Our A&R Can't Hear AI — and That's Exactly the Right Problem

Somewhere in the last year, artificial intelligence stopped being a headline in production and became furniture. Pattern generation, sound-design experiments, rapid prototyping, mix reference — the tools are simply part of how records get made now, the way a compressor or a preset library always was. For a label, this creates a question we are asked constantly and answer honestly: can you hear it? Not reliably. And that is the point.

There is no waveform test

A&R is not forensics. We are not running spectral analysis to determine how a hi-hat was conceived, and any label that claims to have an AI detector in its workflow is describing aspiration, not practice. The tools have moved faster than the tests. What we can hear is something else entirely: whether a record has a point of view.

And that is a much harder thing to fake than authorship. A track assembled with heavy assistance can still be a real record — if a human made thousands of decisions about what to keep, what to cut, what to refuse. A track made entirely by hand can still be worthless — if no one would defend it.

The three questions we actually ask

• Does this have a stake in the ground? A record needs to take a position: this tempo, this texture, this refusal to do the obvious thing. Tools do not have positions. People do.
• Would the artist defend this in an argument? Every track we sign, we interrogate. If the producer cannot tell us why a bar exists, or what they cut to get there, the record is a coincidence.
• Does it survive without its novelty? Whatever made it feel fresh on first listen — does the core still work on the fifth? Novelty expires; structure does not.

We are not signing the process. We are signing the judgement that shaped it.

What a label owes the scene

There is a commercial temptation in all of this. If a label could generate release after release with minimal human input, the margins look attractive — until the whole catalogue sounds the same and listeners quietly stop caring. The market for novelty is shallow and unforgiving. The market for identity is deep. Our incentive, as a curator, runs the opposite way to the seductive shortcut: the more capable the tools become, the more the value concentrates in the people who know how to say no.

That also changes what we ask of artists. Disclosure is not a purity test here — it is a courtesy that keeps trust intact. If a record leans on machine-assisted composition, we would rather know, so we can judge the result honestly and represent it accurately. What we will not do is pretend we can police it when we cannot.

The honest position

AI will keep dissolving the line between tool and collaborator, and most of that is neither scandal nor salvation — it is just the next chapter of a studio story that started with drum machines. The label's job was never to certify how a sound was built. It was to certify that the sound deserved to exist in a catalogue with our name on it. That test has not changed. It is just harder to hide behind technique now — for the artists, and for us. In a world where anyone can generate a plausible track, taste becomes the only scarce resource. We intend to spend it carefully.