Andrew Ng introduces a new course on building AI agents with LangGraph, taught by Harrison Chase and Rotan Blyth. You'll learn to construct agents from scratch, then use LangGraph's graph-based approach for workflows you can control and debug. The course covers multi-agent setups like architect, coder, and reviewer agents working together.
I'm delighted to introduce AI agents in Langgraph, built in partnership with LangChain and Tivoli, taught by Harrison Chase and Rotten Weiss. When writing code, you don't just type out the program you want from start to finish without ever using backspace, at least icon code like that. If I were building something together with Parson and Rotem, maybe I'll try to create an outline of the architecture. Harrison May code up, V1, Rota may do a code review and make suggestions, and then I might edit it some more, and so on. AI agents use a similar step by step iterative workflow. For example, you may build a software architect agent by prompting an LLM to propose the overall architecture. A second coder agent implemented by prompting an LLM to write code might then thaw the code and the third agent could be instructed to act as a co reviewer and try to critique the software. These different agents can work iteratively together to produce the work product. This iterative agentic approach often produces much better results than a single shot application of prompting an LLM. LangChain has been a very popular open source framework for building applications. They recently added Langdraft, an extension that adds features specifically aimed at supporting agentic workflows. I'm delighted to introduce one of our instructors, Harrison Chase, who's CEO of LangChain. Thanks Andrew. LangChain has supported agents almost from the start, but as agents have proliferated, we found that we could better support their development, debugability and maintenance by directly supporting a graph structure that underlies so many of their implementations. This resulted in Langgraph. In this course you will write an agent from scratch and then you will rewrite it using Langgraph. You will learn about the components of Langgraph and how to combine them to build agents and other flow based applications. Langgraph is aimed at letting developers create highly controllable agents. You can specify exactly the state and the control flow of the agent or agents. This is extremely useful when you need a high amount of control to get your agent to perform reliably. That controllability aspect is a real strength of Landgraaf. And when you build your own agents, you find that many agents might need to use tools such as Search to enhance their built in knowledge. But standard search returns links to web locations where you might find an answer to a query. But this isn't directly usable by an agent to make this process easier. I'm also delighted to introduce another of our instructors, Rotan Blyth, who is co founder and CEO of Tuvali. Thanks Andrew an agentic search returns multiple answers to a query rather than links to answers. It formats the results in an agent friendly way. In this course, we'll show you how agentic search works and how you can use it in your application. AgentIC AI workflows are an emerging cutting edge way to build really powerful systems. I hope you take this course and gain skills that let you be a real change agent.
