Cursarium logoCursarium
intermediateFree

AI Agents in LangGraph

by Tavily Team & Harrison Chase · DeepLearning.AI

4.6
(5,100 reviews)
100K+ enrolled1 hourUpdated 2024-10

Our Verdict

Worth it — with caveats

AI Agents in LangGraph is a free, ~1.5-hour intermediate short course from DeepLearning.AI taught by LangChain co-founder/CEO Harrison Chase and Tavily co-founder/CEO Rotem Weiss, and it is genuinely worth taking for the right learner: you build a ReAct-style agent from scratch in Python, then rebuild it in LangGraph to understand nodes, edges, state, persistence, streaming, and human-in-the-loop control. It holds a real 4.7/5 rating (273 ratings on Class Central, ~310 on Coursera as of mid-2026), and the standout value is hearing the framework's actual author explain why and how to use it. The biggest caveat, confirmed repeatedly in student reviews, is that the code is dated: it was published in mid-2024, and LangGraph reached a breaking v1.0 in October 2025, so the notebooks now show library incompatibilities and some examples (notably the Human-in-the-loop lesson) fail without manual fixes. Treat it as a high-quality conceptual primer rather than a copy-paste-ready, up-to-date codebase.

The teaching is excellent and free, but because it predates LangGraph v1.0 (Oct 2025) the code is partly outdated and several reviewers report broken examples and library conflicts. It is a strong 'take' for learning concepts from the framework's creator, but only if you are comfortable adapting deprecated APIs to current LangGraph yourself.

Best for: Developers with solid intermediate Python who already grasp basic LLM/API calls and want a fast, authoritative mental model of how agentic loops and the LangGraph framework actually work (state graphs, tool calling, persistence, streaming, human-in-the-loop). Ideal for engineers evaluating LangGraph for a project who want it explained directly by LangChain's CEO, and for people who learn well by building a from-scratch agent before adopting the abstraction.

Skip if: Complete beginners to Python or LLMs (it assumes you can read and debug intermediate Python and is not a gentle on-ramp), learners who need plug-and-play code that runs unmodified against the latest libraries, anyone wanting a verifiable certificate from the free DeepLearning.AI version, and those seeking deep coverage of production concerns like multi-agent orchestration, evaluation, deployment, or LangGraph's current v1.0 API and middleware patterns.

About This Course

Build autonomous AI agents that can reason, plan, and use tools using LangGraph and LLM function calling.

What You'll Learn

Build an LLM agent from scratch in Python and understand the division of labor between the LLM and the surrounding control code
Re-implement the same agent using LangGraph's core components (nodes, edges, and shared state) to make it more controllable
Use agentic search (via Tavily) that returns structured, multi-result answers instead of raw links to ground an agent's responses
Add persistence and streaming so agents can manage state across threads, switch conversations, and reload prior states
Incorporate human-in-the-loop checkpoints to review or interrupt an agent's actions before they execute
Assemble a multi-step essay-writing agent that mirrors a researcher's plan, research, write, and revise workflow

Curriculum

Introduction

Course overview and framing of why LangGraph adds controllability over plain LLM loops (video, ~6 min).

Build an Agent from Scratch

Implement a ReAct-style agent in plain Python with an LLM to expose the underlying loop before any framework abstraction (video + code, ~12 min).

LangGraph Components

Rebuild the agent with LangGraph's nodes, edges, and state to make it more controllable and debuggable (video + code, ~19 min).

Agentic Search Tools

Integrate agentic search (Tavily) that returns predictable, structured multi-answer results rather than links (video + code, ~5 min).

Persistence and Streaming

Add state persistence across threads and stream intermediate agent steps; switch and reload conversation states (video + code, ~9 min).

Human in the loop

Insert manual approval/interrupt points into the agent graph for oversight; several learners report this lesson's code errors out and needs fixes (video + code, ~14 min).

Essay Writer

Build a multi-node agent that plans, researches, drafts, and revises an essay, replicating a researcher's workflow (video + code, ~18 min).

LangChain Resources

Pointers to LangChain/LangGraph documentation and further resources (video, ~2 min).

