The right kind of AI video is the boring kind

I wanted a short explainer video for the operator-workspace piece. The obvious route — point an AI video generator at a script and let it produce a talking avatar over stock motion — was also the wrong one. This site's whole register is calm, diagram-led, low-hype. A glossy generated video would have undercut exactly the thing the site is trying to demonstrate. On a page about judgement, a flashy video is an own goal.

So the interesting decision wasn't "make a video." It was which kind. There are two AI video paths and they look nothing alike. One generates pixels — an avatar, synthetic motion, the house style of every AI demo. The other generates nothing: you write the video as a web page, in the same fonts and colours as the site, and a renderer films it frame by frame into an MP4. Same input, same output, every time.

I took the second path. The video is HTML and CSS — the site's own charcoal, white, and muted teal, the same Inter and JetBrains Mono, the same boxes and arrows the diagrams already use. Nothing in it is "generated" in the sense people mean when they're nervous about AI video. It looks like the site because it is the site, set in motion.

The lesson generalises past video. When AI offers you a powerful, impressive-looking default, the more senior move is usually the narrower, duller one you can fully control — the one where you still own every pixel, and the output can't drift somewhere off-brand without your say-so. Restraint is the feature.