There is an old phrase I keep returning to lately: the proof is in the pudding. You can argue about a recipe forever — the ingredients, the technique, the confident promises printed on the box — but eventually someone has to taste the thing. And when I listen to the loud, certain claims being made about AI music tools right now, that anyone can make a song and the musician is about to be automated out of existence, I reach for that phrase. Because the pudding is sitting right in front of us, and it is not made of generated audio. It is made of people, in rooms, watching other people play.
I should declare my bias early, because I have a large one. I am an engineer by training and an entrepreneur by trade, but I have played keyboard and guitar since I was small — through the long, shapeless summers when there is nothing to do but sit in a bedroom and play the same four bars until your fingers stop arguing with you. So I am not a neutral party here. I love these machines. I produce electronic music as a hobby. This is not a sermon against the tools, and I am not trying to drag anything down. It is a reflection on what the tools are, and on what they quietly are not.
The Rooms Are Full
Start with the evidence, because it is overwhelming and it is public. If generated music were genuinely replacing the human kind, you would expect the appetite for live performance to be cooling. The opposite is happening. Tours are booking out. Artists are crossing the planet, playing as many nights as their bodies will allow, and audiences are flooding into venues — not despite the flood of recorded and now generated music, but, I suspect, partly because of it. The more frictionless and infinite the supply of audio becomes, the more people seem to crave the one thing it cannot manufacture: the presence of another human being, sweating under the lights, occasionally getting it wrong, doing it anyway, with you in the room.
in a single record year
vs. pre-pandemic 2019
the largest ever recorded
The headline revenue and attendance figures come from Live Nation's own year-end reporting, so I treat the superlatives as the company's claim rather than settled fact. The growth and single-tour numbers come from independent tour-tracking (Pollstar), which corroborates the direction even where it tempers the hype: the rooms really are fuller than they were.
I want to be careful with these numbers, because the most quotable of them come from the companies that profit when they look good, and a chief executive announcing the biggest year ever is making a claim, not stating a law of physics. But the independent trackers tell the same story from a different seat, and what both describe is not a public quietly abandoning live music for an infinite jukebox. It is a public paying more, travelling further, and showing up in greater numbers to be in the same space as a performer. That is the pudding. Everything else is the recipe argument.
A Tool Is a Tool
So where does that leave Suno, and the dozens of platforms like it, promising finished songs on demand? Exactly where it leaves every tool that arrived before them: useful. Genuinely, sometimes thrillingly useful. These are instruments you can fold into a production flow to get somewhere faster, to sketch an idea, to hear a possibility you would not have reached alone. I am not interested in pretending they are toys.
But I have watched this film before, and so has anyone who has spent time around a studio. The synthesizer was going to make the pianist obsolete. Multitrack tape, then Pro Tools, then GarageBand — each was announced as the moment the musician became unnecessary, the moment the craft got automated. And each one turned out to be an instrument, not an executioner. They did replace some older things; the upright piano lost ground, certain session jobs vanished. But they did it by handing us new sounds and new ways to make them. Electronic music did not kill music. It gave us more music, and I am one of the people grateful for it.
"Every new instrument arrives wearing the costume of the thing that will replace us. It almost always turns out to be the thing we get to play next."— GlamBon Lifestyle
Where the Feel Lives
Here is the part I can only describe from the inside of my own hands. There is a difference between a person playing and a grid playing, and I want to be honest about it because the honesty is the whole point. The lazy version of this argument is "you can always hear it." I do not believe that is true. In a genuine blind test, I would not reliably pick the human every time, and neither would you — especially in the heavily produced electronic styles where the grid is supposed to win. Detection is not where the difference lives.
The difference lives in feel. A person playing pushes and drags against the beat by milliseconds, leans on some notes and ghosts others, and never plays the same bar twice in exactly the same way. Quantization — snapping everything to a perfect machine grid — is a gift when you want it and a flattening when you do not. Drag the control below and listen with your eyes: watch what happens to the groove as the human hand is pulled toward the perfect grid.
Where the Feel Lives
Drag from human timing to machine-perfect grid
Slide toward the grid and the dots line up into a clean machine pulse. Slide back toward the hands and they scatter - early here, late there, never quite repeating. Neither is better. But they are not the same thing, and a body in a room can tell.
Notice what the slider does not claim. It does not say the grid sounds worse, or that you could spot it blindfolded. It says the human version carries variation and risk that the perfect grid smooths away — and that whether you can name that difference is a separate question from whether you feel it.
