The free open-source courses we read, and who they are not for.
Our curriculum was not written from thin air — the three tracks' outlines draw on the open-source courses and roadmaps below. This page lays out what we read: who each resource suits, what it teaches, and where it falls short. When choosing, “watch out for” is usually more useful than “what's inside”.
Resources
14
All free
14
Our adaptations
9 lessons
Last updated
2026-09-29
How we source content
01
What we may adapt, we adapt and attribute
Licences such as CC BY 4.0 permit adaptation. We rewrite in our own words, swap in local examples, and add our own measurements; the source, link and licence always appear at the end of the article.
02
Rights-reserved material we only review
Courses from DeepLearning.AI, LangChain Academy and similar are rights-reserved. For those we write only our own commentary and link out — never copying or rewriting the material.
03
Numbers only after we ran them
Where an external course conflicts with our own measurements, ours wins and the difference is noted. We never publish a number we did not run.
Licence information always defers to each repository's own LICENSE file; the labels here are for quick filtering only. If you spot a missing or incorrect attribution anywhere on this site, tell us and we will correct it.
Our take：The most complete free agent intro course available, running from "what is an agent" through tool use, RAG, multi-agent systems, and production deployment. Each lesson ships with video, notes, and runnable code.
Good for
Developers meeting agents for the first time
Anyone who wants a clear lesson order and assignments
Readers comfortable with well-structured English material
Watch out
Examples centre on Azure OpenAI and Microsoft Foundry, so other providers require swapping the SDK calls. Some lessons move faster than the code, which can lag the latest SDK.
Our take：A foundations course on generative AI covering prompt engineering, embeddings, RAG, evaluation, and responsible AI. More fundamental than the agents course — good for understanding LLM behaviour first.
Good for
Anyone who wants to understand the LLM itself, not just a framework
Business-track readers filling in technical foundations
Readers who need the responsible-AI and evaluation chapters
Watch out
Broad rather than deep, with relatively little on agents. If your goal is multi-agent systems, follow this with AI Agents for Beginners.
Our take：The only free agent course that issues a certificate, and it is framework-neutral — teaching smolagents, LangGraph, and LlamaIndex side by side so you can see how their abstractions differ.
Good for
Readers comparing agent frameworks
Anyone who needs assignments and a certificate to stay on track
Those past LLM basics and ready to build
Watch out
Difficulty jumps sharply between units; without solid basics you will stall at the multi-agent unit. The certificate requires completed, submitted assignments.
Our take：The most structurally complete Chinese-language agent development tutorial: LLM fundamentals, RAG, memory, tool use, function calling, multi-agent, LangChain / LangGraph, MCP, and agentic RL — with daily automated arXiv tracking.
Good for
Chinese-speaking developers
Readers wanting full depth from basics to the frontier
Anyone tracking the latest papers
Watch out
Large and fast-moving, with some chapters closer to paper summaries than runnable tutorials. You will need to set up environments and dependencies yourself.
Our take：Official short courses on LangGraph and agents, with the clearest treatment of state-machine-style agent orchestration. Best once you understand basic agents and want controllable orchestration.
Good for
Building human-in-the-loop flows
Agents that must be interruptible and resumable
Teams already in the LangChain ecosystem
Watch out
Tied to the LangChain / LangGraph ecosystem with heavy abstractions. If you prefer a lightweight build, take the concepts without adopting the framework. Rights reserved — we only review and link, never rewrite the material.
Our take：Short courses taught by the authors of each framework (LangChain, CrewAI, LlamaIndex, OpenAI, Google, Anthropic, and more), one to two hours each on a specific topic. The shortest path to understanding a framework's design intent.
Good for
Learners with limited time who want one topic at a time
Hearing design tradeoffs from framework authors themselves
Every course reflects one vendor's perspective and carries some promotional framing; cross-framework comparison is on you. Rights reserved — we only review and link, never rewrite the material.
Our take：Unusually combines agentic RAG (with knowledge graphs and caching) and multi-agent coordination (sub-agents, orchestration, MCP, A2A) in one course. For those past single-agent work and into coordination design.
Good for
Building agentic RAG
Hands-on MCP and A2A protocol work
Designing sub-agent division of labour
Watch out
Assumes fluency in Python and basic agent concepts, with no ramp-up. A2A is still evolving, so implementations may lag the latest spec.
Our take：The classic (non-generative) AI foundations course: search, symbolic reasoning, neural networks, NLP, and computer vision. Worth reading to understand the AI that predates agents, including symbolic methods.
Good for
Filling in classical AI and symbolic reasoning
Seeing how agent decisions relate to classical search
Academically inclined readers
Watch out
Largely unrelated to generative AI and LLMs. If your goal is shipping a working agent quickly, the return is low — read selectively rather than end to end.
Our take：A trilingual (Traditional Chinese / Simplified Chinese / English) agentic-AI roadmap with 240+ resources arranged by stage, each with required exercises and reading. Our three-track, four-stage structure borrows heavily from its layering.
Good for
Traditional Chinese readers
Anyone asking "what should I read next"
Readers who want exercises to test understanding
Watch out
It is an index rather than a tutorial and carries no full text. Resource quality varies, so judge each entry yourself.
Our take：Datawhale's Chinese-language agent roadmap and resource collection, including a list of major open-source agent projects (SWE-agent, OpenHands, UI-TARS). Useful for quickly deciding which open-source project to read.
Good for
Chinese-language readers
Finding open-source agent projects worth reading
Community study resources
Watch out
Mostly link aggregation; depth requires following through to each project. Update cadence depends on community maintenance.
Our take：A learning index that separates tutorials, courses, and reading lists, and tags which framework each resource uses (smolagents, LlamaIndex, LangGraph). Easier to pick from than a giant awesome-list.
Good for
You have chosen a framework and need matching material
You prefer book-list style deep reading
Comparing the framework slant of different courses
Watch out
An index with no body text. Its taxonomy reflects the author's habits and does not map exactly onto our three tracks and four stages.
Our take：The clearest index of the engineering side of agents: benchmarks such as AgentBench, GAIA, and MCPAgentBench, plus observability, deployment, and failure modes. Required reading for stages 3–4.
Good for
Building your own evaluation harness
Observability and failure modes
Engineers moving from demo to production
Watch out
A pure index with no tutorial text. Papers and tools move fast, so confirm links still resolve.
Our take：Tracks 300+ free agentic-AI resources ranked by stars and rising momentum, updated through 2025–2026. The lowest-effort way to see what people are actually using now.
Good for
Understanding the current ecosystem
Filtering projects by stars and growth
Periodically scanning for new tools
Watch out
Stars are not a proxy for quality or maintainability. The list is long — set your own filter or risk collecting without building.
Our take：Organised around a single formula — Agent = LLM + context + tools — running from fundamentals through context engineering, tools, and memory to multi-agent systems and engineering practice. The 28 runnable projects are its greatest asset.
Good for
Readers wanting a systematic Chinese-language book
Those who value runnable code over theory
Deepening understanding at stages 2–4
Watch out
Centred on specific frameworks and the author's own practice — separate universal principles from the author's choices. Confirm the open-source licence scope before quoting.
Applies to：Developer Track · Business TrackStages：S2 · S3 · S4