13. Human-in-the-Loop (HITL)
Mini-Project: Expense Approval with Human-in-the-Loop
An expense approval agent that auto-approves low-value requests and pauses for human review on high-value ones, using LangGraph's interrupt mechanism to capture the human decision before resuming.
Description
Human-in-the-Loop (HITL) integrates human oversight into an agent's execution flow. At critical decision points, the agent pauses and requests human approval, input, or correction before proceeding. This is essential for high-stakes operations (financial transactions, data deletion, sending emails) where autonomous execution is too risky. LangGraph supports HITL natively through interrupt points and checkpointers that persist state while waiting for human input.
When to Use
- High-stakes actions (payments, deletions, external communications)
- When agent confidence is low and human judgment adds value
- Compliance requirements mandating human approval
- Iterative workflows where human feedback steers the agent
Benefits
| Benefit | Description |
|---|---|
| Safety | Prevents costly autonomous mistakes |
| Trust | Users maintain control over critical decisions |
| Quality | Human judgment supplements LLM reasoning |
| Compliance | Meets regulatory requirements for human oversight |
Architecture Diagram
flowchart TD
A[User Input] --> B[Agent Plans Action]
B --> C{High-Stakes Action?}
C -->|No| D[Execute Automatically]
C -->|Yes| E[Pause: Request Approval]
E --> F[Human Reviews]
F -->|Approved| D
F -->|Rejected| G[Agent Replans]
G --> B
D --> H[Return Result]
style A fill:#4CAF50,color:#fff
style B fill:#2196F3,color:#fff
style C fill:#FF9800,color:#fff
style E fill:#E91E63,color:#fff
style F fill:#9C27B0,color:#fff
style D fill:#00BCD4,color:#fff
style G fill:#F44336,color:#fff
style H fill:#4CAF50,color:#fff