There is a moment, around the twentieth track in Tuesday's demo batch, when something makes you pause. It's not a drop or a breakdown: it's a feeling. A texture that doesn't quite close. A melody that resolves too perfectly. An arrangement that seems straight out of a sound design manual on algorithmic steroids. And then the uncomfortable question appears: did a person make this, or a machine trained to sound like one?
At Eclipse Records we receive between eighty and one hundred twenty demos per week. The number fluctuates, but the trend is clear: more and more material arrives with some degree of artificial intelligence intervention. We're not talking about tracks generated entirely from a prompt—though we've received those too—but productions where AI acts as an invisible co-writer: drum patterns suggested by a model, bass lines completed by a generative assistant, automatic EQs that promise the ideal spectral profile for Beatport.
The tool is not the problem
Let's be clear from the start: at Eclipse we have nothing against technology. We are an electronic music label, not a 1993 vinyl club. The sampler was heresy, the DAW was heresy, sync was heresy. Every new tool went through its season in purgatory before becoming standard. AI is no different in that regard.
What changes is the scale. When a producer can generate forty variations of a groove in the time it used to take to program one, the volume of material in circulation multiplies. And that changes the A&R function in ways we are still understanding.
Listening is no longer just listening
Traditional A&R operated with a relatively simple filter: Is this good? Does this fit the catalog? Does this person have something to say? The first question still stands. The second too. But the third has suddenly become more complex, because now it involves discerning whether what we are hearing is a genuine artistic statement or a statistical optimization of what works in the charts.
We have developed some signals. The most reliable, as old-fashioned as it sounds, is non-deliberate imperfection. A fade that cuts a millisecond earlier than expected. A resonance that no algorithm would have let through. An awkward silence between two sections. AI tends to polish, to resolve, to close gestures. The human producer, even the most technical, leaves cracks. And it is those cracks that interest us more and more.
What no model can simulate
There is something AI cannot replicate, and it's not a matter of technical capability but of context. A track written at four in the morning after a club session that blew your mind. A mix decision made by intuition, not by training on ten thousand references. A sample that entered the project because it was playing at the corner bar while the producer walked home. That cannot be trained; it is lived.
At Eclipse we have started asking for contexts, not just files. An email line that tells where the track comes from, what triggered its creation, what was playing in the room when it was composed. It's not a bureaucratic filter: it's a way to reintroduce narrative into a process that technology tends to make abstract.
The human ear remains the competitive advantage
Paradoxically, the more production is democratized by AI, the more valuable human curation becomes. Anyone can generate a track that sounds professional. What not everyone can do is build a criterion, sustain an editorial line, understand what a catalog needs to have coherence without becoming predictable.
That is the real work of the A&R in 2026: to be the human counterweight to an ecosystem that produces more music than anyone can listen to. The machine accelerates, the ear slows down. The machine averages, the ear dissents. The machine optimizes, the ear takes risks.
And at the end of the day, when we turn off the monitor after reviewing the demo batch, what we remember is not the track that best followed the rules. It's the one that made us forget we were working.