New Research Reassesses the Value of Agents.md Files for AI Coding
7 hours ago
- #developer tools
- #AGENTS.md
- #AI coding agents
- ETH Zurich research finds AGENTS.md files often hinder AI coding agents, recommending omitting LLM-generated context files.
- Human-written context files offer marginal gains (4% success rate increase) but raise costs by up to 19%.
- LLM-generated context files reduce task success rates by 3% and increase inference costs by over 20%.
- Context files lead to unnecessary actions like more tests and file reads without improving task resolution.
- Developers debate the utility of AGENTS.md, with some arguing they're invaluable for domain-specific knowledge.
- Writing context files helps articulate codebase specifics, benefiting both AI agents and human team members.