I mapped AI Agent adoption across 217,000 GitHub repositories (alteredcraft.com)

A scan of 217,000+ active public GitHub repositories (200–430,000 stars) in early October 2025 found a 4.9% overall adoption rate for AI coding agents based on committed configuration files (e.g., .cursorrules, CLAUDE.md). The data surfaces three clear patterns: Claude Code—launched Feb 2025 as a CLI-first agent—overtook early leader Cursor by late June and widened its lead; GitHub Copilot showed an adoption uptick from August 2025 and surpassed Cursor thereafter; and new repositories created in 2025 adopt agents at ~13.7% vs. 4.3% for older projects (≈3× higher). Most repos use one agent (86%), while 13.4% maintain multi-tool setups, indicating active experimentation or transitional migrations. Methodologically, the study tracked committed per-repo config files and recent activity (commits within 90 days), so it measures deliberate, repository-level choices rather than all agent usage—private repos, global IDE settings, and uncommitted configs are excluded, meaning real-world usage is likely higher. Technically and strategically, the rise of a CLI-centric agent (Claude Code) and Copilot’s late surge underscore two adoption vectors: workflow affinity (CLI/automation-friendly tools) and distribution/embedding within existing platforms. Continued monthly scans and maintainer interviews aim to reveal the “why” behind these trends and whether agents become core development infrastructure or remain task-focused aids.