Lesson 9: LangChain & LangGraph — Stateful Agent Workflows

Build AI Agents: From Zero to Production

Part 4: Agentic Frameworks

xdev.asia

Introduction

Building an agent from scratch is great to understand the concept, but production needs a framework. LangGraph (by the LangChain team) is the most powerful framework currently available for stateful, graph-based agent workflows.

---

1. LangChain vs LangGraph

	LangChain	LangGraph
Paradigm	Chains (linear)	Graphs (cyclic)
State	Stateless	Stateful
Best for	Simple pipelines, RAG	Complex agents, multi-step
Control flow	Sequential	Conditional routing

2. LangGraph Fundamentals

from langgraph.graph import StateGraph, MessagesState, START, END

def call_llm(state: MessagesState):
    response = model.invoke(state["messages"])
    return {"messages": [response]}

def use_tools(state: MessagesState):
    # Execute tool calls
    ...

# Build graph
graph = StateGraph(MessagesState)
graph.add_node("llm", call_llm)
graph.add_node("tools", use_tools)
graph.add_edge(START, "llm")
graph.add_conditional_edges("llm", should_use_tools, {"yes": "tools", "no": END})
graph.add_edge("tools", "llm")

agent = graph.compile()

---

Summary

• LangGraph = state machine for agent workflows
• Nodes = functions, Edges = transitions, State = shared data
• Conditional routing for complex decision trees
• Human-in-the-loop built-in support
• Persistence for long-running workflows

Exercises

1. Implement research agent with LangGraph
2. Add human-in-the-loop approval step
3. Implement retry logic when tool fails
4. Build an agent with 3+ conditional branches