Stop 15 · AI Agents & OrchestrationTopics · 2/9

LangChain

In short

Swapping OpenAI for Claude, adding a parser, plugging in a retriever: LangChain gives you ready parts so you are not rewriting the same glue code in every project.

A widely-used framework for chaining LLM calls, tools, memory and retrieval into structured applications — helpful for prototyping though often criticized for abstraction overhead at scale.

The blob swaps one small bowl on a round steel thali for another while the other bowls stay in place.

Think of it like this

A thali with standard katoris

A thali has fixed katoris: dal goes here, sabzi there, rice in the middle. Change the dal and the thali still works. LangChain gives every piece (prompt, model, parser, retriever) the same shape, so you can swap one without touching the rest.

Watch it happen

One chain.invoke() call

Prompt

Model

Parser

Output

1/4{"topic": "recursion"} fills the template and becomes a list of chat messages.

The big ideas

What you need to know

Runnables and the pipe

Prompts, models and parsers are all Runnables. The | operator joins them into a chain, read left to right.

Swap providers easily

init_chat_model lets you change provider with one string. Your prompt and parser code stays the same.

Integrations for everything

Loaders, vector stores and tools for hundreds of services live in separate langchain-* packages you install as needed.

LangGraph for agents

For loops, branches and state, the LangChain team points you to LangGraph, which builds agents as graphs of steps.

See it in code

Walk through the code

python

1from langchain.chat_models import init_chat_model2from langchain_core.prompts import ChatPromptTemplate3from langchain_core.output_parsers import StrOutputParser4 5prompt = ChatPromptTemplate.from_template(6    "Explain {topic} to a first-year student in 3 lines."7)8model = init_chat_model("gpt-5-mini", model_provider="openai")9chain = prompt | model | StrOutputParser()10 11print(chain.invoke({"topic": "recursion"}))

1/4A template with a {topic} slot. Every call fills it with a fresh value.

A three-part chain. Needs langchain, langchain-openai and an OPENAI_API_KEY. Change the model string to try another provider.

Save yourself the pain

Common mistakes

Oops

Copying a 2023 LangChain tutorial with LLMChain and old imports.

Do this instead

The API has changed a lot. Use the current docs, langchain_core imports and the pipe syntax.

Oops

Adding LangChain to a project that makes one API call.

Do this instead

For one call, the provider SDK is simpler. Bring in LangChain when you need swapping, retrieval or many integrations.

Your mission

Build and swap a chain

~30 min

Quick check

In prompt | model | StrOutputParser(), what does the parser receive?

Go deeper

1Build a simple chain and a simple agent using LangChain's core abstractions.
2Learn LangChain's building blocks: chains, tools, memory, retrievers.
3Understand when going 'framework-free' with raw API calls is actually simpler for your use case.