Back to Blog
ResearchMarch 22, 2026 · 5 min read

We Scanned 13 Popular Open Source Repos for AI-Generated Code. Here's What We Found.

Deno is 41% AI-coded. Claude Code appeared in 99% of AI commits. Data from 6,500 commits across Next.js, Ruff, LangChain, Ollama, and more.

Published by GitIntel Research

The Results

| Repository | Language | AI Commits | AI % | Agent | | --- | --- | --- | --- | --- | | Deno (101K) | Rust/TS | 206/500 | 41.2% | Claude Code | | Claude Code (30K+) | TypeScript | 63/500 | 12.6% | Claude Code | | Next.js (131K) | TypeScript | 27/500 | 5.4% | Claude Code | | Ruff (38K) | Rust | 20/500 | 4.0% | Claude Code | | Mem0 (48K) | Python | 18/500 | 3.6% | Claude Code | | shadcn/ui (82K) | TypeScript | 11/500 | 2.2% | Claude Code | | CrewAI (46K) | Python | 8/500 | 1.6% | Claude Code | | Biome (17K) | Rust | 7/500 | 1.4% | Claude Code | | Astro (50K) | TypeScript | 7/500 | 1.4% | Claude Code | | LangChain (96K) | Python | 6/500 | 1.2% | Claude Code | | Ollama (162K) | Go | 4/500 | 0.8% | Devin + Claude Code | | Pydantic (23K) | Python | 1/500 | 0.2% | Claude Code | | Tauri (90K) | Rust | 0/500 | 0.0% | -- |

Total: 378 AI-assisted commits across 6,500 scanned (5.8% overall)

Five Things That Surprised Us

1. Deno is 41% AI-coded

The biggest outlier by far. More than 4 out of every 10 recent commits in Deno's repo were co-authored by Claude Code. That's not a tool being used occasionally — that's a fundamental part of the development workflow. Deno is a Rust + TypeScript runtime, and AI is writing a significant chunk of both.

2. Claude Code dominates everything

Across all 13 repos, Claude Code appeared in 99% of AI-assisted commits. The only exception was Ollama, where Devin contributed 3 commits. No Cursor. No Copilot. This isn't because other tools aren't being used — it's because Claude Code is the only major tool that consistently leaves a Co-Authored-By trail.

3. The attribution gap is massive

Our scan only detects explicit Co-Authored-By trailers. But many AI-assisted commits don't have these — developers copy-paste from ChatGPT, use Copilot inline suggestions, or accept Cursor tab completions without any attribution marker. The true AI percentage could be 2-5x higher than what we detected. We're measuring the floor, not the ceiling.

4. AI adoption follows the team, not the language

Tauri (Rust) has 0% AI commits. Ruff (also Rust) has 4%. Biome (also Rust) has 1.4%. Deno (also Rust) has 41.2%. Same language, wildly different adoption rates. AI adoption is a team decision, not a technology constraint.

5. Even Anthropic's own repo is only 12.6%

Claude Code's own repository — built by the team that makes the AI coding tool — still has 87.4% human-authored commits. AI isn't replacing developers. It's augmenting them at varying rates.

What This Means

For engineering leaders: You can't answer "how much of our code is AI-generated?" without measuring it. And the answer might surprise you — it could be 0% or 41%.

For compliance teams: AI-generated code is already shipping in production at major open source projects. If your organization has policies around AI code review or licensing, you need tooling to enforce them.

For developers: The adoption curve is steep. Deno went from 0 to 41% AI commits. If your team hasn't adopted AI coding tools yet, you're competing against teams that have.

How We Measured This

We used GitIntel v0.1.0, scanning the 500 most recent commits on the main branch of each repository:

gitintel scan --format json --limit 500

Limitations:

What's your repo's AI percentage?

Run gitintel scan and find out.

# Install
curl -fsSL https://gitintel.com/install.sh | sh

# Scan any repo
cd your-repo
gitintel scan

View on GitHub

Open source (MIT) · Local-first · No data leaves your machine

Data collected March 22, 2026. All repositories scanned at latest main branch HEAD.

Related reading on GitIntel: