Choose your path¶
Start here · 10 min · no code
After this: AI 101 Two routes through the site. Pick either; you can switch at any point.
There are two ways through this site, and they are both complete. Pick by what you want to have at the end: an accurate picture of how these systems work, or an agent running on your own machine.
You can switch between them at any point. Every page says which layer it belongs to and links to the same idea one level up or down.
The reading path — understand AI, no code¶
Ten pages, four to five hours, nothing to install. For analysts, product managers, leaders, students, and for developers who want the map before the territory.
Each step needs only the ones before it. Prompting needs models. Retrieval is "give the model your data". An agent is "let the model act", which needs both. All ten pages live under Understand, and you can read them there in any order once you have the first two.
| # | Read | Time |
|---|---|---|
| 1 | AI 101 — what AI, ML and an LLM actually are, and the three things a model cannot do | 40 min |
| 2 | How models work — tokens, context windows, inference | 30 min |
| 3 | Prompting — system prompts, few-shot, chain of thought | 30 min |
| 4 | Retrieval and data — RAG, embeddings, vector search, GraphRAG | 30 min |
| 5 | What an agent is — components, and when not to use one | 40 min |
| 6 | Agentic AI — protocols, memory, orchestration, human-in-the-loop | 40 min |
| 7 | Enterprise AI patterns — copilot, autonomous agent, agentic RAG | 30 min |
| 8 | Safety and responsible AI — hallucination, injection, guardrails | 30 min |
| 9 | Fine-tuning and training — skim; the answer is usually RAG | 15 min |
| 10 | Infrastructure and operations — skim; MLOps, drift, cost | 15 min |
Safety comes after agents on purpose. The failure that matters most is an instruction arriving through a tool result, which only makes sense once you know what a tool result is.
The build path — build an agent, thirteen labs¶
Roughly 12 to 15 hours. Everything runs against a model on your own machine: no account, no API key, no cost. It assumes you can read Python and use a terminal, and it assumes nothing about machine learning.
The build path has the full module list. In short: set up a local model, then tool calling, the agent loop, the harness, context engineering, retrieval, evaluation, observability, safety, production.
If you have not read pages 1, 2 and 5 of the reading path, do those first. That is about 100 minutes and the modules assume them.
If you only have an hour¶
AI 101, then What an agent is, then Safety and responsible AI.
That is enough to follow any conversation on this subject and to tell a real claim from a marketed one.
How pages are marked¶
Every page opens with a block like this one:
Understand · 30 min · no code
Before this: How models work · After this: What an agent is Hands-on version: Module 5, Retrieval · In depth: Retrieval in depth
It tells you the layer, how long the page takes, whether you need to write code, and where the same idea lives at other depths. The four layers:
| Layer | What it is | Code |
|---|---|---|
| Understand | The concept, in plain language | None |
| Build | The same idea as something you run and break | Every page |
| Go deeper | Design-level depth for builders and architects | Some |
| Reference | Glossary, vetted reading, official sources | None |
You can also browse by tag to see everything on one topic across all four layers.
Not sure¶
Start with AI 101. It takes half an hour and it is the right first page whichever route you end up taking.
If you have been here before, What's new lists what changed and when.