How it works
LangChain abstracts LLM operations into composable units: chains (sequential prompts), agents (LLM with tool use), retrievers (data lookup), and memory. The original abstractions felt heavy in production, which led the LangChain team to ship LangGraph for stateful agent work.
Example
A 2023-vintage production agent built on LangChain initialize_agent with zero-shot-react-description. In 2026 the same team would rebuild it on LangGraph with explicit state nodes.