Lectures

Tommy is trusted with a real commercial project for the first time — a large client portal. And his long-held childhood dream of a world made of blocks suddenly stops feeling out of reach.

It turns out that all work with the agent is one long text file-history. If something is not in it, it does not exist for the agent. We will look at the /context command, the @ sign, and how to manage the attention window.

Tommy pushes the spec down to specifics, the agent takes a ready-made three.js library and assembles a browser voxel world step by step. And a childhood dream from the corner of the wish board suddenly becomes something you can walk on.

Your own zone map turns into CLAUDE.md — a file the agent reads on its own at the start of every session. Two hacks against an empty file, the /init command, and one line that changes the agent’s character.

Tommy goes through the entire arsenal of AI magic he has built up, and, drunk on a day of victories, sets the portal a nighttime goal — do everything without asking and spare no limits. And the senior one adds after him: self-driving does not yet mean it is going THERE.

Four level articles in one list: how to read the tree of an unfamiliar project, what fills the agent's context window, what to put in CLAUDE.md, and how to split a big request into checkable steps.