Human-in-the-Loop Agents

Build agents that know when to stop and ask for human judgment.

Why Human-in-the-Loop?

Critical for:

• High-stakes actions: Financial transactions, data deletion
• Compliance: Audit requirements, approval workflows
• Quality control: Review before publishing, sending
• Edge cases: When agent confidence is low
• Trust building: Users control what agents do

Core Patterns

Pattern 1: Interrupt Before Action

from langgraph.graph import StateGraph, END