Has GitHub become more uniform since ChatGPT? A study of code from 2019–2025

A Habr author set out to test the hypothesis that open-source code on GitHub has become more uniform since the arrival of ChatGPT and the spread of AI coding tools such as GitHub Copilot, Cursor, Claude Code and Windsurf. To do so, he analyzed platform data from 2019–2025.

Why test code for uniformity

Since ChatGPT launched and AI assistants went mainstream, development has sped up noticeably: developers can now generate code, tests, READMEs, comments and small utility functions in seconds. But there is a less obvious downside. If millions of developers rely on similar models and similar prompts, open-source code risks becoming more homogeneous. This is a hypothesis, not a settled conclusion: the author frames it as an open question that needs to be tested against real data, not as an established fact.

How the study works

The author based his analysis on GH Archive and BigQuery data from 2019–2025, examining commit messages, README contents, function names and other text artifacts from repositories. To compare different periods, he introduced his own metric, the GitHub Uniformity Index, which estimates how similar code and its accompanying text became before and after AI tools entered mainstream development.

The 2019–2025 window is deliberate. It covers the years before ChatGPT, its release, and the years that followed, when AI coding assistants, from GitHub Copilot to Cursor, Claude Code and Windsurf, grew from niche plugins into everyday developer tools. That makes it possible to look for a sustained trend spanning several years of adoption rather than a one-off spike.

Why it matters for open source

If the hypothesis holds, it would mean the convenience of AI assistants comes at a cost: less diversity of approaches and styles in open repositories. For an ecosystem that has historically valued multiple solutions to the same problem, that would be a meaningful signal about the side effects of bringing generative models into everyday development at scale.

• Analysis period — GitHub data from 2019–2025
• Data sources — GH Archive and BigQuery
• Subjects of study — commit messages, READMEs and function names
• New metric — the GitHub Uniformity Index
• AI tools mentioned — GitHub Copilot, Cursor, Claude Code, Windsurf

What GitHub code can actually tell us

GH Archive records public platform events such as commits, pull requests and issues, and pairing it with BigQuery lets researchers run SQL queries across years of that data without downloading entire repositories. This approach makes it possible to compare commit style, README structure and function naming in "pre-ChatGPT" and "post-ChatGPT" slices at statistically significant volumes, rather than on a sample of a handful of projects.

What it means

The question of code uniformity is not about aesthetics. It is about whether open-source code is losing diversity of approaches and solutions as development speeds up. The GitHub Uniformity Index methodology offers a way to measure this effect on real data instead of just eyeballing it. The very framing of the question shows that the debate over AI coding tools is gradually shifting from a simple "it got faster" to subtler side effects of mass adoption.