Archive
Earlier guides, still online, no longer part of what I offer.
These were written while I was working on two adjacent problems: building products with AI without a large team, and designing recommendation systems that can show their grounds. Both fed into the legibility work, and neither is something I sell.
They stay published because they are cited, because the methods still hold, and because removing work you did is a strange way to argue for a legible public record. If you are here for what I do now, that is the Legibility Sprint and ongoing legibility work.
The Multi-Model Workflow
Which model to use for which task, how to hand context between them, and how to keep one chat from turning a strong model into a mediocre generalist.
The Tech Team Moment
How to brief work and accept it without reading code: specify behaviour rather than implementation, and keep acceptance on the behaviour.
Readable AI Decisions
How a recommendation system can make its grounds checkable rather than asking to be trusted. Where the Indifference Test comes from.
Also here
The Canonical Buffer Audit and Visibility Without Budget are not archived. They are part of the current method and live alongside the Sprint.