A 7-week path where you actually prompt the models, break things, fix them, and understand what's happening. Read the orientation in each week, run the exercises in your own terminal, ship the proof-of-learning at the end. No grade, no quiz — just whether your build runs.
Two weeks have full content (Week 1 + Week 2). Five more are outlined so you see the destination. New week unlocks every 3-4 days.
Strip the marketing. What a language model actually does, token by token. Why "AI is just autocomplete" is both true and useless. You'll prompt a model 5 times the same way and 5 times completely differently to feel the difference yourself.
Why where information sits in your prompt matters more than what's in it. You'll discover attention by breaking it — long prompts where the model "forgets" the middle, structured prompts that survive 10x the length. The lesson: structure is leverage.
Beyond prompting — when to load CLAUDE.md, when to use system messages, when to load files, when to use sub-agents. The skill that separates "prompt user" from "AI engineer."