The Selves of Us is a South African literary platform — essays, poetry, a podcast. It was founded by two people who care about writing more than almost anyone I have worked with. Tshego Khatri is a Botswana-born poet. Roy Michael Hewett is a retired principal and editor. Between them they had a rare voice and no infrastructure.

They also had a clear position, held without apology: writing should be done by humans. In a web filling up with generated text, they wanted to stand for the opposite. That stance shaped everything we built — including, especially, the artificial intelligence.

The voice came before the system

The engagement did not begin as engineering. It began with brand-voice extraction — sitting with their essays and poems until we could describe, precisely, what made the writing theirs. Cadence. Restraint. The way a sentence turns.

That work mattered because everything downstream depended on it. You cannot build a home for a voice you have not learned to hear. Only once we understood the voice did the conversation become an engineering one: what would it take to give this brand a platform equal to its writing?

A custom literary publishing platform

We built the whole thing end to end — an 87-page custom platform, 45 API routes, on a 25-table Postgres schema, with row-level security on every table. Not a theme. Not a page builder. A publishing system shaped around how two editors actually work.

At its centre is a three-role editorial CMS — founder, editor, studio — running a real workflow: draft, review, version-tracked edits, publish. Every change to a piece is kept. A poem that goes through six passes carries its history with it. The rich-text editor was built for people who care where a line breaks.

The Listening Room, the archive, the newsletter

Around the CMS we built the rest of the brand's life. The Listening Room — a podcast platform with a custom audio player, made for the work rather than borrowed from a generic embed. Full-text search across the entire archive, so nothing written disappears once it scrolls past.

Then the newsletter engine: draft, approve, send, with open-tracking built in. That last part is where the numbers speak. The newsletter is opened by 68 percent of the people who receive it — several times the industry norm. That figure is not a marketing outcome. It is what happens when a real voice reaches people who chose to hear it.

For revenue, we wired in PayFast — book sales and recurring supporter subscriptions, so readers who want to sustain the work can.

AI that reflects, never writes

This is the part I am most careful about. Their whole stance is anti-generated-content. So the AI had one rule: assist, never author. It is not allowed to write a sentence that goes out under their name.

We built six Claude features inside that rule. The Librarian answers readers' questions from the live story archive — a way into the work, not a substitute for it. The Mirror reflects a writer's own work back to them, showing them their patterns without rewriting a word. Editor's Eye supports Roy's review without replacing his judgement.

Every one of them points inward — at the writer's own thinking — rather than producing text to paste in. AI as a lamp held up to the work, never a hand on the pen. On a platform built to defend human writing, that constraint was the point, not a limitation.

Running the brand from one system

We also built a white-label social module: a token-authenticated client approval portal and scheduled publishing. It lets the studio run the brand's social presence from the same system the writing lives in — drafts go up, the founders approve, posts ship on schedule. One system, one source of truth, no scattered tools.

What it added up to

A literary brand that started with a voice and no infrastructure now has a platform equal to its writing: a publishing system, a podcast, a searchable archive, a shop, and an AI that refuses to write for them. The proof is quiet and it is real. The newsletter is opened by 68 percent of its readers.

That is what we do. We arrive already knowing, we build the system underneath the voice, and we hold the line the brand cares about — even when the easy thing would be to let the machine write.