Build an AI Agent
from Scratch
Start with basic Python. Finish with your own agent framework and a coding agent that fixes bugs on its own. You don’t need any experience with AI.
You leave with working code,
not just notes.
Your first agent
A model that calls tools in a loop.
Your own agent framework
Pluggable models and tools, with tracing and evals.
RAG, memory and MCP
Retrieval, planning, and an MCP server and client.
A coding agent
It fixes bugs on its own. Then you compare it with real frameworks.
If you can write a Python function, you can start.
- YESYou use AI tools and want to know what’s actually happening inside them.
- YESYou know basic Python, or you’re happy to learn it in Part 0.
- NOAny experience with AI or machine learning. We start from zero.
- TIPAlready fluent in Python? Skip Part 0 and start at Part 1.
9 parts, from Python
to a coding agent.
- 00Up next
Setup and Python primer
Everything you need from Python, with a way to skip it if you know it.
- 01Coming
LLM foundations
Tokens, model calls, prompting and structured output.
- 02Coming
The first agent
Tool calling and the agent loop.
- 03Coming
From script to framework
Pluggable models, tools, tracing, reliability and evals.
- 04Coming
RAG
Embeddings, vector search and better retrieval.
- 05Coming
Agent capabilities
Context, memory, planning and durable execution.
- 06Coming
MCP
Build an MCP server and client.
- 07Coming
Multi-agent and production
Multi-agent systems, security, model switching and cost.
- 08Coming
Capstone
Build a coding agent, then compare with real-world frameworks.
Watch
A short captioned video that opens with a 30-second recap.
Read
An article that walks through real code and the output it really produced.
Build
Exercises that check themselves, so you know when you’ve got it right.
Questions before you start
Do I need a paid API key?
Not to get started. The course runs a model on your own computer with Ollama, so you need at least 8 GB of memory and about 10 GB of free disk space. A Gemini API key comes in at chapter 31, when you learn to switch models.
I already know Python. Can I skip ahead?
Yes. Part 0 covers everything you need from Python, with a way to skip it if you already know it.
How long does it take?
About 85 hours across 35 chapters. It’s fully self-paced, so you choose how fast to go.
Do I need to know Git?
No. You can download the course code from GitHub as a ZIP file, and you don’t need a GitHub account.
What are the videos like?
They’re silent, with captions on screen, and each one opens with a 30-second recap.
Can I reuse the material?
Yes. The code is under the MIT License, and the articles and videos are under CC BY 4.0, so you can reuse them with credit.
Open your first black box.
Up next: Part 0: Setup and Python primer. Follow along so you catch it the day it’s released.