Why AI Agents Need Evidence Before They Retry

AI agents can mistakenly repeat actions like refunds when they lose confirmation of success, risking duplicate effects. Research shows agents often report success despite errors, highlighting the need for evidence-based retry safeguards.