Research: The AGENTS.md file can reduce AI programming agent runtime by 29% and decrease output tokens by 17%

BlockBeats News, March 3 — According to 1M AI News, research teams from Singapore Management University, Heidelberg University, Bamberg University, and King’s College London published a paper on arXiv, first quantifying the impact of warehouse-level configuration files AGENTS.md on AI programming agent efficiency. AGENTS.md is a directive file stored in the root directory of a code repository, used to inform AI Agents about project architecture, build commands, coding standards, and operational constraints. It is similar to CLAUDE.md used by Anthropic Claude Code and copilot-instructions.md used by GitHub Copilot. Over 60,000 GitHub repositories have adopted it.

The research team conducted paired experiments using OpenAI Codex (gpt-5.2-codex) on 124 merged pull requests (each with no more than 100 lines of code changed) across 10 repositories, under conditions with and without AGENTS.md. Results showed that with AGENTS.md, median runtime decreased from 98.57 seconds to 70.34 seconds (a 28.64% reduction), median output tokens dropped from 2,925 to 2,440 (a 16.58% reduction), and there was no significant difference in task completion behavior (Wilcoxon signed-rank test, p < 0.05).

The researchers noted that AGENTS.md shifts agent guidance from “brief prompts” to “version-controlled, reviewable, collaboratively maintained configuration artifacts,” recommending that development teams adopt it as a standard practice in repositories. Limitations include testing only a single agent (OpenAI Codex), a small sample size of PRs, and no comprehensive code correctness evaluation.