Free AI Agents Courses in 2026, Compared by Framework and Prerequisites

By Pooja Goenka · Published 2026-10-08 · Updated 2026-10-21

No one free AI agents course is best for everybody, because they teach different frameworks and expect different starting points. If you know basic Python and want a structured free path, the Hugging Face Agents course is the most complete one with free certificates. If you want Andrew Ng's teaching, DeepLearning.AI's Agentic AI course is free to watch, with the labs and certificate behind a paid plan. If you work in Azure, Microsoft's 18-lesson course is the fit. If you want to start from zero and build one assistant end to end, LogicWiz is built for that. The details are below, every figure comes from each course's own page, and the table near the end adds which courses cover the A2A protocol and which give a certificate.

Disclosure: I run the LogicWiz GenAI course, so I have an interest. I read every page in the first week of October 2026, checked them again on 14 October 2026, and quote only what they print. Where a page did not say something, I note that. Courses change, so check the page before you commit.

How did I compare free AI agents courses?

People mean quite different things by an "agents course", so I asked four questions of each one.

First, what you need before you start. Some begin at Python basics and some expect intermediate Python and a working knowledge of LLM APIs.

Second, which framework it teaches. LangGraph, smolagents, Microsoft Agent Framework and the Model Context Protocol are separate skills.

Third, what "free" includes. Free to watch, free with an account, free during a beta and free including the certificate are four different offers.

Last, what you build. One connected project teaches something different from a set of separate exercises.

Hugging Face Agents course

The course page calls it a free course "from beginner to expert". It has five chapters (onboarding, agent fundamentals, frameworks, use cases and a final assignment) plus three bonus units on fine-tuning, observability and a game agent. It asks for basic Python and basic LLM knowledge, and a Hugging Face account. The frameworks it teaches are smolagents, LangGraph and LlamaIndex. Each chapter is designed for one week at roughly three to four hours.

It offers two free certificates: one for the fundamentals unit, and one for completion that needs a use-case assignment and the final challenge.

It suits someone who can already code a little and wants a guided path that compares frameworks, with a community around it.

DeepLearning.AI: Agentic AI and AI Agents in LangGraph

The Agentic AI course, taught by Andrew Ng, is 9 hours 55 minutes across 31 video lessons, last updated on 31 August 2026. It covers reflection, tool use, planning and multi-agent workflows, and it asks for intermediate Python and a basic understanding of LLMs and APIs. The page says the videos and community forum are free. Quizzes, interactive labs, graded assignments and the certificate sit in the Pro membership, listed at $25 a month billed annually or $30 a month billed monthly.

AI Agents in LangGraph, made with LangChain and Tavily, is 1 hour 42 minutes in nine video lessons. Its page says access is free while the learning platform is in beta, with the certificate and graded quiz on Pro, and it asks for intermediate Python.

DeepLearning.AI also has A2A: The Agent2Agent Protocol, built with Google Cloud and IBM Research: 1 hour 27 minutes in 14 lessons, at intermediate level. Its page does not say the course is free, and the certificate is a Pro feature.

This pair suits developers who learn well from video and want the clearest explanation of agent design patterns. You have to be fine with the free tier being watch-only.

Microsoft AI Agents for Beginners

This is a free, MIT-licensed GitHub course with 18 lessons. The lesson table runs from agent use cases and design patterns (tool use, planning, multi-agent, metacognition) through agentic RAG, protocols (MCP, A2A and NLWeb), context engineering, memory, computer-use agents, deployment and securing agents. The code is built around the Microsoft Agent Framework and Microsoft Foundry, and the page says a Foundry (Azure) account is required. No certificate is mentioned. Its protocols lesson is the only free, official material I found that teaches A2A without a paywall on any part.

Best for: developers who will build on Azure and want the most complete topic list in a single free syllabus.

LangChain Academy: Introduction to LangGraph

LangChain's own course is free, with six hours of video across 55 lessons, and you enrol through a register or sign-in step. It covers graph basics, state, memory, human-in-the-loop, streaming, breakpoints, parallelization, sub-graphs and long-term memory. A second free course, Project: Deep Agents, adds two hours and 30 lessons on long-running agents, todo lists, files and subagents.

If you already know you want LangGraph, this is the maker's own material.

Anthropic Academy: MCP, agent skills and subagents

Anthropic's free courses on its Skilljar site include Introduction to Model Context Protocol, an advanced MCP course, Introduction to Agent Skills and Introduction to subagents. They need a free Skilljar account. The MCP courses ask for working Python and a basic understanding of JSON and HTTP, and the skills and subagents courses are about Claude Code. The academy says it uses Skilljar so it can issue completion certificates, but the course pages do not say which courses award one.

Best for: developers who want to build MCP servers or work inside Claude Code, as a complement to a general agents course.

freeCodeCamp on YouTube

Two long videos are worth knowing. "AI Agents For Beginners" is three hours, published on 7 July 2026, and builds agents from the core ideas of LLMs, tokens and temperature through workflows, guardrails and sandboxing. "Agentic AI using LangGraph" is 24 hours, published on 30 July 2026, and goes from single agents to multi-agent systems, RAG, streaming and production deployment. Both are free on the channel, and neither comes with a certificate.

They work if you like following along with one long video at your own pace.

