Agent engineering for developers who outgrew prompt demos

What it does

zero2Agent is a GitHub-hosted curriculum—131 articles across 12 modules—that walks developers through Agent engineering from first principles to deployment. It pairs deep dives into frameworks like LangGraph, OpenAI Agents SDK, and DeepSeek Harness with a 500-question “Green Book” of real interview problems from companies like Ant Group, ByteDance, and Tencent. The whole thing is published as a static site, with code derivations starting from as little as 60 lines of Python.

The interesting bit

Most tutorials stop at a working demo; this one treats “why the demo breaks in production” as a first-class topic, dedicating modules to Loop/Infra, context compression, and sandboxed runtimes. The interview section is equally methodical, sorting 757 questions by real-world interview frequency and contrasting “novice” versus “expert” answers so you can see exactly how shallow your understanding is.

Key highlights

• 131 articles covering 12 modules, from Agent basics to training and deployment.
• Framework autopsies: LangGraph, Semantic Kernel, Eino, Mastra, AutoGen, and others, plus first-principles derivations like a 60-line OpenClaw core.
• Deep dives into specific runtimes: Claude Code, Pi Coding Agent, Codex CLI, and DeepSeek Harness.
• “Agent Interview 500 Questions” PDF (CC BY-NC-SA), covering 17 dimensions with real questions from Alibaba, ByteDance, Tencent, Baidu, and more.
• A final capstone project (OfferPilot) building a 10-layer interview-diagnosis Agent from PRD to deployment.

Caveats

• The “Agent 应用实战” module is still listed as work-in-progress.
• Advanced modules like DeepSeek Harness and Pi pin their explanations to specific external API versions and commits, so details may drift as upstream evolves.
• Every article is in Chinese, which is great if you read it, less so if you don’t.

Verdict

Worth bookmarking if you are a Chinese-speaking developer trying to level up from prompt engineering to production Agent engineering—or if you have a big-tech AI Agent interview in the next 48 hours. Skip it if you are looking for a drop-in framework; this is a syllabus, not a library.