Agents are the topic everyone is searching for right now and most of the material out there is either marketing fluff or a wall of framework documentation, so a tight ninety minute course from the person who actually built LangChain is worth paying attention to. The teaching choice that makes this one land is the order. Harrison Chase and Rotem Weiss have you build a simple agent from scratch in plain Python first, the loop where the model decides on an action, you run a tool, you feed the result back, and round it goes, and only after you have felt how fiddly that is by hand do they reintroduce the same agent built in LangGraph. That contrast is the whole lesson, because suddenly the graph, the state, and the controllability are not abstract features, they are answers to problems you just hit yourself.
From there it moves into agentic search, which is the idea of giving your agent a search tool that returns clean structured results rather than raw web pages, and into persistence and human in the loop, so the agent can pause and wait for your approval before doing something consequential. It is hands on the whole way, you are running cells and watching the agent reason, not watching slides. The honest caveats are about level and scope. This is not a beginners course, you need to be fine with Python and you should already know how to call a model from code, ideally after doing one of the introductory prompting courses first.
Because it is short it does gallop, and a couple of the meatier ideas around state management could happily fill a session of their own. And naturally it teaches LangGraph, so part of what you learn is that specific framework rather than universal agent theory, which is fine as long as you go in knowing it. For free, taught by the source, with that build it yourself then rebuild it structure, I think it is one of the best on ramps to actually building agents available today.