Week 0: The Tutorial You Found Is Already Behind

The four lines you probably already have open in another tab

If you searched "LangGraph quickstart" some time in the last year, you have a tab open right now with this import:

from langgraph.prebuilt import create_react_agent

It still runs. That's the trap. LangGraph's own current reference page for that function (reference.langchain.com/python/langgraph.prebuilt/chat_agent_executor/create_react_agent) marks it deprecated in favor of create_agent from the langchain package, as of the v1.0 stable release in October 2025. Deprecated code doesn't throw an error, though. It just quietly stops being the path anyone maintaining LangGraph wants you on. You can build a working agent on the old import today, ship it, and never see a warning. You'll be building on a foundation its own maintainers have moved off of, using a keyword (prompt) that the new factory silently ignores in favor of system_prompt. The failure mode isn't a crash. It's an agent that runs fine and simply has no system prompt.

That's not a hypothetical for this course. Lesson 01 fetched that same reference page directly and covers the full audit.

What actually moved, and what didn't

The high-level agent factory moved. The low-level graph-construction layer underneath it (the pieces you build a graph, its state, and its message-merging rules from) didn't change at all. You don't need to know those names yet, only that they're the stable ground this course's other seven lessons build on. That's the pattern this course teaches you to spot on your own: which layer of a framework actually shifted, and which layer just looks unfamiliar because a tutorial explained it badly.

The gap this course closes

Most LangGraph material teaches the concepts once and freezes the code at whatever the API looked like on publish day. The concepts (state as a typed contract, conditional routing, tool binding, parallel execution, human-in-the-loop pauses) don't move nearly as often as the surface does. This course teaches both together: the durable ideas, and which specific import, keyword, and model string are current as of when each lesson was checked against LangGraph's own live reference pages, not assumed current from memory.

If you've already built something on LangGraph and it feels like it's held together with workarounds you don't fully trust, that's usually not a sign you misunderstood the framework. It's a sign the framework moved and the tutorial you learned it from didn't.

What you can do after Week 1

By lesson 08 you'll have built a stateful, multi-tool agent with graph-native parallel execution and a human-in-the-loop pause point, on the current API, with the reasoning for every migration decision shown, not asserted. You'll also be able to look at someone else's LangGraph code and tell, quickly, whether it's current or quietly stale. That second skill outlasts this specific API version. The framework will move again, and knowing how to check is more durable than memorizing today's import path.

Eight lessons, no padding. Lesson 01 starts by finishing the audit above: exactly what changed, what didn't, and why the parts that didn't change are the parts worth learning first.