langgraph-fundamentals

LangGraph state updates overwrite previous data when reducers are not defined for list fields, causing default replacement behavior instead of accumulation. Defining custom reducers ensures state merges correctly across multiple nodes and parallel execution steps.

Use the Send object within a conditional edge to dispatch multiple state instances to parallel worker nodes, enabling fan-out execution in a compiled StateGraph. Ensure your state includes reducers to properly accumulate results from these concurrent workers.

Conditional edges evaluate state to determine the next node, while Command objects allow simultaneous state updates and goto routing within a single node return. Command provides explicit control over both data mutation and graph flow progression.

You should return a partial update dict from LangGraph nodes rather than mutating the full state object directly. Partial updates trigger the defined reducers to safely merge new values with existing state, preventing lost updates and unexpected graph behavior.

Stream intermediate progress from LangGraph agent workflows by using the stream method instead of invoke during graph execution. This outputs state changes and node outputs in real-time, allowing you to display step-by-step updates to a user interface.

Handle LangGraph execution failures by implementing explicit error-handling strategies that distinguish between transient, tool, user-fixable, and unexpected errors. Design dedicated fallback nodes or conditional routing paths to gracefully manage these common failure modes.