How AI is used here

New here? Start with five pieces →

The guardrails, as principles

The guardrails, as checkable rules

These are the ones a build can enforce. A rule worth stating is worth testing, so each of these is checked every time the site is built, and a page that breaks one does not reach it. The friction is deliberate: a false stop costs me a minute, a false pass is public.

AI disclosures

One was in the room for nearly all of it. Here is the part that matters, which is what it did not do.

AI-assisted drafts

A post with the 🤖 AI-assisted draft chip starts with me and is written up by Claude. The idea, the angle and the stories are mine (usually an outline, or my answers to an interview where Claude asks one question at a time). Claude turns that into a first draft in my voice.

Then it comes back to me. I read the draft, edit it, and change anything that does not sound like me or that I do not think is true. Claude checks the citations and links along the way. Nothing goes up until I have re-read the whole thing and approved it.

These drafts are how I get an idea out quickly, while it is still fresh. Then I distill it and polish it, pass by pass, until it reads as one coherent post. The goal is for most of it to end up in my own words. When it gets there, the chip comes off; until then, it stays. Whether a post has got there is my call, not a word count.

The bubbles in the margin

They started as a joke about Pop-Up Video, the VH1 show that ran little bubbles of trivia over music videos, and turned into the most honest part of the site: a place for the machine to say the thing I left out, or argue with me where you can see it. There is one robot. The colour tells you what kind of remark it is making.

On a wide screen they sit out in the margin beside the paragraph they belong to. On a narrow one they stack underneath it. They are part of the page rather than an overlay, and they need no JavaScript to read.