Cisco study: GitHub stars are a poor measure of AI framework adoption

A longitudinal analysis of 15 major open-source AI agent frameworks from late 2022 to early 2026 reveals that GitHub star counts are a poor predictor of ecosystem health. Researchers from Cisco and Indiana University examined 808,042 stars, 73,997 pull requests, 86,241 commits, and 987,330 user profiles across repositories including AutoGPT, LangChain, MetaGPT, Pydantic-AI, and OpenAI Agents Python. They found that hype cycles and inorganic activity inflate star counts. For example, AutoGPT gained 111,967 stars in one month but converted fewer than 9 contributors per 1,000 stars (contributor density), compared to LangChain’s 41. Lower-profile frameworks like Pydantic-AI showed higher contributor density, indicating deeper adoption.

Three key findings emerged: First, awareness (stars) and adoption (contributor density) often diverge. MetaGPT and LangFlow have contributor density ratios below 5 despite high visibility, and OpenAI’s own agents-python repository shows limited community depth despite institutional backing. Second, LangChain acts as shared infrastructure, attracting 82.5% of all cross-ecosystem contributors, suggesting it serves as a de facto standard for multi-agent orchestration. Third, contributor retention drops most sharply in the first 30 days after initial contribution and stabilizes near 90 days. The study concludes that ecosystem health is better measured by contributor density, cross-ecosystem engagement, and retention—metrics that offer engineering teams a more robust basis for framework evaluation than star counts alone.