Why this guide exists
There is a strange gap opening up in the world. On one side are a few thousand people who build these systems; to them, the behaviour of an AI is not magic — it is mechanism. They know why it makes things up, why it's confident when it's wrong, why it's brilliant at some tasks and hopeless at others that look easier. On the other side is nearly everyone else — including people making large, consequential decisions about tools they can describe but not explain.
This guide is for the second group — for the ones who refuse to stay there. The serious technical literature is extraordinary and almost entirely closed to outsiders: the reference this series is built from runs to six hundred pages of equations, code, and hardware diagrams. What I've done is take the real ideas inside it — not a watered-down cartoon, the actual load-bearing concepts — and rebuild them in language you can hold. Semi-technical, never non-technical. I won't protect you from the ideas; I'll hand them to you with the maths taken off, so you can pick them up.
By the end, you won't be impressed by AI the way a person is impressed by a magic trick. You'll be impressed the way an engineer is impressed by a bridge — because you can see how it holds itself up, and exactly where it would fall down.
That second part matters more than the first. Understanding where these systems break is what turns you from a spectator into someone who can think clearly about what they should and shouldn't be trusted to do.
Every set ends with the honest questions you'd still be asking — answered plainly, so you leave with understanding, not a vague unease.
One floor at a time. Each set assumes the one before — we never meet an "agent" before you understand it's a next-word predictor with a tool-belt.
Every chapter carries custom diagrams. The hard ideas get a picture to land on, because a clear figure does what three paragraphs can't.
Strategic, not hype. Where the field is unsettled — faithfulness, alignment, evaluation — the guide says so plainly.
The journey runs in seven stages: how it works, how it's taught, how it reasons, how we measure it, and then the two-part frontier — how it becomes an agent that acts in the world. A final, more technical set covers the engineering that makes all of it affordable at planetary scale. Start at the bottom; the climb is the point.
What do we mean by "agentic AI"?
Almost every AI you've used so far answers. You ask a question; it replies; you're done. It's a brilliant responder, but it sits still — it can't do anything in the world. An agent is the next thing: an AI that can take actions to accomplish a goal — use tools, search the web, run code, fill in forms, call other software, work through many steps, and check its own progress — with only light steering from you.
The difference is the difference between advice and action. Ask a normal assistant "how do I book a flight to Tokyo?" and it tells you the steps. Ask an agent and it opens the site, compares the options, fills in your details, and books it. A chat model can write you some code; a coding agent edits the files in your project, runs them, reads the errors, fixes them, and tries again — on its own, in a loop, until the job is done. That loop — decide, act, observe what happened, decide again — is the whole idea. The model stops being a clever oracle and becomes a worker.
So why spend four sets on language models before we get to the agents? Because an agent is not a different kind of machine — it's everything in this guide, wired together and pointed at a goal. At its core sits a language model (Set 1), taught to be helpful (Set 2), able to reason through the steps a task needs (Set 3). Bolt that mind to tools, a memory, and a loop that lets it act, and you have an agent (Sets 5–6). You genuinely cannot understand — or trust, or govern — the agent until you understand the mind running inside it. That's why we build from the bottom.
Free — subscribe to unlock
Drop your email and the full library opens. You'll get each new set the day it's published, and nothing else — no noise, no spam, unsubscribe anytime.
Prefer one continuous read? Open the full book → (read online, or print / save as PDF)
Enter your email to start reading
Three full sets — fifteen illustrated chapters — are ready now, with four more on the way.
You're in. Welcome.
The library below is unlocked. Pick any available set and start climbing — Set 1 is the foundation everything else rests on.
All seven sets, end to end
The complete structure is laid out below so you can see where the journey goes. Available sets open to read; the rest are in production and arrive in order.
- The Machine That Only Predicts the Next Word
- First, It Breaks Your Words Apart — tokens
- How It Pays Attention — the Transformer
- How It Decides What to Say — generation & hallucination
- The Machine Underneath the Machine — silicon & the chip wars
- The Three Stages of Raising a Model
- Reading the Whole World — pretraining & scaling laws
- Teaching by Example — fine-tuning & cheap adapters
- Learning From What People Prefer — RLHF
- The Shortcut, and What It Costs — reward hacking, sycophancy, the alignment tax
- The Blurt — why instant answers fail
- Thinking Out Loud — chain-of-thought & test-time compute
- Discovering Reasoning — the "aha moment"
- Checking the Work — many paths, and grading the process
- How Much Thinking Is Worth It — economics & limits
- Why Measuring Is Hard
- Catching the Confident Fabrication — hallucination detection
- The Judge Is Also a Machine — LLM-as-judge & human annotation
- Benchmarks and Their Lies — contamination & Goodhart's law
- Grading an Agent — evaluating multi-step behaviour
- From Answering to Acting — the agent loop
- Looking Things Up — retrieval (RAG)
- Remembering — agentic memory & the Amnesia Tax
- The Harness — context management & orchestration
- Patterns That Work — agent design patterns
- Plugs and Sockets — the Model Context Protocol (MCP)
- Teaching an Agent New Tricks — skills
- Agents Talking to Agents — A2A
- The Org Chart of Machines — multi-agent systems
- Building and Showing — frameworks & agentic interfaces
- A Panel of Specialists — Mixture of Experts
- Shrinking Without Lobotomising — compression & quantization
- The Fast Intern — speculative decoding
- Keeping the Hands Fed — the memory wall, revisited
- The Data-Centre Reality — scale, parallelism & failure
- The line that actually matters — what gets automated, what stays human
- What becomes more valuable, not less
- How to actually get good at using them
- If the thing adopting this is your workplace
- Where this is going — open problems & the road ahead
- "Check Your Understanding" — self-tests for every part
- The Glossary — every term, in plain language