A new dataset tracking commits co-authored by Claude — visible via GitHub's co-author metadata — reveals a striking distribution: roughly 90% of all Claude-linked commits land in repositories with zero stars, and about 98% in repos with fewer than two stars.
The finding, surfaced by researchers analyzing public GitHub commit data and discussed widely on Hacker News this week, offers the first large-scale glimpse into where AI-assisted coding is actually happening.
The Long Tail of AI Development
The numbers look dramatic at first glance. But some context is essential: GitHub's star distribution itself follows a steep power law. Out of roughly 15 million unique repositories with at least one star, fewer than 2,000 have over 10,000 stars. The vast majority of GitHub activity — human or AI-assisted — occurs in the long tail of obscure, personal, and experimental repositories.
Still, the concentration is notable. It suggests that the primary beneficiaries of tools like Claude Code aren't the large, well-resourced open-source projects with active maintainer communities. They're the solo developer building a side project at midnight, the startup founder prototyping an MVP, the student learning to code with an AI partner.
What the Data Shows
The dataset covers recent months of GitHub commits tagged with Claude's co-author signature. Breakdowns show:
- ~14.9 million unique repositories with 1 star
- ~1.2 million with 10+ stars
- ~213,000 with 100+ stars
- ~28,000 with 1,000+ stars
Claude Code's activity skews heavily toward the first bucket — zero-star projects that have never been publicly surfaced or discovered.
The Democratization Argument
Proponents argue this is exactly what transformative technology looks like early on. The printing press didn't immediately produce bestsellers — it produced pamphlets, personal correspondence, and religious texts read by small communities. The same logic may apply here: AI-assisted coding is first and most visibly empowering people who previously couldn't build at all.
For every well-publicized enterprise deployment of Claude in a Fortune 500 engineering org, there appear to be thousands of quiet individual builds happening in repos no one will ever star.
Implications for the Ecosystem
The pattern raises interesting questions about where AI-driven code growth will show up next. If the bulk of Claude Code activity is in zero-visibility repos, the impact on public open-source ecosystems may be slower to materialize than headline deployments suggest. But the sheer volume of new projects being built — however small — represents a meaningful expansion of who gets to make software.
The claudescode.dev live feed, which streams recent Claude-linked commits in real time, makes the scale viscerally apparent: dozens of new commits per minute, spanning languages, domains, and time zones.
