A Staged Learning Map from LLM Basics to Reliable Agent Systems

You've probably typed a prompt into a chat window, gotten a decent answer back, and then wondered what actually separates a chatbot from an "agent." Or maybe you've read enough threads about tool use and agent loops that you want to build one yourself—but every tutorial assumes you already know three things you don't. The gap between "I can call an LLM" and "I can ship a reliable agent system" is real, and it's mostly a sequencing problem. awesome-agentic-ai-zh is a learning roadmap built to close that gap in order, one stage at a time.

What It Does

This is a curated learning map for agentic AI, written primarily in Traditional Chinese with Simplified Chinese and English versions available. It's not an encyclopedia—the README is explicit that when you need full chapters, it points you to official docs, the Datawhale Hello-Agents project, or relevant cookbooks rather than re-writing them. What it provides instead is a structured path, selected resources, and small runnable exercises.

The structure runs from Stage 0 through Stage 8, plus a Stage 7.5 advanced reading stop—ten learning stations in total. Stages 0–2 are a shared foundation: getting your tools set up, understanding what LLMs, tokens, and context actually are, and calling a model from either the cloud or your own machine. After that the map forks. There's a Track A for CLI power users who want to get work done with tools like Claude Code, Codex, or OpenCode, and a Track B for people who want to build their own agents, tool loops, workflows, and services. The two tracks share later stages and then branch again into role-specific routes. There's also a separate route for everyday users who just want to use AI safely without writing code.

One detail worth noting: technical terms get explained in plain language the first time they appear, with the formal English term kept alongside. There's a glossary you can jump to when you forget something.

Why It's Cool

• It solves sequencing, not just curation. Awesome-lists are everywhere. What's harder to find is something that tells you what to learn first, and what you can safely skip. The staged design (with an explicit note that people who already know Python, Git, and APIs can skip Stage 0) does that work for you.

• The routing table is genuinely practical. Instead of one linear path, you pick your entry point by what you want to do: complete tasks with a CLI agent, build your own agent systems, or just use AI safely day to day. That's a more honest model of how people actually arrive at this topic.

• The definition of "agent" is refreshingly strict. The README draws a clear line: a chatbot that answers once, or a script with every step hardcoded, isn't necessarily an agent. An agent pursues a goal, decides its next step, uses tools when needed, and continues, corrects, stops, or hands control back to a human—all within the rules and permissions it's given. Starting from that definition sets the right expectations for everything that follows.

• Reliability is treated as part of the curriculum, not an afterthought. The stated end goal is adding permissions, evals, human approval, observability, and failure recovery. Plenty of beginner material stops at "look, it called a tool!" This map keeps going.

• It's honest about scope. Pointing to external docs and cookbooks instead of duplicating them keeps the project maintainable and keeps you close to primary sources.

How to Try It

The fastest route is the online documentation site, which the README recommends for mobile reading:

https://wenyuchiou.github.io/awesome-agentic-ai-zh/

If you'd rather work locally:

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

Then open stages/00-foundations.md—or jump directly to the station that matches you:

1. Never written code before: start at Stage 0 (foundations), paired with the zero-background setup guide.
2. Already comfortable with Python, Git, and APIs: start at Stage 1 (LLM basics).
3. Not sure which track: check the Track A / Track B selection table in the README before committing.
4. Just want to use AI safely in daily life: go straight to the everyday users guide.

Either way, Track A and Track B readers should confirm Stages 0–2 before moving on. Each exercise also explains how to connect to a model via cloud or local paths, so you're not left guessing at setup.

The project is MIT-licensed and lives at github.com/wenyuchiou/awesome-agentic-ai-zh.

Final Thoughts

This is best suited to developers who've touched an LLM API but haven't built anything agentic yet, and to people who keep bouncing off scattered tutorials without a sense of progression. If you're already shipping reliable agent systems, you're past most of this. The honest caveat is language: the primary content is in Traditional Chinese, and while Simplified Chinese and English versions exist, that's worth confirming against your own reading comfort before you dive in. But if the sequencing problem is the one blocking you, a map that tells you where to start and what to skip is worth more than another list of links.