Analytics Vidhya: AI Agents Learning Path

The learning path is listed as free, with a login by one-time code, 12 hours of self-paced modules and projects that simulate enterprise agents. It starts with Python fundamentals and covers agent design, prompt engineering, memory, tool use, RAG, orchestration, evaluation and deployment, with LangGraph, CrewAI, AutoGen and Amazon Bedrock named. The page says you get a certificate when you complete the course and pass the final assessment.

Best for: beginners who like a structured path that compares several frameworks.

Google Skills: Introduction to Agents

Google's path "Introduction to Agents and Google's Agent Ecosystem" has five activities, including an introductory course on how AI agents work, and uses labs in the Google Cloud console. A second path, "Develop Agents with Agent Development Kit (ADK)", has eight activities and covers building agents with the Agent Development Kit. Both end in a skill badge rather than a certificate. The pages I read did not print a price, so I do not claim one. Both need you to sign in.

It makes sense if you are targeting Google Cloud and want hands-on practice in the console.

LogicWiz GenAI course

The course page says every lesson is free, with no upgrade needed to read or run any of them. The early chapters, which teach Python foundations, open without an account, and a free sign-in keeps your progress in the later ones. Across the whole course you build one assistant, Nova: it starts as a first Python script, gains prompting and an agent loop, then retrieval over a private archive, then multi-agent orchestration with LangGraph, and finally a voice interface deployed as an API. Tool use, RAG (embeddings, chunking, indexing and retrieval), MCP and A2A, evaluation and guardrails are covered, and lessons come with labs that run in the browser. At the end there is a free certification exam (60 questions, 75 minutes, 70% to pass) with a certificate anyone can verify online. Most labs use your own OpenAI API key, and do you need an OpenAI API key to learn GenAI explains what that costs and the alternatives.

If you are not sure which chapter fits you, the Zero to AI Agent Line draws the course as a metro line: pick where you are today and it shows the chapters still ahead.

Best for: a complete beginner, or a developer who has never built an agent, who wants to grow one project from start to finish.

Which free AI agents courses cover A2A and give a certificate?

Checked on 14 October 2026 against each course's own page. "Not covered" means the syllabus I read does not list A2A.

How do you choose a free AI agents course?

I would choose by where you are starting from, since the best-known names mostly assume you can already code.

If you are new to Python, pick a course that teaches it, such as LogicWiz or the Analytics Vidhya path, and do not start with a course that assumes intermediate Python.

If you can code and want the strongest explanation of agent patterns, watch the DeepLearning.AI Agentic AI course, and add the LangChain or Hugging Face course to build.

If you know where you will work, choose the matching stack: Azure with Microsoft, Google Cloud with Google, Claude Code or MCP servers with Anthropic.

If you need a certificate, check each page. Hugging Face lists two free certificates, Analytics Vidhya gives one after a final assessment, Anthropic's academy mentions completion certificates without saying which courses award one, DeepLearning.AI's sit behind Pro, and LogicWiz has a free certification exam at the end of the course. Before you choose on the certificate alone, read what recruiters can check on a free GenAI certificate.

Whatever you pick, build something of your own afterwards. Following a course's agent and designing one yourself are different skills, and a second project outside any course is how you find out whether you have the second. For the retrieval side of agents, the plain-Python RAG walkthrough and our wider comparison of free GenAI courses are good next reads.

Common questions

What is the best free AI agents course?

It depends on your starting point. Hugging Face's Agents course is the most complete free path with free certificates, DeepLearning.AI's Agentic AI course is the strongest video teaching with labs behind a paid plan, and LogicWiz builds one assistant from Python basics. Pick the one that matches what you already know.

Do I need to know Python before an AI agents course?

For most of them, yes. DeepLearning.AI asks for intermediate Python, Hugging Face for basic Python and LLM knowledge, and Anthropic's MCP courses for working Python. LogicWiz and the Analytics Vidhya path start with Python fundamentals.

Is DeepLearning.AI's Agentic AI course free?

Its page says the videos and community forum are free. Quizzes, interactive labs, graded assignments and the certificate are part of the Pro membership, listed at $25 a month billed annually or $30 billed monthly.

Which free course teaches LangGraph?

LangChain Academy's Introduction to LangGraph (6 hours of video, 55 lessons, free with sign-in), DeepLearning.AI's AI Agents in LangGraph, freeCodeCamp's 24-hour video, Hugging Face's Agents course and the LogicWiz course all teach it.

Is there a free MCP course?

Yes. Anthropic Academy has free MCP courses with a Skilljar account, Microsoft's agents course covers MCP among its protocols, and LogicWiz covers MCP in its GenAI course.

Is there a free course on the A2A protocol?

Microsoft's AI Agents for Beginners has a free lesson on agent protocols that covers MCP, A2A and NLWeb, and the LogicWiz GenAI course has a chapter on MCP and A2A. DeepLearning.AI has a dedicated A2A course built with Google Cloud and IBM Research, but its page does not say it is free and the certificate needs Pro.

Which free AI agents course gives a certificate?

Hugging Face's Agents course gives two free certificates, Analytics Vidhya's path gives one after a final assessment, and LogicWiz has a free certification exam (60 questions, 75 minutes, 70% to pass) that anyone can verify online. For two-line answers to questions like these, see our quick answers page.