Colophon

Every essay here ends with a short dated note: what was happening in the world, in technology, and in AI when it was written, and where I was standing at the time.

That’s deliberate. Writing about AI dates badly, because you usually can’t tell what the author knew at the time. A claim that looked bold in early 2026 can read as either obvious or naive two years later, and without the dateline you can’t tell which. Pinning each piece to its moment lets you judge it fairly. It also lets me come back later and mark my own homework: did the pattern get stronger or weaker as the models improved? That test, run honestly, is worth more than the original argument was.

And it’s the whole point. I write about agentic AI and security, and about what it takes to ship real systems inside organisations that can’t afford to get it wrong: architecture, governance, regulation, the occasional detour through cryptography. Most of it comes back to context, one way or another, and a site about context that threw its own away would be embarrassing.

There’s a page called Positions that does the opposite job. The essays stay pinned to their dates. That page holds what I currently think, and gets rewritten whenever something changes my mind. Keeping both means I can revise a view without pretending I always held it. A dated essay is a record of what I thought then; Positions is a claim about now, and it carries its own revision date so you can see how fresh the claim is.

The site itself is built with Hugo, statically generated, and hosted on GitHub Pages. Self-hosted fonts, self-hosted images, and one third-party service: Plausible, for analytics. It sets no cookies, collects nothing personal, and builds no profile of you, which is why there’s no consent banner in your way. It tells me which pages get read and where readers came from. That’s all I want to know, and all it can tell me. The type is IBM Plex.

This page used to say no third-party calls of any kind, and that was true when it said it. I gave it up deliberately, because publishing into a void teaches you nothing about whether the work reaches anyone. Naming the trade seemed better than quietly dropping the claim. It’s the same discipline I’d want from anything I put in front of real users.