AI & Open Source

Under the supervision of Assistant Professor Nathan TeBlunthuis, I study how AI adoption changes the way people contribute to open-source software. A growing number of projects include a file called CLAUDE.md or AGENTS.md that gives AI coding tools instructions and project-specific guidance. Previous research has focused largely on how much code AI produces. My study instead compares projects with and without these instruction files to examine differences in participation and activity.

I used 2025–26 data from GH Archive, a public record of activity on GitHub, to identify projects with at least two contributors and a minimum level of development work or discussion. The resulting dataset contains 3,734 projects, including 358 with an AI instruction file. I then used statistical models to examine how the presence of these files relates to overall project activity and activity per contributor, while accounting for differences in team size and the influence of a few exceptionally large projects.

Because of the volume of data, I run the analysis on the university’s high-performance computing system and have learned to work with datasets too large for a typical laptop. The initial results showed no statistically clear relationship between AI instruction files and the number of developers participating in a project (p = 0.925). However, projects with these files recorded approximately 30 more actions per contributor on average, including code updates and discussion activity (p < 0.001). At this stage, the findings suggest that instruction files may be associated with greater activity among existing contributors rather than attracting new ones. Next, I plan to expand the dataset roughly tenfold and compare activity before and after each file was introduced so I can evaluate possible cause and effect more carefully.