Subgraphs & Multi-Agent Systems
Master LangGraph Subgraphs and Multi-Agent Systems — design parent-child compiled graphs, choose between isolated vs shared state, and scale complex agentic workflows.
As agentic applications scale, a single compiled graph with dozens of tools becomes unmaintainable. This module details dividing labor by nesting graphs within parent graphs (Subgraphs) to create decoupled multi-agent architectures.
Subgraphs & Encapsulation
Why this matters
Subgraphs encapsulate complex flows inside a parent graph — key for multi-team agent architectures.
A subgraph is a compiled graph used as a node inside a parent graph — encapsulates a sub-workflow.
sub_builder = StateGraph(SubState)
sub_builder.add_node("translate", translate_node)
sub_graph = sub_builder.compile()
builder.add_node("translator", sub_graph) # subgraph as node- Use when a team owns a reusable flow (translation, scoring, approval).
- State can be isolated or mapped between parent and child schemas.
Common mistakes
- Forgetting to compile the graph with a checkpointer when persistence is required.
- List state fields without reducers — updates overwrite instead of append.
- Infinite loops in cyclic graphs with no max iteration or termination edge.
Interview checkpoints
- Q: Explain subgraphs in LangGraph. A: One-sentence definition + one API.
- Q: Common bug? A: State, checkpointer, or routing loop issue.
Practice
- Basic: Sketch a minimal subgraphs example.
- Intermediate: Run a notebook cell demonstrating Subgraphs.
- Advanced: Break Subgraphs intentionally and read the LangSmith trace.
Recap
- You can explain subgraphs clearly.
- You know one mistake to avoid.
- You see how this connects to the next lesson.
Next: Multi-Agent
Multi-Agent Production Systems
Why this matters
Multi-agent systems coordinate specialized subgraphs with shared or isolated state for scale.
Multi-agent systems assign roles to subgraphs (researcher, writer, critic) coordinated by a supervisor node.
- Shared state vs message-passing — trade isolation for simplicity.
- Production: rate limits, error boundaries, subgraph-level retries.
Common mistakes
- Forgetting to compile the graph with a checkpointer when persistence is required.
- List state fields without reducers — updates overwrite instead of append.
- Infinite loops in cyclic graphs with no max iteration or termination edge.
Interview checkpoints
- Q: Explain multi-agent in LangGraph. A: One-sentence definition + one API.
- Q: Common bug? A: State, checkpointer, or routing loop issue.
Practice
- Basic: Sketch a minimal multi-agent example.
- Intermediate: Run a notebook cell demonstrating Multi-Agent.
- Advanced: Break Multi-Agent intentionally and read the LangSmith trace.
Recap
- You can explain multi-agent clearly.
- You know one mistake to avoid.
- You see how this connects to the next lesson.
Next: Back to LangGraph Hub