Two layers, not two rivals
“LangChain vs LangGraph” used to mean picking between two frameworks. Today it’s mostly about picking a level.
- LangChain is the high-level layer. Its
create_agentfunction is “a minimal, highly configurable harness”: the prompt, the tools and the middleware around the model loop. - LangGraph is the low-level layer. It’s an orchestration framework where you define the graph yourself: nodes that do work, edges that decide what runs next, and state that flows through it.
LangChain’s agents are built on LangGraph, so a LangChain agent already gets LangGraph’s durable execution, human-in-the-loop support and persistence.
When LangChain is enough
Most agents follow the same loop: call the model, run the tools it asks for, and repeat until it’s done. If that describes your app, create_agent covers it, and middleware lets you change behavior without rewriting the loop. LangChain’s standard interfaces also make it easy to switch between model providers, embeddings and vector stores.
When to reach for LangGraph
LangChain’s docs recommend LangGraph for “advanced needs combining deterministic and agentic workflows”. In practice that means:
- fixed steps before or after the model, such as validation, routing or approval;
- loops and branches you design rather than leave to the model;
- long-running work that must survive a crash or wait for a person;
- several agents, where you decide exactly who acts when.
Where Deep Agents fits
Deep Agents is a third, higher layer: a batteries-included harness built on LangChain agents, with planning, subagents, context compression and a virtual file system. It’s worth a look for long, open-ended tasks where you’d otherwise build those pieces yourself.
The bottom line
- New to the stack, or building a standard tool-using agent or RAG app: LangChain.
- Designing your own workflow, or running long, resumable, multi-agent processes: LangGraph.
- Long, open-ended tasks that need planning and subagents: try Deep Agents.
Comparing agent frameworks more broadly? Read CrewAI vs LangGraph vs AutoGen. Weighing code against a visual builder? Read Dify vs LangChain.
How we compared
Facts come from the official LangChain documentation, the LangChain and LangGraph repositories, PyPI and the LangSmith pricing page, checked on 1 November 2026 and listed below. Check the official pages before committing, and tell us if something has changed.