Study Finds AGENTS.md Often Fails Improving Agents

A recent study evaluated whether AGENTS.md repository context and extended instructions improve coding agent performance and cost-effectiveness. Researchers found that adding lengthy repository context and extra instructions often failed to yield proportional accuracy gains while substantially increasing compute costs and response latency. The results suggest developers should empirically validate context utility, prioritize concise task-relevant information, and weigh cost-benefit trade-offs before broad rollout.