Home › Claude AI › claudearchitect › claudeai_agents › When should a human-in-the-loop confirm step fire?
When should a human-in-the-loop confirm step fire?
ABefore any tool call, always
BBefore irreversible actions (push to prod, send money)
COnly when the model is unsure
DNever; agents should be fully autonomous
Answer & Solution
Correct answer: B. Before irreversible actions (push to prod, send money)
Cheap to add, saves you once; require explicit human confirmation for actions that can't be undone.
Related questions