Until the world quiets down 🎛️

I press a pad on the BeatStep twice, and the chord I'm holding stays in the air. My hands leave the keys. A knob comes under my right thumb, the filter opens slowly, the pad breathes. I haven't added the lead yet. One small thing is changing in the sound, and my hands know what to do next without me having to tell them.

I've been chasing this feeling for twenty years.

The first and second posts in this series were about how Arcturus got built. This one is about why, and the answer goes further back than the code.

A lifelong fiddler

I'm not really a keyboard player. I don't read music and I don't have piano hands. If you sat me at a grand piano and asked me to perform anything I'd apologise and walk out of the room. What I am is a fiddler: put me alone with a knob, a slider, or a waveform, and I will disappear into it for hours.

I started making music as a teenager on a Windows computer running Cubase. ⊕ I remember the Cubase splash screen better than I remember a lot of my early relationships. Late-night CRT glow; a Sound Blaster card doing its best; the quiet hum of a PC tower that had no business being as loud as it was. I wasn't really writing songs. I was exploring. I'd land on a synth patch I liked, route it through an effect I didn't fully understand, change one number, and listen again. An hour would pass. Another hour.

The tools changed underneath me over the years. Cubase on Windows, then a stint on Linux, eventually Logic Pro on a Mac, and a handful of web and hardware experiments in between. The pattern underneath never changed. I was still looking for that specific, quiet feeling of being inside a sound.

The KeyStep I quietly loved

I bought the Arturia KeyStep years ago, long before I had a plan for it. I used it the way a fiddler uses a keyboard: not to perform, but to type in the occasional MIDI note while chasing a patch. ⊕ It's the same way I type at a keyboard, honestly. Not-quite touch-typing, eyes halfway on the keys, making up for the lack of virtuosity in patience. Something half-formed was happening in my head, I wanted to hear it, and poking the KeyStep was the fastest way to get it out of my head and into the air.

It never fought me. It did one thing and did it cleanly. Even now, when Arcturus boots and I start playing, the KeyStep feels the same as it did on day one.

And then the BeatStep happened

The BeatStep was different. The BeatStep was love at first sight. Sixteen pads in a square on the left, sixteen encoders in a square on the right, a master knob in the middle. ⊕ I bought it mostly for how it looks. I can admit that now. It is a very handsome rectangle. The promise was so obvious. Knobs to turn. Pads to press. Surely, surely, a device with this exact shape would just work as a control surface for the software I was already using.

Reader, it did not just work.

The software was always in the way

Every piece of software I tried to glue the BeatStep to turned out to be a small war. Logic Pro has a MIDI mapping interface that, I think, was designed in a room I was not invited to. I spent weekends on forums. I kept maps. I wrote scripts. I named presets things like please-god-just-work. ⊕ I'm sure there are people who have the Logic MIDI mapper completely figured out. I would like to meet them. I would like to shake their hand. I would like to know what I missed. Every so often I'd get something working, feel briefly powerful, and then an update or a new project would reset the state and I'd be back on the forums.

My original fantasy, back when I bought the BeatStep, was that Arturia would meet me in the middle. They make lovely soft synths. They make these handsome controllers. Surely the combination of their controller driving their synths, with the KeyStep on my left for notes and the BeatStep on my right for everything else, was the thing their software was designed to make easy. It was not, at least not in a way I ever managed to reach.

For a while I had a pleasant little consolation. I got the BeatStep working as a control surface for OBS when I was toying with streaming. Scene switches, camera choices, channel volumes. ⊕ For a brief golden period I had the BeatStep as my Stream Deck. Then I stopped streaming, and the BeatStep stopped being useful again. This is how a lot of my hobbies end. The pads finally felt like buttons; the knobs finally felt like knobs. But it was a novelty. It wasn't why I'd bought the box.

The quiet years

And then I got busy with life. The boxes went into a drawer. The music mostly stopped. ⊕ The drawer was half-metaphor, half-actual-drawer. You know the feeling: a hobby you meant to keep up with, quietly waiting, out of sight.