Conclusion

Wrap-up and next steps, followed by a short reading quiz (video, ~4 min).

Prerequisites

  • Intermediate Python (able to read/write and debug functions, classes, and type hints)
  • Basic familiarity with LLMs and calling a chat model API (e.g., OpenAI)
  • Conceptual comfort with tool/function calling and the ReAct (reason-act) pattern is helpful but introduced in the course
  • An OpenAI API key and (for agentic search) a Tavily API key to run the notebooks yourself

Instructor

Tavily Team & Harrison Chase

Instructor · DeepLearning.AI

Pros & Cons

Pros

  • Taught directly by the framework's creator (Harrison Chase, LangChain CEO) plus Tavily's CEO, giving authoritative first-party insight into design intent
  • The build-from-scratch-then-rebuild-in-LangGraph structure builds real understanding of what the abstraction is doing, not just API memorization
  • Free on the DeepLearning.AI platform and very time-efficient (~1.5 hours) with runnable Jupyter notebooks and six code examples
  • Strong real rating (4.7/5 across ~273-310 ratings) with reviewers praising clear explanations and a solid conceptual starting point for agents
  • Covers practical, production-relevant agent concepts: tool use, persistence, streaming, and human-in-the-loop oversight

Cons

  • Code is outdated: published mid-2024, it predates LangGraph v1.0 (Oct 2025), and multiple reviewers report library incompatibilities and that examples fail in a fresh environment
  • Specific reported breakage in the Human-in-the-loop lesson, where learners hit errors even after reloading and rerunning
  • Short and fast-paced: it is a conceptual primer, not deep coverage of multi-agent systems, evaluation, or production deployment
  • The free DeepLearning.AI version issues no certificate, and on Coursera it is a Guided Project gated behind subscription/audit

Alternatives To Consider

Frequently Asked Questions

Is AI Agents in LangGraph free?

Yes — AI Agents in LangGraph is free to access. Free to take on the DeepLearning.AI learning platform (no certificate). The same content is also offered as a Coursera Guided Project, where it requires a Coursera Plus subscription or one-time purchase; auditing options on Coursera are limited. You also need your own OpenAI and Tavily API keys to run the code, which can incur small usage charges.

Who is AI Agents in LangGraph for?

Developers with solid intermediate Python who already grasp basic LLM/API calls and want a fast, authoritative mental model of how agentic loops and the LangGraph framework actually work (state graphs, tool calling, persistence, streaming, human-in-the-loop). Ideal for engineers evaluating LangGraph for a project who want it explained directly by LangChain's CEO, and for people who learn well by building a from-scratch agent before adopting the abstraction.

What will you learn in AI Agents in LangGraph?

Build an LLM agent from scratch in Python and understand the division of labor between the LLM and the surrounding control code; Re-implement the same agent using LangGraph's core components (nodes, edges, and shared state) to make it more controllable; Use agentic search (via Tavily) that returns structured, multi-result answers instead of raw links to ground an agent's responses; Add persistence and streaming so agents can manage state across threads, switch conversations, and reload prior states.

What are the prerequisites for AI Agents in LangGraph?

Intermediate Python (able to read/write and debug functions, classes, and type hints); Basic familiarity with LLMs and calling a chat model API (e.g., OpenAI); Conceptual comfort with tool/function calling and the ReAct (reason-act) pattern is helpful but introduced in the course; An OpenAI API key and (for agentic search) a Tavily API key to run the notebooks yourself.

Is AI Agents in LangGraph worth it?

The teaching is excellent and free, but because it predates LangGraph v1.0 (Oct 2025) the code is partly outdated and several reviewers report broken examples and library conflicts. It is a strong 'take' for learning concepts from the framework's creator, but only if you are comfortable adapting deprecated APIs to current LangGraph yourself.

How we reviewed this course

This is an independent editorial assessment by Cursarium, based on DeepLearning.AI's published course materials and aggregated public learner feedback (last reviewed 2026-06). We have not independently completed the course. Links to providers are standard references, not paid placements.