Does AI Help or Homogenize? Thoughts on Knowledge Production in Open-Source

One of the questions driving my current research is deceptively simple: when developers use AI coding assistants, do they become more productive — or do they just start sounding more alike?

The Productivity Story

The productivity case is intuitive. AI tools reduce the friction of writing boilerplate, looking up syntax, and debugging trivial errors. In theory, this frees contributors to focus on harder, more creative problems. Early studies on GitHub Copilot suggest meaningful speed gains on routine tasks.

The Homogenization Worry

But there's a flip side. If everyone draws from the same model trained on the same corpus, the diversity of approaches in a codebase might shrink. Open-source communities have historically derived much of their value from the variety of perspectives contributors bring. If AI nudges everyone toward the same "average" solution, are we quietly eroding that?

What I'm Trying to Find Out

With Abhishek Nagaraj and Eunae Yoo, I'm using variation in AI tool adoption across open-source projects to try to answer this. The empirical challenge is separating the effect of the tool from selection — projects that adopt AI tools early may already differ in important ways.

More to come as the data starts talking back.