100 Questions of AI Agent

Free, continuously updated books of practical, judgment-first questions covering AI agents, Large Language Models (LLMs), RAG, vibe coding, agent engineering, etc.

⭐ Star this repository

If this collection saves you an afternoon of scattered searching, a star is the easiest way to say thanks — and it helps the next person find it too. No account beyond GitHub needed.

Book list

Everything in this repository lives on GitHub. Star or watch the repo to follow new books — I post summaries on X, but the source always lands here first.

Contact

• WeChat: AIRay1015
• GitHub: akira82-ai
• X: — I post updates there, but the full books and newest chapters always land on GitHub first. Star here so you never miss one.

What this repository is for

This repository is a growing collection of 100 questions books on the practical side of AI: Agents, LLMs, RAG, vibe coding, evaluation, and Agent engineering.

The idea behind it is simple.

On one hand, I wanted a more engaging way to collect material. When a topic is just scattered notes, it easily turns into a pile of links, screenshots, and fragments that I am too lazy to revisit later. The 100 questions format forces structure, which makes reading, searching, organizing, and writing far more actionable.

On the other hand, I wanted to use the process to crystallize my own thinking, judgment, and hard-won lessons. What is truly valuable is rarely “I have read about this concept”; it is “how I understand it”, “where I stepped on a rake”, and “why I later changed my mind”. Left unorganized, that dissipates quickly.

So this repository is both a knowledge base and a discipline: it pushes me to write my judgments down clearly. The end goal is not a stack of summaries, but a set of topic documents that carry the traces of real thinking.

Who this is for

• AI / Agent engineers who need a structured map of the field, from prompt design to production deployment.
• Product managers, founders, and operators shipping AI features and trying to tell hype from reality.
• Anyone building with LLMs — vibe coders, forward deployed engineers, and self-learners who want hard-won judgment, not just concept summaries.

Topics covered

Seventeen books, 1700 questions, covering the full AI Agent practitioner stack plus essential developer tooling:

• Loop Engineering and Graph Engineering — how to orchestrate and supervise multi-agent systems.
• Codex and vibe Coding — building software with agents, from first prototype to production.
• The AI Product Manager and the Forward Deployed Engineer (FDE) — shipping AI inside real organizations.
• WorkBuddy and Lark CLI — agent-native office and CLI tooling.
• GitHub, Obsidian, and Harness Engineering — developer workflows, personal knowledge systems, and the systems around agents.
• Agent Skills — giving agents reusable capabilities: SKILL.md authoring, progressive disclosure, triggers, host compatibility, and skill security.
• Context Engineering — curating what enters the model’s window: attention budgets, context rot, compaction, memory systems, cache economics, and evaluation.
• Doubao Work — delegating real work to an AI coworker: capability boundaries, task delegation and verification, subscription economics, Feishu-powered team rollout, and the office-agent turf war.
• Eval Engineering — measuring and improving AI quality end to end: error analysis and golden sets, LLM-as-judge calibration, agent outcome vs trajectory metrics, production monitoring and eval debt, and the benchmark trust crisis.
• Jev and System One models — delegating fast, structured decisions to a model that never generates text: the Choice/Score/Noul primitives, atomic question decomposition, confidence thresholds and calibration validation, real cost arithmetic, and the open-source alternative ecosystem.

Themes running through all of them: RAG, evaluation and guardrails, context engineering, agent architecture, and the gap between a demo and something people actually use.

How to use this repository

Treat it as a directly readable topic library.

Read it to learn, to build a quick overall picture of a topic, or to distill the questions, structures, and source paths into your own articles, courses, checklists, or research frameworks.

If you are building Agents, content systems, or knowledge bases, or writing your own skills, this repository works as a content reference. Many of the topic-breakdown methods, chapter organizations, question designs, and source-to-text paths are reusable.