Where the ideas come from

Every essay on this site starts in the same place: me, sitting with a set of ideas and experiences I want to put together. Something I read that connected to something I built. A conversation with a client that echoed a conversation from ten years ago. A pattern I noticed across three different companies. The work of thinking — putting disparate things next to each other and seeing what the collision produces — is mine. It has to be. Nobody else has stood where I've stood.

That's the starting point. Only then do I bring AI into the loop.

What I use LLMs for

Once I have a position — a claim, a frame, a piece I want to write — I use large language models to sharpen it, never to originate it. I have the model attack the argument to find where it wobbles, pressure-test the structure, tighten the prose, and catch what I've stopped seeing — typos, broken links, a dropped thread. It also helps me triage what to read: sorting papers and transcripts so I spend my time on the sources worth it. I read the sources myself; I accept or reject every edit; the model publishes nothing on its own.

What I don't use them for

  • Originating positions. See above. The take is mine; the model doesn't hand it to me.
  • Originating facts. Every named person, quote, statistic, and historical claim comes from a source I've read or a fact I already knew. Never from a model's memory. If a claim can't be verified, it doesn't run.
  • Ghost-writing. No piece is a model draft dressed up. Every sentence has passed through my hands, usually several times.

The agentic layer

Some of what I run is more than a chat window — my tools can reach the web, my notes, and data from services I use. For essays this means faster research and cleaner drafts; it does not mean the essays are automated. The argument, structure, and voice come from me sitting with the piece. Where a particular one involved something unusual — heavier synthesis, an automated pipeline, generated prose that survived editing — the note at the end of that piece will say so.

The commitment

  • Positions and arguments are mine.
  • Facts are verified against primary sources before publication.
  • Sources are cited at the end of each piece so you can check the work yourself. If a claim isn't cited and isn't obvious, ask — I'll point you to it.
  • The voice is mine. If a sentence sounds like a model wrote it, that's a bug — please tell me.

A note on older essays

This policy describes current practice. Older essays on this site predate its formalisation; the principles above — positions and facts originate with me, LLMs sharpen work I have already begun — held then too, but not every past piece involved AI in any form. The "On AI" link at the top of every page ensures any reader can reach this policy from anywhere on the site, past or present.

Last updated: 11 August 2026