Welcome
Everyone is talking about AI agents. Very few people can explain the machinery that makes one actually work — the layer between “a language model” and “a system you’d trust with real work.” This guide explains that layer, from the ground up.
By the end you’ll understand agents well enough to read the docs for any framework — LangGraph, Temporal, the OpenAI Agents SDK — and recognize exactly what each piece does and why. Not because you memorized vocabulary, but because you watched every piece get built.
Who this is for
Section titled “Who this is for”- Engineers new to agents who want the real mechanics, not hand-waving.
- Non-engineers — PMs, leads, founders — who need to genuinely understand what their teams are building.
If you write code, there are short code samples to make ideas concrete. If you don’t, you can skip every code block and lose nothing — the concepts and the interactive demos carry the whole story.
How to read it
Section titled “How to read it”Go in order. Each lesson assumes the one before it, and the payoff compounds — the failures you feel early are healed, one by one, later.
- Part 0 · Foundations — what a language model actually is, and why it needs a loop.
- Part 1 · The Naive Agent — build the simplest possible agent, then watch it fail six ways.
- Part 2 · The Harness — fix each failure, one capability at a time. This is the heart of it.
- Part 3 · The Map — connect what you built to the real tools, plus a glossary.
Most lessons follow the same rhythm: feel the problem → step through a demo → see the minimal code → name the real-world tool.
The demos are the point
Section titled “The demos are the point”Scattered through the guide are interactive demos — a loop you step through, a process you can crash and restart, an approval you grant. They run entirely in your browser:
- No setup, no accounts, no API keys. Just press the buttons.
- They’re deterministic simulations, not live AI. That’s deliberate — a scripted demo lets you freeze time and see the mechanism, which a real, noisy model would hide.
Poke at them. Crash things. That’s where the understanding sticks.
The one idea to hold onto
Section titled “The one idea to hold onto”Everything in this guide is an elaboration of a single sentence:
Agent systems are workflow systems. The model decides what to do next. The harness decides how it actually gets done — safely, durably, and without losing its mind halfway through.
Keep that split in mind and the rest follows.
Ready? Start with 0.1 — What is an LLM, really?