AI agents, from zero to mastery —
free, and open to inspection.
Three tracks, four stages — pick the one that matches your background. Every lesson states its method, raw data and traceable sources; anything adapted from an open-source course is attributed with its licence at the end.
- Tracks
- 3
- Lessons
- 242
- Readable now
- 242
- Total hours
- 52.599999999999994h
- Tools catalogued
- 71
- Scenarios
- 30
The Agent Learning Roadmap: eleven stops from first model call to shippable agent
A sequenced curriculum adapted from the MIT-licensed awesome-agentic-ai-zh: goals, required reading, curated resources, and hands-on practice at every stop — with 95 lessons from this site mapped into the line. Checkboxes share the same progress as the tracks below.
Start here: which one are you?
Four starting points, each linking straight to one article rather than to an index page. If none fits, pick a track below instead.
Full learning paths
Both end in the same place: being able to judge for yourselfDeveloper Track
How do I build an agent and ship it
From your first LLM API call to building your own multi-agent framework and evaluation harness. Code, architecture diagrams, and numbers we measured ourselves.
Business Track
How much time can AI save my team, and how do we roll it out
No code required. From deciding whether AI is worth it, to handing over a real workflow — including cost models and rollout pitfalls.
Domain specials
Scenario recipes without a full course track. Pick by the work in front of you, not by your job title.
Which track is mine?
Not sure? Match yourself below. Choosing wrong costs nothing — the three tracks share the same articles, so your progress carries over.
| Your situation | Take this | What this track is like |
|---|---|---|
| I want to write the code myself, build an agent, and ship it | Developer Track | Code, architecture diagrams, and numbers we measured ourselves |
| I want to automate my team's processes without writing code | Business Track | Completable with no code; focused on ROI, adoption friction, and pitfalls |
| I have a domain-specific professional workflow to automate (investment research, legal contracts, healthcare, manufacturing…) | Finance Track | The finance track is the most complete today: every conclusion ships with verification gates and traceable sources, so numbers are not left to the model to guess. For other industries, start with the business track and then check the matching domain in the scenario library |
Four stages: beginner → expert
Every track uses the same stage definitions, so depth means the same thing whichever track you switch to.
- 1Beginner
You know the vocabulary, what it can do, and have run it once
- 2Intermediate
You complete standard tasks alone, pick tools, and grasp the principles
- 3Advanced
You design solutions, make tradeoffs, and handle messy real-world data
- 4Expert
You build your own frameworks, benchmark rigorously, and judge the frontier
71 tools, reviewed honestly
Every entry states who it is for and who it is not for. Amounts appear only after manual verification; otherwise we say so and link to the official pricing page.
Browse the tool library30 pain-point → tool → step recipes
No theory — just how to finish one thing. Each recipe states how long it takes, which tools to combine, and what you end up with.
ScenariosKeep going
Everything here is free, and every piece ships with its method, raw data and traceable sources. Not sure where to start? Learn the vocabulary first, then pick a track.
Start learning- Three learning tracks