L23- LangGraph Introduction | State, Nodes, Edges, Compile & Build Your First Agentic Workflow

Lecture 23 of the AI for Software Engineers series — Bipin Kumar introduces LangGraph, the most powerful agentic AI framework in the industry. Today's class covers the three core building blocks — State, Nodes, and Edges — and a complete working workflow is built from scratch.

🧠 What's Covered:
Why LangGraph — Not Just create_agent:
create_agent works for a single step — one query, one answer. When your application needs multiple steps, branching logic, or shared memory across actions, you need LangGraph. LangGraph lets you design the full flow of your application like a block diagram, then execute it.
LangGraph vs Other Frameworks:
Google ADK (Google), AutoGen (Microsoft), Strand (AWS), n8n, CrewAI, OpenAI Agent SDK — all are agentic frameworks doing the same job. LangGraph is the most powerful, gives the most control, and is the most asked about in interviews. Learn one framework fully and switching to another becomes easy.
The 3 Core Components:
State — a shared notebook that all nodes can read from and write to. You define it as a TypedDict class with the variable names your workflow needs. For example: question, search_result, and answer. At the start, all values are empty. As the workflow runs, each node fills in the values it is responsible for.
Node — a Python function that performs one step in the workflow. The input is always the State. The output is a dictionary updating only the keys that node cares about. Other keys are left unchanged. For example, a search node reads the question from State and writes back the search result.
Edge — the connection between nodes. add_edge(A, B) says "after A finishes, run B." START and END are built-in nodes provided by LangGraph automatically.
4 Steps to Build Any LangGraph Workflow:
Step 1 — Define the State (what information flows through the workflow).
Step 2 — Write the Node functions (each function = one step).
Step 3 — Add edges to connect nodes in the right order.
Step 4 — Compile the graph with builder.compile(), then invoke it with your input.
The Workflow Built in Class:
START → search_node → answer_node → END.
search_node reads the question from State and returns a search result. answer_node reads the search result from State and generates the answer. When invoked with a question, both nodes run in sequence, each updating their part of the State. The final State contains the question, the search result, and the completed answer.
How State Flows:
At START: question is set, search_result and answer are empty.
After search_node: search_result is filled, answer is still empty.
After answer_node: answer is filled. Workflow ends.
Note: this class builds the workflow structure without an LLM inside the nodes — dummy values are used to understand the concept cleanly.

⏭️ Next Lecture (Lecture 24):
👉 Adding an LLM model to nodes, conditional edges, and real agent logic inside LangGraph

💬 Questions about how State updates across nodes? Drop them in the comments — Bipin replies!
📌 Subscribe so you never miss a class.

#LangGraph #AgenticAI #State #Nodes #Edges #LangChain #AIforEngineers #BipinKumar #Workflow #AgenticWorkflow #Python #GenerativeAI #GoogleADK #AutoGen #AIInterview

Видео L23- LangGraph Introduction | State, Nodes, Edges, Compile & Build Your First Agentic Workflow канала Zero to Deployed