Notes on AI and Music, from a Lifetime of Making It

This isn't an attempt to predict what AI will do to music. It's a look at that question through almost a lifetime spent making it.

Falling into the details of a song

I remember back when radio was still one of the main ways people found new music, I could never understand the ones whose answer to "what do you listen to?" was "I listen to station X." How could anyone name a channel instead of an artist? How could you not fall all the way into the details of a song and then try, tirelessly, to make something like it yourself?

I've spent some thousands of hours making music. Grinding out MIDI tracks and home recordings at the kitchen table. Recording in the bunker, alone or with a rotating cast of friends and bands. Sweating over demos and reshaping them with other people. As a teenager I could spend days without leaving the house, because I was making songs.

Hundreds of half-finished demos

And after all of it what I have is hundreds of half-finished demos on a hard drive. Not one of my songs ever found an audience, and I've never played them live. That was at least partly a conscious choice. Part laziness, part clear-eyed knowledge of how much money, effort, and luck real success demands, and part something simpler: the point was only ever to make the music. It's just that much fun.

The first time I heard my demo as a real song

I still value real, human-made "art." But I can't pretend I didn't feel something the first time I heard one of my own demos produced by Suno. What used to take blood and sweat in a studio appeared in front of my eyes. All the instruments played, the lyrics, and vocals. A demo carried across the line into something you could actually sit down and listen to as a real song and iterate with feedback.

Music for the everyday musician

I think the world is full of what you might call everyday musicians: people with a real drive to make art, but not necessarily the skill to play or sing it, or the network, or the money to realize their ideas. For people like me, AI offers a way to see our songs taken one step further. That is where it produces genuine value, and genuine joy.

My own skills lean less on technique and more on obsession, and on a flexible imagination for hearing how a song wants to take shape. Maybe that's why I've come allergic at a certain type of AI criticism: the attitude that making music belongs only to some elite. AI democratizes music. It frees a lot of people into their own creativity.

The process isn't the point, the result is

Seen that way the question of what's human and what's machine stops being so relevant. The process isn't the point. The creative work and the result are. It's no different with people: my own ideas have often only come alive after jamming with others, after someone widened the horizon of what an idea could hold. Working with AI, at the composition stage, is the same thing.

It's worth remembering that in music the actual ideation (the composing, the creative leap) and the phase where a song is played, recorded, and produced are usually two different things, and they call for different skills. And it's in that first phase, being able to spin an idea into a version fast enough to know immediately whether it's worth anything, fast enough to put it in front of a band, that I think AI will be most useful.

The dark side: flatter, duller, and the "Suno EQ" sheen

But I see the dark side too. What happens if human contact fades out of music-making? That would make the world, and the art, flatter and duller. At the same time, streaming platforms fill with AI bloat prompted out of thin air, thinning the whole catalogue. We'll see and are already seeing abuse, and fights over artists' rights. And the more raw, prompt-and-publish material gets pushed out, the more songs start to resemble each other until some listeners get used to hearing only a certain sound, that washed-out, hissy "Suno EQ" sheen. That's one of the parts I dislike most.

What about artists who live off music?

I can also try to understand artists whose living depends on music. I know that making money in this world takes authenticity, promotion, all of it. That doesn't mean AI won't increasingly sit in the labels' toolbelt when material is produced for artists, but how that plays out in revenue models, in how we value artists, in virtual artists and the rest, remains to be seen. Nowadays I guess it's all about profits and efficiency, unit economics.

If I were genuinely skilled on an instrument, I don't think I'd be worried. I'd be trying to figure out how to stress-test the new technology inside my own process. There are going to be unfair advantages here, for the artists and producers who get it. And even then the production can be 100% human-made, with AI simply helping to widen the ideas.

It's also worth remembering that technology, synthetic instruments, algorithms and the rest have been part of music for a long time already.

New platforms and models AI might create

When we ask what AI does to music, it helps to step outside the black-and-white split and accept that there is no single correct way to make or listen to music. They're all valid. And while AI changes the ways of producing and distributing music, it may also create entirely new platforms and models. Off the top of my head:

Personalized songs, a new hit every day. I don't think this kind of listening needs big artists at all, but it answers exactly that "I just want a certain vibe" or "I want this mood right now" impulse. It makes music more exciting, but also disposable, snackable. Is that wrong? At worst, listeners' patience for staying with a single song drops, and artists put out faster, more easily tailored content. That's a real headache for Spotify and the rest, and maybe how Suno vs. everyone else shakes out is anyone's guess.

Music you make yourself. The masses probably won't touch Suno yet. It costs real money and takes a real desire to make something. But we'll surely see "I make my own music" listeners emerge. I notice I'm partly in that camp myself, because it's genuinely exciting to see my own material turned into songs that fit a different genre or subject. I've even built my own AI agent I talk through topics, lyrics, and production with. It can take a single guitar recording and hand me ten different directions to develop in seconds.

These notes may already be out of date

My guess is that some of these thoughts will already be out of date in a few months. Maybe by tomorrow.