Tag
The author describes a three-week experiment testing AI agent trustworthiness, finding that agreement between agents using the same model is unreliable, and outlines a system with human approval for critical decisions.