Most of what appears here begins away from the screen: a dictated voice memo, a page of notes, a handful of references and the argument I already mean to make. An AI assistant, Anthropic’s Claude, is then used to transcribe, to tidy the prose and to hold the house style steady across the essays. The thinking is mine, and so are the claims, the judgements and the errors. I treat whatever the machine returns as a conjecture and nothing more: it is checked against the sources, argued with, and often thrown out. Nothing stands here on the model’s authority, because nothing stands on authority at all. And where an essay corrects something I have argued before, the correction is mine too. The podcast that accompanies a post is a synthetic reading of the finished essay, generated with ElevenLabs: the same text, spoken, and not a recording of my own voice.