Take Away the Training
There is a second thing people say less carefully than they think, and I have said it myself in a hurry. Strip the training data out of a music model and there is nothing left: no patterns, no taste, no sense of which note should follow which. This is completely true. A generative model is, in the most literal sense, a recombination of human-made music it was shown. Remove what it was shown and it has nothing to say.
But here is where the comparison breaks, and the break is the whole point. A machine and a musician are not in the same situation, because a musician feels the music. Strip every song I have ever heard out of my head and I am not left with silence — I am left with a body that still responds to sound, that still reaches for a note because it feels right rather than because some pattern says it comes next. I could sit at a keyboard knowing nothing and still find my way to something that pleases my own ear. That is not a thought experiment. It is the actual origin of music: someone, with no catalogue to copy from, made a sound that felt good and then made it again. On the first day there was no training data. There was only feeling, and a hand.
A machine cannot do that, because there is no ear to please and nothing it would ever call good. Remove its corpus and it does not improvise from feeling — it simply stops. That is the asymmetry the convenient phrase is designed to hide. Yes, I learned from records and teachers and a friend's older brother who knew one cool thing; but I passed all of it through something that wanted to say something. The model passes it through nothing. It inherits the grammar of music and not the reason music exists — and that reason is not a part you can bolt on afterwards. It is the whole thing the patterns were invented to carry.
Filler and Fire
Let me turn the knife on myself, because it is only fair. I produce electronic music. I love it. I will defend a great electronic record against anyone. And yet, when I am honest, a great deal of it — including some of my own, and including plenty of what sits inside a DJ set — functions for me as filler. I put it on while I work. It fills the room beautifully without asking anything of me. I do not mean that as an insult. Filler is a real and useful category; not every piece of music needs to reorganise your insides, and a room that is pleasantly full of sound is a good room.
But it is a different thing from the music that does reorganise you — the live take, the voice cracking on the note it should not quite reach, the band locked together and pulling a thousand strangers into the same breath. I am describing what those two things do to me, not running a detector over them. One is wallpaper I am grateful for. The other is an event I will remember in ten years. The tools are very good at the first. They have so far shown me nothing of the second, and I am no longer sure the second is the kind of thing a tool can be good at.
The Proof Is in the Playing
So, to be clear, because I do not want to be misread as a man shaking his fist at a synthesizer: are AI music tools real and worth using? Completely. Do I use them? Already. Are they instruments in the long, honourable line of instruments that were each supposed to end music and instead enlarged it? Yes. The case against them as creativity-killers is, I think, mostly a failure of memory about every previous tool that earned the same funeral.
But the larger question — whether a generator replaces the musician — was never going to be settled by argument. It was always going to be settled by the rooms. And the rooms have answered. They are full of people who travelled and paid and queued to stand in the dark while a human being plays, imperfectly, in real time, the patterns they spent their childhood learning to feel. You can generate the song. You cannot generate the wanting to be there. The proof, in the end, is in the playing.
Frequently Asked Questions
On current evidence, they relocate effort rather than replace people. Tools such as Suno are powerful instruments that speed up parts of production, but demand for human performance is rising, not falling. Live touring has hit record revenues in the same years generation became abundant — which suggests audiences increasingly value the thing a generator cannot supply: a person performing in real time, in a shared room.
The likeliest reason is that scarcity has moved. When audio becomes effectively infinite and frictionless, the rare and valuable thing becomes presence — being in a room while a human performs. Independent tour-tracking shows top worldwide grosses up roughly 71 percent since 2019, so the abundance of recorded and generated music has coincided with greater, not lesser, hunger for the live experience.
In function, yes. Every major music technology — the synthesizer, multitrack tape, Pro Tools, GarageBand — was introduced as a replacement for musicians and instead became an instrument that widened what musicians could make. AI generation fits that lineage: it changes the production flow and adds possibilities rather than removing the human creative role. Some older tasks shrink; the craft adapts and continues.
No — without training it produces nothing, because it has only patterns to recombine and no feeling to draw on. People are not in that position. Strip away everything a musician has heard and they can still reach for a sound that pleases their own ear; that is how music began in the first place, with no catalogue to copy. So the honest distinction is not merely where patterns come from. It is that human music starts in feeling, and a machine has none.
Not always, and the evidence is mixed — in controlled blind tests many listeners cannot consistently tell the two apart, particularly in highly produced electronic genres. The more honest claim is about feel rather than detection: performed music carries micro-timing, dynamic variation, and risk that machine-quantized audio tends to smooth away. Whether someone can name that difference is a separate question from whether they feel it in the room.
