AI Agents Crash Course: a free 5-day course on building agents
This free, MIT-licensed course takes learners from no experience to deploying multi-agent systems in 5 days. It uses Buddy, our production Google Ads agent, as the working example throughout.
- Author
- John M. Williams, founder of It All Started With An Idea
- License
- MIT
- Built
- Feb 2026, v5.0 Mar 2026
- Format
- Web, Word, PDF and Markdown
- Length
- 42 pages, 5 days plus setup and capstone
The idea
Most AI agent tutorials assume you already code, or they stop at a toy demo. Beginners and working marketers both get lost.
We wanted a course that works for every level and ties each idea to a real agent running in production.
How we built it, step by step
- Planned 5 days plus setupThe course covers AI agents, tools and MCP, context engineering, agent quality, and production and multi-agent systems, with a setup day first and a capstone at the end.
- Wrote 3 learning pathsObserver takes 2 to 3 hours a day, Builder 4 to 6 and Architect 6 to 8, so people can go as deep as they want.
- Tied concepts to BuddyEach concept maps to a real part of Buddy, such as its sub-agents and its CONFIRM step for safe writes.
- Ended each day with a gameLearners build a small game by prompting at the end of each day, in Tetris, Zelda, football, Super Mario Bros and Duck Hunt styles, then an AI Game Master RPG as the capstone.
- Added checkpoints and quizzes'What You Should See' checkpoints, 6 quizzes with 21 questions, XP levels, 6 badges and daily streaks keep people on track. Progress saves in the browser.
- Linked further readingThe course builds on 5 Google and Kaggle agent white papers, links 13 free Anthropic Academy courses made by Anthropic, and includes 16 references and a 22-term glossary.
- Published in several formatsThe 42-page course is available as a web version, Word, PDF and Markdown, with the README in 7 languages.
What it does
- 5 days plus setup day and capstone
- Observer, Builder and Architect paths
- 5 daily game builds and a capstone RPG
- 6 quizzes with 21 questions
- XP, badges and streaks
- 22-term glossary
- Web, Word, PDF and Markdown versions
How people find it
- Free web course on ahmeego.com
- MIT-licensed source on GitHub, README in 7 languages
- Links to free Anthropic Academy courses
- Built on Google and Kaggle agent white papers
Questions people ask about this work
Where can I learn to build AI agents for free?
Our AI Agents Crash Course is free and MIT-licensed. It runs 5 days, from your first agent to multi-agent systems, with paths for beginners and experienced builders. It also links 13 free Anthropic Academy courses.
What is MCP and how do AI agents use tools?
MCP, the Model Context Protocol, is a standard way to give an agent tools it can call, such as an API or a database. Day 2 of the course covers tools and MCP using Buddy's Google Ads tools as the example.
What is context engineering for AI agents?
It is the work of deciding what information an agent sees at each step, including instructions, memory and retrieved data. Day 3 of the course covers it with examples from a production agent.
Services in this project
Related projects
Buddy by Ahmeego
Buddy is a chat-based agent that reads a Google Ads account, stages changes, asks you to confirm them, makes them through the Google Ads API and checks the result.Open-source tool · AI agent toolkit for advertisingAdvertising Hub and Agent Toolkit
This is our open-source toolkit for running advertising work through AI agents.
ContextOS
ContextOS is a Python MCP server and CLI that gives AI agents memory, routes retrieval by how fast data changes and adds a reasoning step between retrieval and generation.
Built and written up by John M. Williams, founder of It All Started With An Idea in Queen Creek, AZ. 15+ years and $350M+ in managed ad spend.