Orchestration

On my first day on a new contract, a colleague asked the standard new-hire question: "What do you do outside of work?" I make music, I said, and watched him land on the usual guess for someone who looks like me: something loud and aggressive. Not wrong, exactly — I came up through metal, and still reach for it often. But I'm just as interested in synthesisers and field recordings, and loud and aggressive covers far more ground than people think: the sound of a falling tree, or an animal's roar, can be as heavy as any wall of guitars, if you listen to it and not the thing that made it. Mostly, I love sound itself. He thought about it and asked the only question anyone asks a musician in 2026: "Do you use AI for music?"

"Yes," I said. "But not to generate sounds or compositions."

That week I'd used it to inventory my studio, to work out what gear to pack for a show by size and weight, and to untangle a signal path across a patch bay, a matrix mixer and a chain of pedals. Logistical things. None of it touched a note.

It took me embarrassingly long to notice that this is exactly the answer I'd give about my engineering.

I've been building for the web since around 2000. The first sites were static pages in FrontPage, with window.alert() popups nagging visitors to sign the guestbook. Since then: jQuery and its callback hell; the polyfills we wrote to paper over browsers that agreed on nothing; the early days of AngularJS and Backbone; and eventually TypeScript, Vue, components, micro-frontends. Twenty-five years of the ground moving under my feet. So I think it's fair to say I've picked up a thing or two — mostly that the tools never stop changing, and that my first instinct with a new one — suspicion — is worth nothing until I've used it.

Last year the next one was agentic AI, and I was suspicious — not of the idea, but of what people were doing with it. The buzzword was "vibe coding": prompt something, ship whatever falls out, don't look too closely. That made machine-generated code sound like exactly what I'd spent years learning to distrust. What I hadn't reckoned with was the other half — the system I could build around it, so every line held to my standard whether I wrote it or AI did. That turned out to be the interesting problem: not whether the machine could write code, but how far I could stretch the discipline that kept it honest. Building that was the real work — the prompting was the easy part. The tooling only ever ran as fast as the rigour around it would let it.

I use it for the parts that were never the point — the boilerplate, the test, the obscure config error, the first pass I'd have taken anyway — and it doesn't get to decide anything. The architecture, the taste, the bar for good enough: those stay mine, behind quality gates I can't talk my way past. It's not bulletproof, either: for a while it kept slipping unwanted comments back into code I wanted to keep clean, and no amount of asking stopped it — only a lint rule that refused them outright ever could.

I'd been applying the same discipline on a client's greenfield product under real deadlines; I carried it into a project of my own, where the stakes were only mine. The project is this website.

I built it because I left social media, and leaving taught me something I hadn't expected: when you delete an account, the work seems to vanish — but it doesn't. It's still there. What vanishes is its visibility. The platform never owned the music; it owned whether anyone could find it. And in an economy built on attention, invisible is close enough to nonexistent that the difference no longer matters. Which left me with an uncomfortable question: if the work isn't on any of the major platforms, does it exist at all?

So I answered it the way I know how — as an engineer. I found a problem and built my way out of it. A website is a throwback; social media was supposed to make it obsolete. But the old, unfashionable form turns out to be the durable one: a place I own outright, that nobody can quietly down-rank or switch off. Which is its own small joke — that it took the newest tool I've got to rebuild the oldest thing I know how to make.

What's here is two decades of work — records, scores, performances, the whole discography — kept in a form I control, next to the engineering that holds it up: a streaming audio player I wrote from scratch, precise enough that a piece's chapters land on the exact millisecond; two languages with proper prerendering; a small serverless backend for the contact form; accessibility checked on every build; and a pipeline that refuses to ship if a single pixel regresses or the quality bar drops. The same system also produces the paperwork both trades run on — a press kit and tech rider for the music, a CV for the engineering — each built from source, never laid out by hand. It runs on free tiers and a domain I renew once a year. It costs me almost nothing, and I own every line.

Nobody asked me to build it. No deadline, no client, no manager. It's simply the standard I hold myself to when no one is making me — which is, I think, the only standard that tells you anything real about a person. It's also the same instinct that runs through the music. Orchestration, in a score, is deciding how the parts combine — what plays, when, against what. In software, it's the literal word for the same thing: coordinating the pieces into something that works as a whole. I've spent my life doing one or the other, and lately both at once. The tools keep getting better at playing. The conducting is still the part I keep.

I'm between things at the moment, which is how I have the time to write this at all. I won't end it with a pitch — that would undo the point. I'll only say that the work is out in the open, where it's always been easiest for me to be seen: through what I've made, rather than in the room. If that tells you who I am, we should probably talk. If it doesn't, no website was ever going to.