Twelve lessons in three groups. Each is one page, under 900 words, with one diagram, and ends by naming the single page to read next. The lessons are being written now — the list below is the finished shape, and a title without a link is a page that does not exist yet.
Who this is for
Someone who can already get an AI coding tool to produce code, but does not yet have a stable model of what it sees, why a long session gets worse instead of better, how files preserve what the conversation forgets, or how to tell a plausible answer from a proven one.
The outcome is not becoming a software engineer. It is making better choices, recognising the common failure modes while they are happening, and finishing small projects with a workflow you can repeat.
Reality
What the technology does today, before any of the mechanics.
The machine
How the thing you are directing actually works. The context window is the load-bearing lesson of the level; the three around it exist to make it usable.
- The context window
- Tokens and false economies
- Models and effort
- How it goes wrong
Your side of the work
The habits that decide whether any of the above helps you.
- Outcome and boundary
- Markdown files are the memory
- Folders a cold session can read
- One bounded session
- Plausible is not proven
Plus the glossary: every term used in a lesson, defined in one line, linking back to the lesson that introduces it.
What is not here yet
Six of the twelve lessons. Lessons 1 to 6 are written, and the glossary holds the terms they define; every other title above is a page that does not exist yet. The lessons are published one at a time, in the order that makes each one readable on its own, which is not always the order they are numbered in.