Human-in-the-Loop

A pattern where sensitive AI actions (send email, delete files, deploy code) require human approval before execution. The agent proposes, the human approves. Critical for building trust and preventing costly mistakes.