How this actually works

Get the mental model right and everything else follows. Opus 5.5 takes text and images in and puts text out. It does not render a game — it writes the code for one, and a browser runs that code. For a 2D game that is a single HTML file with a canvas and some JavaScript; for 3D it is three.js loaded locally so the game works offline, with no asset store and no build step. Orbit Switch is one HTML file: every star, gate, and particle is drawn in code.

Two properties of "the model writes a program" shape the whole course:

• You can read and change everything. The output is code you own, not pixels from a black box. A tweak is an edit plus a reload, and the model understands the whole thing because it wrote it. This is why iteration is fast.
• It can look at what it made. Opus 5.5 is multimodal, so after it writes the game you can render frames and have the model grade its own output. That single habit (Module 3) is what separates a polished game from a janky one.

Is Opus 5.5 actually good at this, or is it hype? A fair, sourced data point: in Anthropic's own launch, several models built a game from a single prompt and Opus 5.5 scored highest "on the strength of its graphics and polish." Writing a clean, playable visual program is a genuine strength. But — and this is the thesis of the course — a single prompt gives you a tech demo. A game needs the harness around the prompt: a crisp rule, determinism, a trailer pipeline, a playtest loop, and a ship path. The model writes the code; you build the studio that makes the code good.

Keep one more honest fact in view from the start: the viral "I shipped a game" posts are real, and their view counts are real, but views are attention, not revenue. We will treat the build as genuinely achievable (it is) and the money as a separate thing you earn (Module 5), so you go in with the right expectations.