The order I'd learn to build AI agents if I started from zero
An agent is a loop: Claude reads the job, calls a tool, reads the result and goes again until it's done. If I had to learn to build them from zero, this is the order I'd go in, with a free lesson or guide for every step.
The courses are free. Running the code needs an API key from the Claude Console: new accounts get a small free credit, then you pay for what you use. Checked on 2 Nov 2026
Prompting
Brief it like a brilliant new hire.
The API
Your own code sends the request.
- Claude Platform 101 (opens in a new tab)13 lessons, 1.5 hours. Checked on 2 Nov 2026
- Going further: Building with the Claude API (opens in a new tab)67 lessons, 9 hours. Checked on 2 Nov 2026
Tool use
Claude asks, your code runs it, the result goes back.
- Tool use with the Messages API (opens in a new tab)9 lessons, 1.5 hours. Checked on 2 Nov 2026
MCP
Plug it into other services without writing every tool.
- Introduction to Model Context Protocol (opens in a new tab)10 lessons, 1 hour. Checked on 2 Nov 2026
The agent loop
Think, act, read, repeat.
Skills
Write a job down once.
- Introduction to agent skills (opens in a new tab)6 lessons, 1 hour. Checked on 2 Nov 2026
- Examples: anthropics/skills on GitHub (opens in a new tab)
Subagents
Helpers with their own memory.
- Introduction to subagents (opens in a new tab)4 lessons, 45 minutes. Checked on 2 Nov 2026
Evals
Check what happened, not what it says happened.
Five of the steps are free Anthropic courses. The other three are their own docs. Checked on 2 Nov 2026
Want the next one?
I send what I'm learning and building with AI every Monday and Thursday. It's free.