Beyond the Prompt: Why AI Music Still Feels Empty to a Producer
Let's talk about being a creative in the times of AI.
Recently, I've been noticing the massive jumps AI is making in fields outside of my primary art.
As a programmer, the progress is crazy. I know many languages and have built apps — but I lack the deep, certified knowledge of a computer scientist. Already, AI is creating code that is more complex than what I can write, and it's certainly doing it faster than I ever could.
Then there is video generation. I recently saw a new model — Seedance 2.0 — that generated a realistic boxing match between Ronaldo and Messi. The faces were coherent from beginning to end. It looked unbelievable.
Seeing that mind-blowing progress in code and video made me wonder: What is really happening with music?
The Problem of Specificity
Perhaps it's just my point of view because I am deeply embedded in making music, whereas I'm only a casual observer of video art. But when it comes to music, I haven't yet heard anything AI-generated that I actually like.
The current output sounds incredibly generic — mainstream pop or country caricatures.
For an electronic musician, the process is defined by extreme specificity. We don't just want "Detroit-sounding drums." We want a bass drum that is very low and has a specific "knock." We want a hi-hat with an exact decay length, processed with a very specific type of high-tech granular reverb — not just "some vintage verb."
Crucially, we want sounds that set us apart from everyone else.
Right now, using generative AI feels a lot like using early image generators: you generate 1,000 images just to find the one in the middle that works. Are you really composing, or are you just fishing for happy accidents? It's rare that you write a prompt and the output sounds exactly like what you heard in your head.
The Addiction of the Loop
When you are just starting out, you probably don't care about that specificity. You are just happy you put a beat together. The magic of opening your DAW or firing up a drum machine, where every session creates something completely different, is incredibly addictive. It makes you happy.
Beginners might get that same feeling of amazement from prompting an AI. But after a while, when you stop doing it just for yourself and start sharing it, criticizing it, and understanding style, you want more.
There is a tactile joy to production that current AI lacks. Hitting drum pads feels nice. Working in a loop, feeling the track develop in real-time, tweaking the filter cutoff on a synthesizer while the sequence runs — that is the fun part. That "live" feeling is essential.
Typing text into an interface and waiting for it to generate a finished file is a vastly different, and much more passive, experience.
The Future is Real-Time
I believe there is a space for a new kind of AI music software. We need tools that bridge the gap between generative power and that "live feeling."
We need AI that sits inside the loop with you, responding in the moment as you tweak parameters, rather than just spitting out fully formed audio files based on a text prompt. (Some existing tools, like XLN Audio's XO, are stepping in the right direction of organizing sounds intelligently, which is fun.)
This has changed my own roadmap. Previously, machine learning felt like high-mighty IT science requiring massive cloud processors and thousands of euros to train models. But I'm realizing it might not be that inaccessible.
For the next music software I create, I plan to focus heavily on machine learning. The goal isn't to replace the musician, but to give them a collaborator that can jam in real-time.
Originally published on Medium