Pattern · intermediate
Human-in-the-Loop Pattern
HITL patterns pause agents for human approval, clarification, or takeover on risky or ambiguous steps.
- hitl
- safety
- operations
Human-in-the-loop (HITL) inserts people at critical points in an agent run: approvals, clarifications, or full takeover.
When to require HITL
- Payments, legal sends, prod changes
- Low-confidence classifications with high impact
- Conflicting evidence
- First-time actions in a new environment
Implementation pattern
if action.risk >= threshold or confidence < min:
pause(run_id, payload_for_human)
wait_for(approval | edit | reject)
resume(run_id)UX requirements
- Show the proposed action and rationale
- Show tool diffs / previews
- Allow edit-then-approve
- SLA timers and escalation paths
Anti-patterns
- Approving every trivial step (alert fatigue)
- No timeout handling
- Approval without enough context
HITL is a control, not a substitute for evals and least-privilege tools.
Frequently asked questions
What is human-in-the-loop for agents?
Human-in-the-loop (HITL) means the agent pauses for a person to approve actions, answer questions, or take over before continuing—especially for irreversible or ambiguous work.
Related reading
Agent Safety and Guardrails
Practical guardrails for agentic AI: permissions, sandboxes, human approval, budgets, content filters, and audit trails.
Routing and Handoffs Pattern
Routing patterns send tasks to the right agent, model, or workflow based on intent, risk, and skills.
AI Agent Production Readiness Checklist
A practical checklist for shipping agentic systems: tools, auth, budgets, evals, observability, HITL, and incident response.