Human-in-the-loop means AI agents don’t operate unchecked. Critical decisions — especially those involving money, data, or safety — require human review or approval before they proceed. It’s a safeguard for trust and accountability.