Human-in-the-loop means deliberately putting a person into an automated or AI-driven process to review, approve, correct or reject a step before it completes, rather than letting the system run all the way on its own. It’s a way to keep automation’s speed while still applying human judgment to decisions that carry real risk, cost, or reputational weight.
Example
An AI agent drafts a reply to a customer complaint, but instead of firing it off automatically, the draft goes to a support agent on Slack for a quick approve or edit first. The automation still does the heavy lifting - a person just keeps final say.
n8n relevance
Human-in-the-loop patterns show up often, and are encouraged, in production n8n workflows involving AI agents, especially wherever a step is hard to undo - sending an email, issuing a refund, posting publicly, updating a customer’s CRM record. n8n supports this with nodes that pause a workflow and wait for approval - a message with approve or reject buttons, say - then resume based on the response. Knowing when to let a step run fully automated versus when to add a human checkpoint is a judgment call n8n’s AI automation training spends real time on, since over-automating risky actions is one of the more common early mistakes.
Related concepts
See also AI Agent, Trigger and Multi-Agent System.