior-level developers. Despite their higher adoption rates, early-career developers show no significant benefits. CONCLUSION: We show that genAI use can be accurately detected in the digital traces of software developers, revealing rapid global diffusion of AI coding tools. GenAI increases output and helps programmers expand into new domains—but only for senior-level developers. Early-career developers, despite being the most enthusiastic adopters, see no measurable gains. This leads to substantial uncertainty for young developers about their career development in the software sector. Global diffusion of genAI coding assistants and their impact.: (Left) The share of AI-written Python functions (2019 to 2024) grows rapidly, but countries differ in their adoption rates. (Right) Comparing usage rates for the same developers at different points in time, genAI adoption is associated with increased productivity (commits), breadth of functionality (library use), and exploration of new functionality (library entry), but only for senior developers, whereas early-career developers do not derive any statistically significant benefits from using genAI. [ABSTRACT FROM AUTHOR]*