My journey

Physics did not give me the way I think, but it prepared me well for it. Studying it left me with the conviction that the real power is not in the equations but in the abstractions underneath them, and in the models those abstractions assemble: every model provisional, improved or replaced rather than defended past its usefulness. That idea has lived rent-free in my head ever since. How I think is as much a product of my brain geometry, my career, and my own life as of that training, and of reading the likes of Hickey, Abelson and Sussman, Alvaro, Wadler, and Steele. What all of it left me with is a predisposition to test an abstraction rather than take one on faith.

I have been building software since the turn of the century and designing it for nearly as long, leading teams through data platforms, web systems, and technical leadership roles. The abstractions only earn their keep when something real is standing on them, and most of a career has gone into making sure they do.

Early in my career someone quoted a line to me, author forgotten: "code was the thing that got in the way of making software". It struck a chord, and has sat at the back of my head like a silent mantra ever since. The work that held up best kept moving toward that sentiment: better abstractions, and more of the power and leverage sitting in models rather than in the code that happens to express them.

The ADHD diagnosis arrived in mid-life, with an ASD assessment pending, after years of explaining my own working methods without the words for them. It also accounts for the shape of my curiosity: cooking, including baking and fermentation; writing, including two co-written comedy pilot scripts that were received well enough without being picked up; and years in and around the scientific scepticism community, including a spell hosting my own Skeptics in the Pub. The useful abstractions tend to arrive when two of those turn out to describe the same structure as something I am building, a connection nobody was looking for. When one of those abstractions is genuinely new, I name it, codify it in open specifications and libraries, and publish the argument in long-form when it needs room.

Since June 2024 I have led architecture and delivery of a confidential client platform, from concept to launch-ready. It launches in late 2026 and the contract runs on beyond that, with me keeping technical leadership as the day-to-day work winds down; the specifics stay under NDA. What is public is the thinking it produced: a distributed graph store specification (DCSGS), semantic modelling, mouldable interfaces, and planning-based interaction chains that surface prerequisite actions instead of blocking the user.

The open source side of that work runs through RelationalFabric. Canon is the usable library for lazy typing, still evolving; Howard and Suss carry the claims and propagator network ideas at an earlier stage. Agent Brain Trust and RPL sit on a parallel track: inspectable reasoning and AI rooms that refuse to skip the useful steps.

Lately the industry has tried to jump straight to the end with AI-assisted coding, and the cracks are showing. Between what that career taught me and the promise of vibe-coding there are steps we skipped, and that is the ground I am claiming. If you want the argument rather than the summary, start with Writing.