How I Use AI

I use AI tools extensively. This page explains what that means in practice — both how they fit into how my mind works, and the standards I hold myself to before anything goes out under my name.

The cognitive prosthesis

My brain moves faster than my fingers. Ideas arrive in clusters, trailing connections to papers I’ve read and conversations I’ve half-remembered, and the window for capturing them before they dissolve is shorter than typing allows. I record audio — a lot of it — and AI transcription is what makes that tractable.

Beyond transcription, I use AI to help me keep track of my own thinking: connecting papers I’ve read, tracing the links between ideas across different domains, and surfacing threads I’d otherwise lose to the general noise. There’s an unexpected benefit here I didn’t anticipate: explaining my thinking clearly enough to get useful output is itself a clarifying discipline — more than once, the act of articulating a half-formed question has resolved it before any response arrived. My commonplace book is the artefact of that practice; AI is part of what keeps it alive.

I’ll also name the less glamorous part: I have procrastinating perfectionism and executive dysfunction. The gap between having a thought and producing something from it is, for me, a real obstacle — not laziness, but a kind of friction that is neurological before it is volitional. AI helps me stay on track when my focus would otherwise have me buried in a rabbit hole that doesn’t serve my original goal, and reduces the cost of starting enough that I actually start. I consider these uses a cognitive prosthesis: tools that compensate for the way my mind works, rather than substituting for it.¹

The competence boundary

Using AI as a prosthesis is different from using it as a ghostwriter. Everything that goes out under my name is something I can explain and defend without AI assistance. If I can’t, it doesn’t go out.

For empirical claims, that means primary sources. If something I’ve written cites research findings or domain advice, I’ve read the source — not a summary the model produced. AI can locate papers and extract structure; the reading and the judgement are mine.

The distinction I find useful: AI reduces the cost of execution for ideas I’ve already reasoned through. It does not supply the reasoning. An argument I can’t make without the model is not my argument, and I have no business putting it in front of readers.

There’s a self-interested version of this commitment too: extended reliance on AI for cognitive tasks can atrophy the underlying capability.² A prosthesis used as a substitute rather than a supplement will eventually weaken the muscle it was meant to support. Staying within the competence boundary is the discipline that keeps the skill alive — not only an obligation to readers, but a condition for remaining someone worth reading.

What this means for you

When you read something I’ve written, you can hold me to the above. If a claim is wrong, or you have a better source, I want to know — because I’m the one who chose to publish it, and that’s on me regardless of what tools I used to get there.

I’m also happy to talk about the tooling in more detail. Ask.