Part of my teaching work: a one-hour, hands-on workshop where participants build an AI agent from scratch, no prior agent experience assumed.
The progression
The material is deliberately structured as three small, runnable steps:
- Talk to an LLM: the minimal LangChain call, demystified.
- Your first graph: modeling a flow with LangGraph nodes and edges.
- A real agent: a multi-node agent that generates HTML.
Everything runs on free tooling (Google AI Studio API key, uv for
environments) so anyone can follow along live or on their own afterwards.
Why this format
Agents are surrounded by hype and abstraction. Building one in an hour from plain Python makes the core ideas (state, nodes, edges, tools) concrete enough that participants can judge the hype for themselves.