跳转至

AI Engineering — English Guide

AI 工程 (AI Engineering) is an open-source book systematically compiled from 4,273 first-hand articles — covering LLM fundamentals, Agents, Harness engineering, MCP, RAG, security, training and inference. Every entry traces back to a primary source, carries a difficulty rating, and the corpus is re-curated daily by an automated pipeline. License: CC BY-SA 4.0.

What is available in English

  • Agent Engineering: Harness, Loop, Graph — a complete 13-chapter book, fully in English. This is the deep dive; start here.
  • This guide page.
  • Site-wide machine translation — turn it on below and every page you open is translated to English in your browser, on the fly. The original Chinese content is never modified.

The curated 20-chapter course and the 1,695 entries are written in Chinese; use the translate icon in the header (it becomes 中文 while the mode is on) to switch back.

🌐 Turn on site-wide English translation

Agent Engineering: Harness, Loop, Graph

A hands-on handbook for senior engineers: from probability models to reliable systems. It dissects the agent loop and its five organs, argues that the loop's quality ceiling is set by its verifier, maps the five layers of a harness, treats context as a supply chain to be engineered, and ends at evaluation, multi-agent orchestration and long-term governance.

Start reading — Preface →

Table of contents

Part Chapters
I · Foundations Ch 1 · From Generation to Delivery · Ch 2 · The Model's Boundary
II · Loop Engineering Ch 3 · Anatomy of the Agent Loop · Ch 4 · Verifier First · Ch 5 · Sustained Loops
III · Harness Engineering Ch 6 · The Harness, Five Layers · Ch 7 · Context Engineering · Ch 8 · Tools, Permissions, Guardrails · Ch 9 · State and Memory
IV · Graph Engineering Ch 10 · From Loop to Graph · Ch 11 · Multi-Agent Orchestration
V · Production & Governance Ch 12 · Evaluation and Observability · Ch 13 · Decay, Sovereignty, and the Engineer's Next Seat
Appendices Glossary · Checklists · References

How the site works

  • Sources: 4,273 first-hand articles (papers, engineering blogs, post-mortems) collected and deduplicated daily; 1,695 curated entries so far, still growing.
  • Difficulty ratings: every entry is graded ⭐ to ⭐⭐⭐⭐⭐, so you can read at your own level.
  • Traceability: every curated entry links back to its primary source — the site is a map, not a replacement.
  • Self-evolving: an automated loop collects, filters and re-integrates new material every day.

The Chinese main course, in five parts

The 20-chapter course below advances by closing capability loops, not by stacking articles. Its chapters are written in Chinese; the primary sources linked inside them are mostly English originals, so the trail is still followable.

Part Theme Chapters
Part 1 · Foundations Understand AI from zero Ch01 AI & LLM Basics · Ch02 Prompt & Context Engineering · Ch03 AI Tools & Product Landscape
Part 2 · Engineer Build AI applications Ch04 Agent Core Architecture · Ch05 Harness Engineering · Ch06 Memory & Context Management · Ch07 Skills, Tools & MCP · Ch08 Multi-Agent Collaboration · Ch09 AI Coding & Code Generation · Ch10 RAG & Knowledge Retrieval
Part 3 · Expert Deep into system architecture Ch11 Cloud Infrastructure & Deployment · Ch12 Security & Governance · Ch13 MLOps & Evaluation · Ch14 Data Engineering
Part 4 · Scientist Research & innovation Ch15 Training & Fine-tuning · Ch16 Inference Optimization & Architectures · Ch17 Multimodal & Generation · Ch18 Robotics & Embodied AI
Part 5 · Master Philosophy, safety & the future Ch19 Frontier Research & Theory · Ch20 AI Philosophy, Safety & the Future