Top AI Product

Developer Louis Abraham ran KL-divergence k-means clustering on the vocabulary of 47,464 GitHub pull request descriptions. Eight clusters emerged. One of them didn’t exist in 2025. Last month, it covered 45% of PRs attributed to humans. In other words: nearly half of “human-authored” PRs were written, or heavily rewritten, by Claude.

“Load-bearing” is the tell

The word “load-bearing” shows up 929 times inside that cluster and only 82 times outside — a 39x gap, making it the clearest shibboleth of Claude’s presence. Other giveaways: “quietly,” “latent,” “genuine,” “seam.” The twist: “delve,” the internet’s favorite AI-detector word, was never a Claude-ism. That one belongs to ChatGPT.

What it actually is

This is a data-analysis project turned interactive dictionary of Claude-speak: pick a word, see its frequency curve and real PR examples. No model, no API — just clustering math on public GitHub data. That’s exactly why it exploded: 694 points and 326 comments on Hacker News, picked up by Daring Fireball and BoingBoing. Nobody had a number for how much Claude ghost-writes open source. Now there’s one, and it’s 45%.

You Might Also Like

• Stripe Minions 1300 Pull Requests a Week Zero Human Written Code
• 27k Github Stars in Weeks Learn Claude Code by Shareai lab Breaks Down ai Coding Agents Into 12 Lessons
• Claude hud hit 5 3k Github Stars Because Developers Were Flying Blind With Claude Code
• 27 Agents 109 Skills 88k Github Stars is Everything Claude Code Genius or Over Engineering
• 708 Github Stars in 48 Hours Claude Token Efficient Universal Claude md and the Fight Over Claudes Most Expensive Habit