The case for being unpromptable
What Angine de Poitrine tells us about building an audience in the age of AI
Two guys in papier-mâché monkey masks and polka-dot jumpsuits just became the most talked-about band on the internet.
Angine de Poitrine are a duo from Saguenay, Quebec. They play microtonal math rock on a custom-built double-necked guitar-bass hybrid that took a local luthier 150 hours to make. The frets are spaced at quarter-tone intervals. The guitarist can barely see through his mask, so the fret markers are oversized and phosphorescent. Their songs are largely instrumental, built on odd meters and drawn from Indian ragas, Turkish psychedelia, and Gamelan. They’ve been playing together since they were thirteen.
In February, KEXP uploaded a video of the band performing at the Trans Musicales festival in France. It now has over 5 million views. Their debut vinyl, self-released in 2024 in a run of 200 copies, started selling for $300-$600 on Discogs. Their entire 2026 tour is sold out.
One of the top comments on the YouTube video reads: “This is the only way we can win the battle against AI.”
The flood
Here’s what was happening in music while Angine de Poitrine were building a grassroots following on the Quebec festival circuit.
By January 2026, Deezer was receiving 60,000 fully AI-generated tracks every day — roughly 39% of all uploads. Deezer’s own study found that 97% of listeners couldn’t tell the difference between AI and human-made music in a blind test. Suno, the largest AI music generation platform, generates more new tracks every two weeks than the entirety of Spotify’s catalogue. Its CEO Mikey Shulman went on a VC podcast and said, with a straight face, that making music “is not really enjoyable” because it takes too much practice. He later said he wished he’d chosen different words.
Doug Shapiro has written persuasively about what happens when content becomes infinite: attention stays fixed, so the value of any individual piece of content converges toward zero. The only things that retain value are filters, curation, and trust. I’ve written about this same dynamic in gaming: 19,000 games launched on Steam in 2025, and the median sold 500 copies. Too much content chasing existing audiences through established distribution channels.
Music just got there faster. AI-generated tracks account for 39% of Deezer’s uploads but only 0.5% of actual streams. 85% of their streams are fraudulent. The content is technically competent and functionally worthless.
Unpromptable
Try typing “microtonal math rock performed by anonymous duo in papier-mâché monkey masks playing a custom double-necked quarter-tone guitar-bass hybrid influenced by Gamelan and Turkish psychedelia” into Suno. Even if it could parse the instruction, what came out would be a statistical average of training data. It would sound like something. It wouldn’t sound like this.
AI music tools are interpolation machines. They produce the weighted average of their training data. They’re brilliant at the middle of the distribution and structurally incapable, I’d argue, of generating true outliers. Everything about Angine de Poitrine exists in the gaps - between notes, between genres, and between what algorithms know how to classify. The music sounds wrong in a way that makes your brain reach for it.
I wrote a while back about how the best products tell people to fuck off.
Why the best products tell most people to f*** off
Elden Ring sold 20 million copies by telling most gamers they weren't good enough to play it.
My argument: in a world of infinite choice, being “pretty good” is a death sentence. You have to be specific enough to create identity. The people who love Angine de Poitrine don’t just enjoy the music, they become evangelists. This is the Marmite principle applied to the age of generative AI: the less your work resembles anything a model could produce, the more it spreads.
Illegibility as distribution
The lesson here isn’t “be weird for the sake of it.” Angine de Poitrine have been playing together for twenty years. The instrument was custom-built over 150 hours. The microtonal vocabulary comes from years spent studying non-Western musical traditions. The weirdness is earned.
The real lesson is about illegibility. The early-stage investments I am most excited about are in companies that can’t be easily pattern-matched - that don’t look like obvious winners until they are. There will be “obvious” winners too, but they are likely to be priced that way.
The same logic now applies to creative work, and increasingly to products. In a world where AI can produce competent versions of anything that’s been done before, the margin is in things that haven’t been done before.
Suno’s CEO says taste is the only thing that matters and skill is going to matter less. I think the opposite is happening. Skill that produces genuinely novel output - not just technically proficient but truly original - has never been more valuable, because it’s the one thing the machines can’t fake.
60,000 AI-generated tracks hit Deezer every day, and a duo in monkey masks just sold out a world tour. I know which side of that trade I’m on.
If you’re building something at the edges — something that couldn’t have been prompted into existence — I’d love to hear from you. That’s what we invest in at F4 Fund.