AI Agents for Beginners
by Microsoft
Microsoft's 18-lesson agent curriculum, from design patterns through production security.
Overview
AI Agents for Beginners is Microsoft's open-source agent curriculum, published under the MIT licence on GitHub with roughly 72k stars and 23.8k forks, and also readable as a documentation site at microsoft.github.io. It runs to 18 numbered lessons plus a 00-course-setup module, and each lesson pairs a written README with Python code samples in a code_samples folder; most lessons also carry a short video, though the most recently added ones do not yet. The sequence starts with what an agent is and where agents are worth using, then covers agentic frameworks and the core design patterns — tool use, agentic RAG, planning, multi-agent and metacognition — before turning to operational material: building trustworthy agents, running agents in production, agentic protocols including MCP, A2A and NLWeb, context engineering, managing agentic memory, computer-use agents, deploying scalable agents, creating local on-device agents, and securing agents. The code has been migrated to the Microsoft Agent Framework running against Microsoft Foundry Agent Service V2, with OpenAI-compatible providers supported as an alternative, so an Azure account is needed for the Foundry path. Translations into more than 50 languages are generated by a GitHub Action, and a sparse-checkout recipe is documented for anyone who wants the repository without them. Microsoft directs absolute beginners to its 21-lesson Generative AI for Beginners course first, and the repo's own framing is that you can start at whichever lesson you need.
At a Glance
- Topic
- Agentic
- Level
- Beginner
- Format
- Course
- Cost
- Free
- Duration
- 18 lessons, self-paced
- Provider
- Microsoft
- Hands-on
- Yes — code/exercises
- Certificate
- None
What You’ll Learn
- ✓Judge where an agent is the right tool and where a plain workflow suffices
- ✓Implement the tool use, agentic RAG and planning design patterns in Python
- ✓Build multi-agent systems and add metacognition to a single agent
- ✓Connect agents over agentic protocols including MCP, A2A and NLWeb
- ✓Engineer context and manage agentic memory across long-running tasks
- ✓Deploy agents on Microsoft Agent Framework and Foundry Agent Service V2
- ✓Run local, on-device agents when cloud inference is not an option
- ✓Apply trust, safety and security controls to agents heading for production
Highlights
- •One of the most-starred agent curricula anywhere, at roughly 72k stars, MIT-licensed and free
- •Every lesson pairs a written README with runnable Python samples, and most add a short video
- •Actively maintained rather than abandoned: the samples were migrated to Microsoft Agent Framework on Foundry Agent Service V2
- •Covers the unglamorous production topics — security, scalable deployment, memory, protocols, local inference — that most beginner courses omit
- •Machine-translated into 50+ languages via a GitHub Action, with a documented sparse checkout to skip them
- •Be clear-eyed about the lock-in: the default path assumes Azure and Microsoft Foundry, and older third-party writeups describe an earlier, smaller 10-lesson version built on GitHub Models
Who It’s For
Best For
- ✓Developers comfortable with Python who are new to building agents
- ✓Teams standardising on Microsoft Agent Framework and Foundry
- ✓Self-learners who want a free, structured path they can dip into by topic
- ✓Engineers who prefer runnable code samples over lecture video
Prerequisites
- •Comfortable reading and writing Python
- •An Azure account for the Microsoft Foundry lessons, or an OpenAI-compatible endpoint
- •Generative AI fundamentals — Microsoft suggests its Generative AI for Beginners course first
FAQ
What is AI Agents for Beginners?
A free, MIT-licensed Microsoft curriculum of 18 lessons that each pair a written README with runnable Python samples and, for most, a short video. It covers agent design patterns, agentic protocols, memory and context engineering, and the production topics beginner courses usually stop short of — deployment, local inference and security.
Is AI Agents for Beginners free?
AI Agents for Beginners is free to access.
What level is AI Agents for Beginners for?
AI Agents for Beginners is aimed at a beginner audience. Recommended background: Comfortable reading and writing Python, An Azure account for the Microsoft Foundry lessons, or an OpenAI-compatible endpoint, Generative AI fundamentals — Microsoft suggests its Generative AI for Beginners course first.
How long does AI Agents for Beginners take?
Expect roughly 18 lessons, self-paced. Most learners work through it at their own pace.
What will I learn from AI Agents for Beginners?
You'll learn: Judge where an agent is the right tool and where a plain workflow suffices; Implement the tool use, agentic RAG and planning design patterns in Python; Build multi-agent systems and add metacognition to a single agent; Connect agents over agentic protocols including MCP, A2A and NLWeb; Engineer context and manage agentic memory across long-running tasks; Deploy agents on Microsoft Agent Framework and Foundry Agent Service V2; Run local, on-device agents when cloud inference is not an option; Apply trust, safety and security controls to agents heading for production.
Topics
Sources
This page was written from 3 sources, 2 on domains other than github.com.