A single README that sorts agent hype from real engineering
An opinionated todo list that drags AI agent development from demo culture back to software engineering.

What it does This repository is essentially one meticulously maintained README that functions as a self-study degree in agent engineering. It breaks the path from “what is an agent?” to production-grade systems into numbered stages, each with reading assignments, concrete deliverables, and curated open-source references. The maintainer, a Datawhale member, treats it as a living document that distills community knowledge, official blogs, and papers into an executable checklist rather than a passive link dump.
The interesting bit The syllabus is deliberately contrarian: it explicitly warns learners not to waste energy on old “crew/role-play” multi-agent frameworks that have devolved into templates. Instead, it steers people toward the unglamorous but critical work of harness engineering—tool protocols, permission gates, context compaction, CI, and evaluation—arguing that an agent without traceability and safety boundaries is just a demo.
Key highlights
- Stages 0 through 7, from building a minimal agent loop in under 150 lines to evaluation, observability, and browser automation.
- Each stage ends with a concrete deliverable, such as a debuggable harness demo, a reusable
SKILL.md, or a research assistant that cites sources. - Heavy emphasis on modern protocols and infrastructure: MCP, A2A, ACP, Claude Code, OpenClaw, and LangGraph.
- Curated project references at every level, explaining why a specific repo fits that stage rather than simply listing it.
- Explicit anti-priorities: old role-play frameworks are noted as worth knowing but not worth mastering.
Caveats
- It is a documentation project; there is no runnable code in the repository itself beyond the rendered README.
- The file is long enough that GitHub truncates the view in-browser, so later stages require clicking to expand.
Verdict Grab this if you are a developer who has played with LLM APIs and now wants a structured, engineering-first path to building reliable agents. Skip it if you are looking for a drop-in framework or a quick copy-paste starter template.
Frequently asked
- What is datawhalechina/Agent-Learning-Hub?
- An opinionated todo list that drags AI agent development from demo culture back to software engineering.
- Is Agent-Learning-Hub open source?
- Yes — datawhalechina/Agent-Learning-Hub is open source, released under the MIT license.
- What language is Agent-Learning-Hub written in?
- datawhalechina/Agent-Learning-Hub is primarily written in HTML.
- How popular is Agent-Learning-Hub?
- datawhalechina/Agent-Learning-Hub has 5.6k stars on GitHub and is currently accelerating.
- Where can I find Agent-Learning-Hub?
- datawhalechina/Agent-Learning-Hub is on GitHub at https://github.com/datawhalechina/Agent-Learning-Hub.