Agentic AI
Build an AI agent from an empty file: the loop, tools, memory, planning, safety and evals. Then plug in a real model.
- 00
Start here
What makes something an agent, and how you'll build one from an empty file.
- 01
A model is a function
Messages in, a message out. Roles, tokens, cost, and prompts you can test like code.
- 02
Structured output
Get JSON your program can use, from replies that aren't always tidy.
- 03
Tools
Give the model hands: describe tools it can call, run them, and report back.
- 04
The agent loop
Call the model, run the tools it asks for, send the results back, repeat until it's done.
- 05
Context engineering
The context window is a budget. Decide what the model sees, and what it doesn't.
- 06
Memory and retrieval
Give the agent knowledge it wasn't trained on: chunk it, find the right pieces, cite them.
- 07
Planning
Break big goals into steps, run them in the right order, and check the work.
- 08
Reliability
Networks fail, tools fail, models make things up. Agents that survive all three.
- 09
Safety
Agents act in the real world. Keep them from being tricked, and from doing what they shouldn't.
- 10
Evals
You can't improve what you don't measure. Test agents like software.
- 11
Many agents
Routers, orchestrators and workers, and when one agent is better than five.
- 12
Real models
Swap the stand-in for Claude on your own computer. Your agent code stays the same.
- 13
Capstone: a research agent
Everything together: an agent that searches a handbook, reads, answers with a citation, and gets graded.
Final exam and certificate
Unlocks when you finish the course: 39 lessons to go.