Something shifted

A few years later, a new thing happened to me, alongside a lot of other people: I became someone whose job and daily practice is working with large language models and agents. I was building prompts and tooling and frameworks. I was reading diffs written by models. Most of my working hours were spent orchestrating AI assistants into something productive. ⊕ I've written at length about how I got good at this in the first post and the second. This one is about what that practice unlocked in me that wasn't about work. I started building small, strange side projects for fun, because the cost of starting had dropped so much that "for fun" became a viable mode again.

Somewhere in that year-or-so of constant practice, the boxes in the drawer started quietly knocking. I don't remember the exact moment. It wasn't a lightning bolt. It was more like a tide slowly coming in. I had a vision I'd been carrying for nearly a decade about what those two boxes should be able to do together. I now had the tools, and the daily fluency, to actually build the software that would make them do it.

The vision lived inside the hardware all along

What the BeatStep was always supposed to be, for me, was very simple: the pads are instruments (press one to switch between them), and the encoders are the sound of the current instrument (sixteen knobs to shape it). The KeyStep is what I play. That's it. No menus. No screen. No MIDI mapper.

That's Arcturus. That's literally the whole thing. Sixteen pads in two rows on the BeatStep: the top row chooses which module's parameters the encoders are showing; the bottom row chooses which instrument is making sound. Double-tap a program pad and the chord I'm holding on the KeyStep latches, so I can walk around sonically on top of the pad I just laid down. ⊕ Under the hood this is a state machine with a voice pool. Under my hands it feels like telling the synth "hold these, please." Which is how I actually think of it when I'm playing. Press a different program pad and I'm on a different instrument, with the same sixteen encoders now showing its parameters. The browser does the bookkeeping. The playing lives entirely on the desk.

Building Arcturus was secretly two projects at once. The surface project was the synth: a thing with a DSP graph and a voice pool and a calibration flow. The deeper project was practising at a new level of agent-assisted work: holding a clear doctrine in my head, delegating aggressively, intervening only when taste required. ⊕ If you've read the other two posts, you already know the machinery. If not, the short version is: markdown files as operating system, a quality-score ratchet, and an offline signal-test harness that made the unsupervised stretches safe. Both projects fed each other. The practice was real and the music was real, and at the end of a month I had a thing I could plug in and play.

Minimal screen, maximal hands

The thing that separates Arcturus from every synth I've tried to tame before is not the sound engine. Lots of software has a good sound engine. It's the shape: I never have to look at the screen to change anything. ⊕ There is a screen. A big waveform, encoder labels in a grid, a little meter that pulses. But I don't need it to play, and that difference is the whole project. My eyes are on whatever they like. My hands are on knobs. If I want the filter slightly more closed, I close it a little. If I want the lead to breathe differently, I touch an envelope knob and it breathes differently. There is no menu between me and the sound.

What I use it for, these days, is pretty humble. I lay down a pad. A low, slow, fat thing that sits under everything else. I double-tap the pad that made the pad so the chord latches. Then I switch to another program pad and make a lead on top of it. And then I just sit there, turning knobs and listening and tinkering.

That's it. That's the whole thing. Nothing is being produced. No track is being recorded. No DAW is open. It's me and two boxes and a browser tab, and nothing is demanding anything of me.

Until the world quiets down

There's a particular kind of tired I've been carrying for most of my adult life. It's the tired of having too many tabs open, metaphorically and literally. Code, chat, Slack, email, decisions, opinions, notifications. ⊕ The browser tab count is usually a pretty honest lagging indicator of how well I'm actually coping with a week. Arcturus runs in exactly one tab. That counts for something. When I plug in the KeyStep and the BeatStep and open this one tab, all of that collapses into something much smaller: one knob, one ear, the next small change.

The first post in this series argued that agents made shipping this thing possible. The second argued that a test harness made agent-trust real. Both are true. But the reason I did the work at all isn't on either of those pages. The reason is this: I wanted a sound machine that asked nothing of me except which knob to turn next.

Twenty years of fiddling. Ten years of a beautiful box in a drawer. A year or so of getting fluent with agents. And now, at the end of a long day, I press a pad on the BeatStep twice, the chord stays in the air, and the world quiets down for a while.

Thanks for reading all three. Go turn some knobs. 🎛️