Course
Everything about agents in one place: how language models work, a 60-line loop, tools, MCP, retrieval, skills and hooks, teams of agents, security, evals and shipping, with Claude, OpenAI and Codex, and code that has been type-checked against the real SDKs.
92 chapters in 13 parts 55.1 hours 481 code samples 72 quizzes
13 parts · 92 chapters · 55.1 hours. Each chapter opens with a plain-English lead and ends with the key points and questions to test yourself.
These learning paths take the chapters you need from this course, and from the others, in a short guided route.
Start hereNo background needed
A 13-week program for a complete beginner: Python and data, your first models, how neural networks and language models work, and your first AI agent, with a hands-on lab every week.
13 weeks64 h75 chaptersBeginner
Developers
A 14-week program from the agent loop to a launched agent: tools, memory and retrieval (RAG), MCP, the Claude and OpenAI SDKs, many agents, evals, safety and shipping to production, with hands-on labs every week and a capstone project.
14 weeks60 h92 chaptersIntermediate
Hands-on, code first
A ladder of twelve projects that grow in ambition, each with the lab behind it: a spam filter, house prices, a digit recogniser, an image classifier, a recommender, a language model, three agents, a simulated robot arm and an ML capstone.
12 weeks72 h24 chaptersAll levels
Agent tooling changes every few weeks, so this course is checked, not just written. Every TypeScript sample is type-checked against the real SDK packages, and the parts that can run without a paid key (the MCP server and client, the vector search, the hook scripts, the orchestration, eval and safety code) were run, with a scripted model standing in for the API where one was needed. Where a feature is in beta, or a detail may have moved, the chapter says so. Last verified 2026-10-27 against:
Model names, prices and product details are deliberately kept light in the prose: they change too fast. If something here no longer matches a primary source, the source wins, and our corrections page is the place to tell us.