In plain language
The automation does the heavy lifting, but a person signs off on decisions that are risky or uncertain.
Example
An AI drafts a refund response, and a support lead approves it in Slack before it is sent.
Related terms
AI Agent · Guardrails · Workflow · Conditional Logic
Common misconception
Adding a human step does not make a process safe by itself; reviewers need enough context and time to catch errors.
Implementation guidance
Send only uncertain or high-impact cases for review, show the inputs and AI reasoning, set response deadlines, and log every approval or override.