datawhalechina/agent-learning-hub
Developer toolsThis page was translated automatically. Technical details may differ from the original. View the original
AI Agent learning routes and database collection
Overview
A curated AI Agent learning roadmap for people who want to build useful, reliable agents instead of collecting random links. This repository maintains only one core display surface: the README. The goal is to compile excellent community shares, official blogs, papers, open-source projects, and real engineering experience into a followable AI Agent learning to-do list. - If you're a beginner: follow the 'Learning Todo List' from top to bottom, ticking off each completed item. - If you already know how to apply LLMs: start from Stage 2 or Stage 3, focusing on Agent loops, tool calls, evaluation, and engineering. - If you want to start a project: just look at the 'Project Ladder' and create one workable project for each level. - If you just want to find information: check 'Curated Resources' and prioritize official documentation and classic papers. The Agent field is changing rapidly. What is more worth investing in now is not the old "role-playing multi-agent framework," but these directions closer to real productivity: It is not recommended to focus on outdated crew/role-play frameworks that have become templates. They can be understood, but should not be the main thread. - [ ] Distinguish between chatbot, workflow, agent, and multi-agent. - [ ] Understand the basic cycle of agents: observe -> think -> act -> observe. - [ ] Know when not to use agents: When tasks are predictable, workflows are stable, and ordinary scripts can solve them, agents actually increase uncertainty. - [ ] Read Anthropic: Building effective agents. - [ ] Read OpenAI: A practical guide to building agents. Output: Write a short note answering, "Why does my scene need an agent instead of a regular workflow?" - [ ] Uses an LLM API to complete normal conversations. - [ ] Allows the model to output structured JSON. - [ ] Defines a tool function, such as search, calculator, read_file. - [ ] Parses the model's tool call / function call.
Overview (machine translated)
A curated AI Agent learning roadmap for people who want to build useful, reliable agents instead of collecting random links.
This repository maintains only one core display surface: the README. The goal is to compile excellent community shares, official blogs, papers, open-source projects, and real engineering experience into a followable AI Agent learning to-do list.
Maintainer
Curated by Chen Sizhou (Datawhale member)
How To Use
- If you’re a beginner: press the ‘Learning Todo List’ from top to bottom, checking off each completed item.
- If you already know how to apply LLMs: starting from Stage 2 or Stage 3, focus on supplementing agent loops, tool calls, evaluation, and engineering.
- If you want to start a project: just look at the “Project Ladder” and create one runnable project per level.
- If you only want to find information: read “Curated Resources” and prioritize official documentation and classic papers.
What To Learn Now
The Agent field is changing rapidly. What is more worth investing in now is not the old “role-playing multi-agent framework,” but these directions closer to real productivity:
| Priority | Learn | Why |
|---|---|---|
| 1 | Claude Code / Codex-style coding agents | Real codebases, shells, file editing, testing, permissions, and context compression are the best agent engineering samples. |
| 2 | Agent harness engineering | A large part of an agent’s capabilities comes from harnesses: tool protocols, permissions, status, feedback, playback, CI, reviews |
| 3 | OpenClaw / Hermes-style personal agents | Long runtime, local priority, cross-application, memory, skills, message entry—more like a “personal operating system.” |
| 4 | Skills / MCP / A2A / ACP | skills is responsible for capability reuse, MCP connects tools, A2A connects agents, ACP connects to host applications. |
| 5 | Evaluation and safety | Agents without eval, trace, or permission boundaries can only be considered demos. |
It is not recommended to focus on outdated crew/role-play frameworks that have become templates. They can be understood, but should not be the main thread.
Learning Todo List
Stage 0: Understand What An Agent Is
- [ ] Distinguishes between chatbot, workflow, agent, and multi-agent.
This page was translated automatically. Technical details may differ from the original.
Recommended Tools
Try a different keyword or remove a filter.
Install
npx skillfish add datawhalechina/agent-learning-hub