ai
Production AI Agents: Tool Boundaries, Evaluations, and Human Oversight
A practical architecture for dependable AI agents: narrow tool contracts, explicit state, bounded execution, adversarial evaluation, and human approval for consequential actions.