microsoft/ai-agents-for-beginners

18 Lessons to Get Started Building AI Agents

What is this project?

AI Agents for Beginners is a free, open-source course from Microsoft that teaches you how to build AI agents—software that can perform tasks autonomously using large language models. It’s designed for people who are new to AI agents and want a hands-on, structured way to learn.

What does it include?

The course has 18 lessons, each covering a different topic. Lessons include:

  • A written explanation (in the lesson’s README)
  • A short video
  • Python code samples that use Microsoft’s Agent Framework and Foundry Agent Service
  • Links to extra resources for deeper learning

Topics range from the basics (intro to AI agents, agentic frameworks) to advanced patterns (tool use, agentic RAG, planning, multi-agent systems, metacognition) and practical concerns (trust, production deployment, security, local agents, memory, and more).

Who is it for?

It’s for beginners who want to start building AI agents. If you’ve never worked with generative AI before, the README suggests first taking Microsoft’s Generative AI for Beginners course (21 lessons).

What technologies are used?

The code examples rely on:

  • Microsoft Agent Framework (MAF) – a framework for building agents
  • Microsoft Foundry Agent Service V2 – a cloud service for running agents (requires an Azure account)

Some samples also support OpenAI‑compatible providers like MiniMax, which offers large‑context models.

How do I get started?

  1. Fork the repository to your own GitHub account.
  2. Follow the Course Setup instructions (in the 00-course-setup folder) to configure your environment.
  3. Run the code examples as you go through the lessons.

If you get stuck, you can join the Microsoft Foundry Discord channel for help.

Languages

The course is translated into 50+ languages (Arabic, Chinese, French, German, Hindi, Japanese, Spanish, and many more). Translations are maintained with a tool called Co‑op Translator.

Community and contributions

  • You can star and fork the repo.
  • Contributions are welcome—you can report issues or submit pull requests.
  • The project follows the Microsoft Open Source Code of Conduct and requires a Contributor License Agreement (CLA) for contributions.

Other courses

The README also links to other Microsoft beginner courses (e.g., Generative AI, LangChain, Edge AI, MCP, etc.) if you want to explore related topics.

Related

  • Project
    rohitg00/ai-engineering-from-scratch

    A free, MIT-licensed AI engineering curriculum with 542 hands-on lessons and 20 phases across Python, TypeScript, Rust, and Julia. Learners implement model internals, retrieval pipelines, and agent runtimes; test them; and keep reproducible evidence. Includes three scaffolded projects with local graders, study routes for Model Context Protocol and Agent Skills, certification tracks, and a web-based tutor that can run through a coding agent.

  • Project
    usestrix/strix

    Strix is an open-source autonomous AI penetration testing platform that scans code and applications to find, validate, and help fix security vulnerabilities.

  • Project
    headroomlabs-ai/headroom

    Headroom is a context compression layer for AI agents and LLM applications. It compresses prompts, tool outputs, logs, RAG chunks, and conversation history locally before they reach the model, cutting token usage by 30–90% while preserving accuracy and allowing reversible retrieval of original content.

  • Project
    ComposioHQ/awesome-claude-skills

    awesome-claude-skills is a curated directory of 1,000+ reusable 'Claude Skills' — instruction packages (a SKILL.md file plus optional scripts) that teach AI agents how to do specific jobs, from editing Word docs to running browser tests. Skills load progressively so an agent can carry hundreds without blowing its context window. The list spans 10 categories and links to external repos, mostly from the community. Composio, the maintainer, also ships an MCP Gateway and a 'connect-apps' plugin so Claude can perform real actions across 1,000+ integrations (email, Slack, GitHub, etc.), with a 3-step quickstart in the README.