A recent study by researchers from Singapore Management University, Heidelberg University, Bamberg University, and King’s College London has demonstrated that AGENTS.md files significantly improve the efficiency of AI programming agents. Published on arXiv, the study reveals that these configuration files, now used in over 60,000 GitHub repositories, reduce AI agent runtime by 28.64% and output tokens by 16.58%. The research involved experiments with OpenAI Codex on 124 pull requests across 10 repositories, comparing performance with and without AGENTS.md. Results showed a decrease in median runtime from 98.57 seconds to 70.34 seconds and a reduction in median output tokens from 2,925 to 2,440, without affecting task completion. The study suggests that AGENTS.md files transform agent guidance into version-controlled, reviewable artifacts, recommending their adoption as a standard practice. However, the study's limitations include testing only with OpenAI Codex and a focus on small pull requests.