Zugzegk

In June, I wrote about a masked math rock duo from Quebec and the question their music kept raising for me: whether authenticity is about the absence of AI or the presence of human intent. I said I'd land on presence, and I still do. But I didn't expect to be back at this so soon, or for the occasion to be Jimmy Fallon standing on a soundstage trying and failing to pronounce a word.

Monday night, Angine de Poitrine made their U.S. television debut on The Tonight Show. It was also the first time anyone outside their touring crowds had heard the song they played. Fallon introduced it, stumbled charmingly over the title, and then Khn and Klek did their triangle hand sign and disappeared into four minutes of looped, microtonal, deeply strange music that somehow still resolved into something you could nod along to. The song is called "Zugzegk." I am not going to pretend I know how to say it either.

Debuting Live, on Purpose

What strikes me most isn't the song itself, though it's a good one. It's the decision to debut it there, unreleased, unrehearsed for an audience, on the single biggest stage of the band's career so far. There was no studio version to fall back on if it went sideways. No prior performance to point to. Just two people committing to something new in front of millions of people who had never heard of them a year ago.

That's a strange kind of risk to take on purpose. It only makes sense if the performance itself is the point, not the artifact it produces. A generated track doesn't have a debut. It doesn't have a first time it could have gone wrong. It exists complete, or it doesn't exist. What happened on Fallon's stage only means something because it hadn't happened before, anywhere, and everyone watching knew that.

The Mask Isn't Hiding Anything

I've been thinking about the costumes differently since June, too. It would be easy to read the masks and the alien mythology as a kind of shield, a way of not being seen. But watching the Fallon clip, that's backwards. Khn and Klek are more legible in character than most performers are without one. The triangle sign before every set. The robotic vocal intro. The specific, accumulated grammar of a persona built over years of actually performing it, live, in front of people, run after run.

That's not obscurity. It's a very deliberate form of control over how they're perceived, one that took real time and real repetition to earn. You can't fake having a signature move you've done a hundred times before. You can only actually do it a hundred times.

Could a Model Write "Zugzegk"?

The question I keep asking myself is whether a generative model could produce something that sounds like this. Probably closer than I'd like to admit. Microtonality is describable. Loop-pedal structures are describable. A model trained on enough of AdP's catalog could likely approximate the surface texture of their sound, maybe convincingly, to a casual listener.

What it couldn't do is choose to play it for the first time on live television, with the specific vulnerability of a debut, in front of an audience it needed to win over on a specific Monday night in September. It couldn't have broken a string on stage in New York the week before, the way Khn actually did, and kept playing anyway. Imitating a sound is not the same as living through the conditions that produced it. The song is downstream of a decade of two people showing up and doing the work in front of people. That part doesn't generate.

So my answer hasn't really changed since June. It's not that AI categorically can't make something that sounds like this. It's that the thing that makes "Zugzegk" matter isn't in the waveform. It's in the fact that it happened once, live, on purpose, and everyone watching knew it.