The AI reading list, in order

This is the material worth reading and watching to get good at building AI, in the order to go through it. Nothing here is filler. If a book or a course didn’t earn its place, it’s not on the list.

You don’t need all of it to start. You need the first few things and a real project to point them at. Reading without building something is how people spend six months feeling busy and learning nothing that sticks.

If you want the version with something to ship at the end of each step, that’s the AI Track. This is the same path with the deadlines stripped out. Just the material.

The first five are the machine learning floor. Get through them fast. The rest is the actual work, and it’s where you should slow down.

The floor

1. 3Blue1Brown’s Neural Networks series. The clearest picture of what a model is actually doing. Watch it early, then watch it again once the rest has sunk in.
2. StatQuest, Josh Starmer’s channel. Put it on when a concept won’t click. Regression, precision and recall, ROC, trees, boosting. He explains all of it slowly and it lands.
3. Google’s Machine Learning Crash Course. Free and structured. Good for the core loop and the data and classification units.
4. Hands-On Machine Learning by Aurélien Géron. Your reference shelf. Read chapters one through seven for the concepts and chapter two for a real end-to-end project. Don’t grind every line of code. Get the shape of it.
5. Jay Alammar’s Illustrated Word2Vec. Where embeddings finally make sense. This one idea carries half of everything in the next section.

The actual work

6. Jay Alammar’s Illustrated Transformer. How an LLM works without the matrix math.
7. Andrej Karpathy’s Intro to LLMs, the one-hour talk, plus his Let’s build GPT and Zero to Hero videos. The clearest free explanation there is. Worth every minute.
8. The Anthropic docs. Prompt engineering, tool use, Building Effective Agents, contextual retrieval. Read the primary source, not someone’s thread about the primary source.
9. The OpenAI cookbook and docs. Function calling and the working patterns. Same idea. Go to the source.
10. DeepLearning.AI’s short courses. Prompting, RAG, vector databases, agents, evals. Free, an hour or two each, hands-on. The fastest way to turn reading into doing.
11. The Hugging Face NLP course. For when you want to go deeper on embeddings and retrieval.
12. AI Engineering by Chip Huyen. If you read one thing on this list, read this. It’s the closest thing to a map of the actual job.
13. Eugene Yan’s essays on evals and LLM-as-judge. The unglamorous skill most people skip. So don’t.
14. Hamel Husain’s Your AI Product Needs Evals. Read it. He’s right.

Keep going

The field changes every month, so none of this stays finished. Budget a few hours a week for primary sources, the Anthropic and OpenAI release notes and a handful of people who actually build, and ship something small every month so the reading has somewhere to land.

That last part is the whole trick. Read to build, not the other way around.

If you want the structured version, with something to show at the end of each step, start here.