Who is Using AI to Code? Global Diffusion and Impact of Generative AI

Between 2019 and 2024, who adopted generative AI for coding, and what did they produce? A team from the University of Technology Vienna analyzed 60,000 programmer-year observations across six countries by tracking commits to public repositories on GitHub.

The Data Collection

The researchers collected all commits made by users with self-reported locations on GitHub during 2019 to 2024. They focused on Python projects — the most accessible language for analyzing AI usage across skill levels and industries. Then they applied an AI detector trained on real AI-generated code to identify when commits likely came from generative AI assistance.

To estimate the impact, they sampled 2,000 programmers per country per year across the United States, five other nations, and analyzed them using the PyDriller tool.

The Surprising Adoption Pattern

What they discovered challenges conventional wisdom about AI adoption. AI adoption is not uniform across experience levels or geographies. Early adopters are not necessarily senior engineers — in fact, junior programmers and students adopted AI coding assistance earliest and most eagerly.

This mirrors other technology adoption patterns: new tools see rapid uptake among those with the least to lose by experimenting. Senior engineers, accustomed to established workflows, adopted more cautiously.

Global Diffusion

The diffusion curve shows rapid adoption across all six countries during 2023 to 2024. But adoption rates vary dramatically by geography. Some countries show significantly higher adoption rates than others, reflecting differences in internet access, programming language preferences, GitHub participation rates, and cultural attitudes toward AI.

The Economic Impact

When one generative AI system handles code that previously required multiple developers, what changes in the labor market? The researchers ran economic models under two scenarios:

Scenario 1: Perfectly Elastic Supply of Code If the supply of code becomes infinitely responsive to demand (because AI can generate unlimited code), wages for programming labor fall sharply, and total employment contracts. The social surplus shifts heavily toward consumers.

Scenario 2: Perfectly Inelastic Supply of Code If programmer time becomes the binding constraint (you still need humans to evaluate, validate, and guide AI output), then higher code productivity increases demand for programming labor, raising wages and expanding employment.

Reality likely falls between these extremes. AI augments programmer productivity without eliminating programmer labor. The question becomes: do organizations hire more programmers (expanding the headcount) or fewer programmers who produce more (consolidating the headcount)? The data suggests both dynamics are occurring simultaneously in different sectors.

What This Means

The global spread of generative AI coding tools is accelerating. Adoption is bottom-up and rapid. The economic impact — on wages, employment, and skill demand — remains uncertain and varies dramatically by geography, industry, and organization.

One pattern is clear: programmers who know how to work with AI have become more valuable. The tool doesn't replace them; it amplifies the leverage of those who can use it effectively.