AI coding agents dilute human contributor share in open-source projects
Newcomer participation drops 3.7% while review depth rises 5.3% post-adoption.
A new preprint from researchers at Karlsruhe Institute of Technology presents the first large-scale causal analysis of how AI coding agents affect human contributor ecosystems. Analyzing 11,097 GitHub repositories from January 2023 to May 2026, the study uses a staggered difference-in-differences design to measure changes after AI agent adoption. The headline result: AI agents do not displace human contributors outright—the absolute number of human contributors remains essentially unchanged (ATT = 0.014, p = 0.224). However, human contributor density drops significantly by 1.9 percentage points (p = 0.002), meaning AI-generated pull requests increasingly crowd out human-authored changes.
More troubling for open-source sustainability: the relative share of newcomers declines by 3.7 percentage points (p < 0.001) immediately after adoption, and this effect persists. Meanwhile, review depth increases by 5.3% (p < 0.001), indicating that human reviewers must work harder to vet AI-generated code. The researchers term this pattern 'augmentation with dilution'—AI boosts output but systematically alters participation, shifting burden to review and reducing opportunities for junior contributors. Effects vary by project size, language, and maturity.
- Newcomer relative participation drops 3.7 percentage points (p < 0.001) immediately after AI agent adoption and stays depressed.
- Review depth increases by 5.3% (p < 0.001), suggesting AI shifts workload from writing code to reviewing it.
- Total human contributor count is unchanged, but human contributor density declines 1.9% as AI-generated PRs accumulate.
Why It Matters
AI agents boost output but reduce beginner on-ramps and increase review burden on maintainers.