A governance pattern in which an AI system may propose and prepare an action in a connected tool — draft the email, open the ticket, schedule the review — but cannot execute it until a human explicitly approves. The gate is what makes agentic capability safe to adopt in a business context: every action carries provenance (which decision produced it), an audit record, and a window to undo.
Approval gating deliberately trades autonomy for reversibility, on the reasoning that the cost of an unwanted action in someone's CRM or inbox is far higher than the cost of one click.