Short answer
Human-in-the-loop is a design pattern where a person reviews, approves, or corrects an AI system's output before or after it acts, keeping human judgment inside the decision rather than removing it. It is the deliberate choice to keep a person on the critical path.The pattern exists because full autonomy is not always the safe or the smart default. Some intents are high-stakes, ambiguous, or rare, and the right move is to draft with AI and let a human send. Human-in-the-loop is how a team automates the confident cases without surrendering the judgment calls.
In Aide, the AI platform for customer service, agent assist is the human-in-the-loop interface: Aide drafts a response inside the helpdesk and the agent reviews and sends. Drafting is native in Zendesk and Front, and on Gorgias drafts post as internal comments. The choice of which intents run autonomously and which stay human-in-the-loop is made intent by intent.
A human stays on any intent that has not cleared testing against real past conversations, so nothing unverified ships unattended. And the harder conversations stay with people for a reason: working them is what keeps a team's judgment sharp.