bryanyzhu/agentic-ai-system-course

This repository is a 22-chapter course on how to design, build, and run AI agent systems, written to be read alongside an AI coding tool rather than as a standalone textbook. The main idea is that you open the course material in something like Claude Code or Codex and use that AI tool as a tutor and collaborator as you work through the chapters, asking it questions and having it translate concepts into working code for your own project. An agentic system, as the course defines it, is an AI setup that can pursue goals on its own by planning, making decisions, using external tools, remembering things across sessions, and adjusting based on what happens, rather than just responding to a single prompt and stopping. The course covers the progression from a single tool call all the way to multi-agent coordination and self-evolving systems, with a final chapter structured as a design canvas for your own project. The course is described as a skeleton: it covers load-bearing concepts and architectural patterns rather than specific frameworks or step-by-step tutorials. The goal is for the material to age slowly, since framework APIs change frequently but architectural patterns do not. The repository never tells you to use a particular library, instead, your AI partner is supposed to suggest the stack that fits your specific project. The material is aimed at both technical and non-technical readers. A technical reader would clone the repo, open it in an IDE, and use an AI agent to go deeper on each chapter. A non-technical reader would use an AI tool to generate code and designs based on the course concepts, asking the AI to explain each piece in plain language. An optional shell script clones four open-source reference agent systems into a references folder for grounded examples. The repository also includes a CLAUDE.md and an AGENTS.md file, which are identical copies of a behavioral guide for whichever AI tool you use to navigate the course.