LangGraph is LangChain's successor framework for building stateful, multi-actor LLM applications. You define agent logic as a directed graph: nodes are steps (LLM calls, tool calls, logic), edges route based on state. More explicit than chain-style frameworks.
LangChain's older chain-based approach couldn't easily express branching, loops, and multi-agent coordination. LangGraph solves this: explicit nodes and edges with conditional transitions.
LangChain has a reputation for abstraction overhead. LangGraph addresses some of this but retains some of the complexity. Evaluate against raw API approaches before adopting.