LangGraph: connect agents through shared state
Compare shared-message collaboration, supervisors and nested agent teams.
Three coordination patterns
The January 2024 walkthrough compares collaboration through shared messages, a supervising agent and hierarchical teams. Its written companion explains who sees intermediate work and who chooses the next worker. Shared scratchpads expose more detail; separate scratchpads let a supervisor receive final responses. These are architecture examples from an early LangGraph release.
Prepare a small graph
Basic Python and an understanding of state will help. Use the separate current Graph API guide for implementation details. As a first exercise, draw two nodes and decide what data passes between them. Define the state, add the nodes and edges, then compile the graph before using it. Check how state updates are combined. Follow the current API rather than assuming the older video code still runs unchanged.
