How it was made

Tetramental and AI

Is it slop?

No. AI is used heavily in making Tetramental, and I won’t pretend otherwise. But “slop” is what you get when AI output is published as it comes out. Nothing here was. Every chapter, track and level has been read, heard, played, questioned and reworked by a person, many times over. AI didn’t remove the hard part of making things. It moved it. My favorite authors say the real difficulty of writing is in the editing, and that’s where my hours went.

Behind the curtain

Some real numbers, as of October 2026:

All told, about 1.3 million words of planning, review and notes sit behind the story and the game. AI drafted much of that writing alongside me; I directed it, read it and decided what stayed.

The book

I wrote the first draft of Tetramental quickly, using AI to get from idea to paragraphs. Within a few months it was finished: twelve chapters, about 74,000 words, start to end. Had I published it then, I’d have called it slop myself. Instead, I read it through again and again and logged every problem I found: 223 recorded revisions, many of them dozens of edits each, across seven full editorial passes. Scenes were added, and long chapters were split until each one ran a natural length. The book grew to twenty-two chapters and about 108,000 words.

Behind it, I wrote dozens of design documents for the world: its characters and how each of them speaks, the spellcraft and the elements, the academy and its grounds. Together they run to about 278,000 words, over two and a half times the length of the book itself. Only after my own passes did I bring AI back in, for editorial reviews from one perspective after another, checked against those documents. I leaned on AI where I fell short, and did the rest myself.

The music

I started making music with Suno in 2024, for other projects. In late 2025 I began to wonder how this world would sound, and turned it to Tetramental between writing sessions. A track might go through hundreds of versions: rewriting descriptions, making covers to shift a sound, then reworking it afterward by adding instruments and reshaping parts until it’s right. I’ve spent hundreds of hours listening to and tweaking these tracks.

The game

I’ve dreamed up games since I was a kid. I daydreamed about building worlds so much in school that my teacher worried about me. I’ve programmed games by hand before, but never shared them widely because I didn’t feel I’d taken them far enough. With Claude Code, I made Can’t Stop the Swap, a puzzle game blending my own code with generated code, refined over months and released as open source. That showed me what was possible, and it led to Tetramental’s game. The free game on this site is designed, tested and tuned by me, played over and over until each level, teacher and challenge feels right.

Not as new as it sounds

Computers have been helping make things for decades; we just didn’t always call it AI. Games have had computer opponents, synthesizers have made music by program, and writers have relied on spellcheck, grammar check and autocomplete. Films are full of 3D renders and digital effects. AI is the next step on that road, and a big one, but the direction still has to come from someone.

Human-led

I’ve spent thousands of hours on this world across the book, the music and the game. I’m a recovering perfectionist with a little ADHD, and AI has helped me finish things I’d otherwise have left half-done. But it has been human-led from first idea to final edit. If you enjoy it, I hope you’ll see the work that went in.

— Matthew Keefe, founder of H4rdw1r3d Studios