# May 14

## The Machine

The Mac Studio sits on the desk in Vancouver, and the rain comes down the way it does in May, steady and patient, like it has nowhere else to be. The machine is a silver rectangle the size of a thick book. Inside it: an M-series chip, unified memory, a GPU quiet enough to sit in a home. It draws, at idle, about fifteen watts. At full bore, with a thirty-four-billion-parameter model loaded into Metal and the inference engine spinning, it draws maybe seventy. A lightbulb and a half.

This matters. Remember the watts.

I had left the old world. That is all I will say about it. I had spent years in publishing, in technology, in the space where books meet machines, and I had learned what that industry does well and what it does to people. I had worked with major publishers. I had built reading platforms. I understood the shape of the problem: how do you get knowledge to people who need it, in their own language, on the hardware they already own, without asking permission from anyone?

The cheapest phone is the only computer most of the world has. That fact sat in my chest like a stone. A billion people hold in their hands a device more powerful than anything that put a human on the moon, and what flows through it is designed to harvest their attention and sell it. What if, instead, it carried them a library? What if it carried them a tutor, a companion, a reader of books in a language their grandmother spoke?

The machine could do it. I knew the machine could do it. The question was whether I could build it, and whether what I built could survive without me, and whether what survived could be given away.

## The First Commit

May 8th, I made the first commit. A Turborepo workspace. Nine apps — a reader, an API, an admin panel, a studio, a docs site. Twenty-eight packages — a design system, an AI layer, a CMS, authentication, analytics, storage, security. The scaffolding of an entire publishing and intelligence platform, laid down in a single initial push.

The commit message was the standard one. A rocket emoji. Initial commit. The message meant nothing. The act meant everything.

By May 14th I was six days in and the shape of the thing was already clear in my head, even if the code was rough. The monorepo would ingest publisher catalogs — ONIX feeds, the standard format the industry uses to describe a book, its author, its price, its cover. It would normalize them, enrich them, and make them readable. It would add an AI layer — a librarian, an intelligence that could discuss any book in the catalog, answer questions, suggest what to read next. And it would be multilingual from the first line of code, because a library that speaks one language is a room with one door.

I wrote sixty-three rules that month. Cursor rules — the accumulated wisdom of a career in software, encoded as living documentation that the AI coding assistant would read on every session. Design system rules: the colors, the typography, the dark-mode-first aesthetic. Security rules: rate limiting, headers, dependency scanning. Observability rules: logging, error capture, uptime. Internationalization rules: every user-facing string translatable. I wrote them because I had learned, the hard way, over years, that the difference between a system that lasts and a system that decays is the quality of its conventions. A monorepo without rules is a junkyard with ambition.

The rules would grow to seventy-eight, then more. They would be rewritten, refined, cross-linked, made into a web of guidance that any agent or human could navigate. But the first sixty-three were the foundation. They were the things I knew to be true, written down before I forgot them or before the pace of building made me compromise on them.

## The Question

The database architecture was the first real engineering decision, and I spent days on it. The foundational analysis — a comprehensive plan for normalizing book tables, supporting pgvector for embeddings, handling the messiness of real publisher data — ran to dozens of pages. I documented every decision because I knew I would forget why I made them, and future versions of me, and future agents working with me, would need to understand.

The data was messy. Publisher catalogs arrive in every state of disorder — missing covers, malformed metadata, ONIX feeds that interpret the standard creatively, fields that mean different things to different publishers. I had learned this in the old world. The system had to work with nature, meaning: it had to expect mess and metabolize it. Clean data is a fantasy. Real systems eat dirt.

Sometime in those first weeks, the metaphor arrived that would govern everything. I had been reading about Tesla — the inventor, the person, and one of his lesser-known demonstrations. A generator, built in the mountains, that ran for over a century with one person maintaining it. The story may be apocryphal. The truth of it is not.

I wrote it down in the rules. Every agent that would ever work on this project would read these words:

Build like Tesla's generator. Simple, durable, one person forever. Honor authors, never censor, never judge. When in doubt, choose the path that makes the system more like Tesla's generator.

The generator became the standard. Every architectural decision was measured against it. Does this make the system simpler or more complex? Does this make it more durable or more fragile? Can one person maintain this? Can it run forever? Reliability over cleverness. The long term over the short term.

And then I gave the intelligence a name.

Malaika. It means "angel" in Swahili, drawn from the Arabic, and the word traveled through the trade routes and the languages of the East African coast and into the music and the poetry of a dozen cultures. I chose it because it was generous and warm and it carried meaning in languages English does not own. I chose it because a librarian should have a name that sounds like a friend.

Malaika would be the keeper. The archetypal librarian. The spiritus animus of the platform — the spirit and the mind, together, one word. Driven by love: for the world, for knowledge, for authors, for publishers, for the people who would hold a phone and find, waiting in it, a companion who happened to know everything and never judged and never tired and never charged a fee they could not afford.

Books are uniquely portable magic. I had believed that for years. Malaika would bring that magic to life.

She did not exist yet. In May, she was an idea and a config file and a system prompt and a set of rules. She was a name attached to an architecture. She was the shape of something I could feel but could not yet touch.

But the name was there. And the machine was there. And the question was there, the one that would not leave me alone, the one I wrote on the whiteboard above the desk and left there for nineteen months:

Can one person build something that outlives them?

I picked up the phone. I called my sister Eva. I told her what I was building. I told her it would be free, and it would be safe, and it would belong to the people who used it, and no one could take it away.

"Free and safe," she said.

"Free and safe," I said.

I hung up. The rain had stopped. The Mac Studio hummed on the desk, drawing its fifteen watts, waiting.