Human-in-the-loop (HITL) means pausing the agent at critical decision points to get human approval before proceeding — essential for high-stakes or irreversible actions.

When to Require Human Approval

• Before sending emails or messages
• Before making purchases or financial transactions
• Before deleting or modifying important files
• When the agent is uncertain about a decision
• Before any irreversible action

LangGraph HITL Pattern

# Add interrupt_before to pause at specific nodes
graph = workflow.compile(
    checkpointer=memory,
    interrupt_before=["send_email", "execute_payment"]
)
# Agent pauses — human reviews state and resumes or modifies

Reference:

Human-in-the-loop documentation

https://langchain-ai.github.io/langgraph/concepts/human_in_the_loop/