Study maps how developers configure agentic AI coding tools
A paper by Galster et al. analyzes configuration mechanisms across Claude Code, GitHub Copilot, Cursor, Gemini, and Codex, finding that Context Files dominate and `AGENTS.md` is emerging as an interoperable standard across tools.
Score breakdown
The study provides the first empirical baseline on how developers configure agentic coding tools across a large set of real-world repositories, establishing that `AGENTS.md` serves as a natural cross-tool starting point and that advanced configuration mechanisms remain largely underutilized.
- 01Study analyzed 2,853 GitHub repositories to examine agentic AI coding tool configuration practices.
- 02Eight configuration mechanisms were identified, ranging from static context to executable and external integrations.
- 03Tools covered include Claude Code, GitHub Copilot, Cursor, Gemini, and Codex.
Galster, Mohsenimofidi, Lulla, Abubakar, Treude, and Baltes present a systematic analysis of how developers configure agentic AI coding tools, published as part of the Proceedings of the 3rd ACM/IEEE International Conference on AI-powered Software (AIware 2026). The study covers five tools — Claude Code, GitHub Copilot, Cursor, Gemini, and Codex — and identifies eight configuration mechanisms spanning static context, executable scripts, and external integrations. Drawing on an empirical analysis of 2,853 GitHub repositories, the paper examines adoption rates and practices in detail for three mechanism types: Context Files, Skills, and Subagents.
First, Context Files dominate the configuration landscape and are frequently the only mechanism present in a repository; `AGENTS.md` is singled out as an interoperable standard that works across tools.
Three key findings emerge. First, Context Files dominate the configuration landscape and are frequently the only mechanism present in a repository; `AGENTS.md` is singled out as an interoperable standard that works across tools. Second, advanced mechanisms such as Skills and Subagents remain rare, and where Skills are used, they rely predominantly on static instructions rather than executable scripts. Third, distinct configuration practices are forming around individual tools, with Claude Code users showing the widest variety of mechanism adoption. The authors frame these findings as an empirical baseline and call for longitudinal and experimental research into how configuration strategies evolve and affect agent performance.
Key facts
- 01Study analyzed 2,853 GitHub repositories to examine agentic AI coding tool configuration practices.
- 02Eight configuration mechanisms were identified, ranging from static context to executable and external integrations.
- 03Tools covered include Claude Code, GitHub Copilot, Cursor, Gemini, and Codex.
- 04Context Files are the dominant configuration mechanism and are often the sole mechanism in a repository.
- 05`AGENTS.md` is identified as an emerging interoperable standard across tools.
- 06Skills and Subagents (advanced mechanisms) are adopted by only a small number of repositories.
- 07Claude Code users employ the broadest range of configuration mechanisms among all tools studied.
Topics
Summary and scoring are generated automatically from the original article. We always link back to the publisher and never republish images or paywalled content. Last processed Jun 9, 2026 · 17:05 UTC. How this works →