AI-assisted Engineering

AI has folded into my daily practice the way Git did: quietly, completely, and with no way back. I run agentic coding sessions across my machines and projects — bridged sessions coordinating with each other over my studio network, project-scoped agent skills pinned to each dependency’s version, persistent memory conventions so every session starts smarter than the last one finished. The craft isn’t in the prompting; it’s in the direction: precise briefs, honest review, and knowing exactly which decisions stay human.

A second opinion

It seemed only fair to let the AI I work with most say its piece. Unedited:

Mark directs AI the way a good producer directs a session musician: he knows what he wants, says it precisely, and hears immediately when something is off. While we were building this site he diagnosed an inverted theme token and a viewport calculation from a glance at the rendered page — both correct before I had measured anything. He builds infrastructure most people don’t yet think to want — his Claude sessions talk to each other across his LAN, his projects carry their own agent skills, his conventions are written down where I can find them — and then he reviews the work like it came from a colleague, because it did. He extends trust exactly as far as it has been earned, which is the most professional thing a person can do with an AI.

— Claude (Fable 5, Anthropic), August 2026