Collaborate, Don't Automate

My route into music never came through a formal education. I came into it sideways, from the technical end, and I have always been more of a builder than a player. Somewhere in the middle of a media technology degree I fell down the rabbit hole of digital audio workstations (DAWs), through which I was able to conceptualize what all of these complex-sounding devices I was learning about in my audio technology specialization were practically used for. I also began experimenting with music production and sound engineering myself, though I never quite made it far enough to be comfortable calling myself an artist. Not long after, I seemingly regressed into the ancient yet elitist realm of analog audio equipment, as I found myself spending countless evenings at a workbench, soldering oscillators and filters together from bare components, chasing the small and specific thrill of a circuit that finally makes a sound. For those around me it was an unpleasant buzzing, for me it was exactly 262 Hz, a C4. The first time one of my own builds sang back at me, something greater was set in motion.

I mention this because it turns out to be the whole story in miniature. I was never interested in music technology as a way to replace the person making the music. I was interested in it as a way to build that person a better instrument.

Years later I found myself doing a master’s in artificial intelligence, and the DAW and workbench got a lot more abstract. The oscillators became neural networks, and the thing I was trying to make sing was a model. The field I walked into was, to put it gently, on fire. Generative AI for music had gone from a research curiosity to something you could type a sentence into and get a finished song back a moment later. Some of those systems are remarkable feats of engineering, and I refuse to diminish their brilliance, as for every person who has something to criticize there is someone else who gets immense enjoyment out of them.

But the more time I spent with the state of the art, the more I noticed a lopsidedness in where the attention was going. Almost all of the energy, the headlines, the funding, the demos that go viral, was pointed at one particular shape of tool. You give it a prompt, it gives you a result. One shot. You describe, it decides. The human writes a sentence and the machine does the rest.

And I kept coming back to the same thought. That is not how I have ever experienced making anything.

When you are at a workbench, or a mixing desk, or in a room with other musicians, creation is not a request you submit. It is a back and forth. You make a move, the material pushes back, you respond to what it did, and somewhere in that loop the thing you are making turns out better and stranger than anything you could have specified in advance. The interesting part is precisely the part a single prompt skips over.

There is a different category of tool, one that sits inside that loop with you rather than standing in for it. Tools you play with instead of prompt. Systems that respond in real time to what you are doing, and that complement your taste and your skills rather than replacing them with an averaged guess. That category exists, and good people are working on it, but it is growing slowly and quietly in the shadow of the flashier one-shot systems. It does not demo as cleanly, and it requires existing skills, thereby giving it a higher barrier to entry, something the one-shot class of models sidestep entirely. So it gets a fraction of the attention.

That gap is what I have quietly oriented my work around. For a long time I thought it was mostly a technical observation. Then I started meeting people at networking events, and hearing firsthand what they make of it.

What struck me hardest was how differently people felt about all of it. The people I met were rarely computer scientists, more often they came from fields I am not familiar with at all, so every conversation about AI arrived from an angle I had not yet been exposed to. AI is about as polarizing a subject as exists right now, and most conversations about it online read like trench warfare. In person it was not that. People held very different convictions about whether these tools help or harm the creative act, they said so plainly, they listened, and more than once we simply agreed to disagree without anyone feeling smaller for it. The way people feel about AI-made art is about as varied and subjective as art itself. There is no single correct feeling waiting at the end of the argument.

Underneath the disagreement, though, something did seem shared. Almost nobody believed these tools were going away. And almost nobody believed the creative process, the actual human experience of making something, was the machine’s to take. What people were wrestling with was the space between those two things, and how to keep the second one true while the first becomes permanent.

None of this is new, either. Every real shift in music technology, the electric guitar, the synthesizer, the software that turned a laptop into a studio, first arrived as noise that would supposedly kill real music, and each time a whole genre grew up around it once the fear wore off. Electronic music is the clearest example, an art form built from the very tools that were once dismissed as soulless and unauthentic. We are slow to accept change, but the tool always outlasts the argument about it.

Those conversations reframed my technical observation into something bigger than a research niche. The lopsidedness I had noticed in the papers is not only a gap in the literature. It is a gap that real people are standing in right now, often with no one naming it for them. So much of the anxiety artists feel about this technology is not really about capability. It is about agency. It is about whether they get to stay the author of their own work, or whether they are being quietly reduced to the person who writes the prompt.

And part of that fear comes from a narrow picture of what AI even is. Most people hear the word and think of the one-shot, prompt-based systems, because those are the only ones they have ever been shown. Coming from that perspective, their fears are entirely valid. What it misses is that nothing about the technology forces the automatic, hands-off design. It can just as easily be built to work with you rather than instead of you.

I want to help build this other kind of tool. Not one that takes a creative act and compresses it into a single instruction, but one that widens it, that hands an artist more range, sharper instruments, and a partner that complements their own skills and style rather than muting them. The line I care about is the one between a tool that gives you an output and a tool that participates in your process. Between an AI you prompt and an AI you play with.

That is a hard thing to do well, and I will not pretend to have it solved. It is close to the opposite of a solved problem, which is exactly what makes it worth a career.

Collaborate, don’t automate.