Human Approval
Human approval should sit immediately before actions where mistakes would be costly, irreversible or sensitive.
Human approval should sit immediately before actions where mistakes would be costly, irreversible or sensitive.
Agent drafts the proposed action.
Person sees evidence and impact.
Only approved actions run.
Understand the working model
Approval is more useful when the reviewer sees exactly what will happen: recipient, amount, record changes, evidence used and any uncertainty. Avoid generic “Approve?” prompts with no context.
Keep preparation separate from execution. The agent may draft an email or refund proposal automatically, but the external side effect happens only after approval.
Practice
Apply the pattern above to one real task you already do. Keep the first version small enough that you can inspect every input and output.
Do not ask for approval after the irreversible action already happened.
Place human approval at the last safe point before high-impact actions execute.