Learn Agentic AI on a 7k-Star Trilingual Roadmap

Most agentic AI material throws docs, courses, and hot takes at you in no particular order. awesome-agentic-ai-zh is a MIT-licensed learning roadmap that sequences the whole path — from what an LLM is to systems that run reliably in production — and it ships the same map in Traditional Chinese, Simplified Chinese, and English.

Why This Skill Matters

The repo organizes agentic AI into 8 thematic stages plus Stage 0 (setup) and Stage 7.5 (advanced reading) — 10 learning stations in total. Two tracks split the journey: Track A is for CLI power users who want to get work done with tools like Claude Code, and Track B is for builders who want to write their own agent loops and workflows. Every stage states its core question, its key vocabulary, and the completion criteria you should hit before moving on.

There is also a role layer: five branch guides for researchers, developers, teachers, knowledge workers, and everyday users, so you can enter the map from the job you actually do.

Installation

git clone https://github.com/WenyuChiou/awesome-agentic-ai-zh.git
cd awesome-agentic-ai-zh

Prefer not to clone? The same map is readable online at wenyuchiou.github.io/awesome-agentic-ai-zh/.

Real Workflow: Pick Your Entry Point and Finish a Stage

1. Answer one question from the README's choice table: do you want to drive CLI agents (Track A), build agents yourself (Track B), or use AI daily without coding (the everyday-user guide)?
2. If you have never programmed, start at Stage 0; if you already know Python, Git, and APIs, jump straight to Stage 1.
3. Open your stage file and read its core question and key vocabulary first.
4. Run the stage's starter.py — the exercises are designed to run offline first, before anything touches a model API.
5. Change one thing in the exercise, rerun the test, and confirm the completion criteria before advancing.

I know Python, Git, and REST APIs. Following the awesome-agentic-ai-zh
roadmap, which stage should I start from, and what is the first exercise
I should complete?

Track A readers follow A1 → A2 → Stage 5 → A3 → Stage 8: pick a CLI agent, build repeatable workflows, learn the Claude Code ecosystem (MCP, Skills, Plugins, Hooks, Subagents), then wire everything into real work safely. Track B readers walk Stages 3–8: tool use and your first agent loop, workflow graphs and frameworks, the same Stage 5 ecosystem hub, memory and RAG, then production engineering.

Tips

• The README estimates Track A at 8–10 weeks and Track B's main line at 16–22 weeks — roughly 5–7 months at 5–8 hours per week. Treat these as planning references, not deadlines.
• Each exercise has a success condition. The roadmap's own advice: do not move on until you hit it — reading is not doing.
• Change one thing at a time and rerun the test immediately, so you know which change caused which result.
• Keep the glossary open. Technical terms are explained in plain language on first use, then kept in English.
• After finishing Track A's A3 or Track B's Stage 7, start the capstone project and log progress in PROGRESS.md.

When Not to Use This

This is a sequenced map, not an encyclopedia. When a topic needs full chapters, the roadmap deliberately links out — to official docs, Datawhale's Hello-Agents, or the relevant cookbook — instead of rewriting them. If you are already shipping agents and only want copy-paste skill libraries for your own stack, a skills collection will serve you faster than a curriculum.

See the leaderboard for more skills.