How it works
LangGraph defines agents as state machines: nodes (functions), edges (transitions, including conditional ones), and shared state. Durable execution means the graph can be paused, resumed, or replayed without state loss. Strong fit for multi-step agentic workflows.
Example
A customer service agent built on LangGraph: entry node classifies intent, routes to specialist node (billing, account, technical), each specialist node has its own tool set, all share the same conversation state, escalation node handles the human-handoff case.