Coming Back to AI Music — Two Years Later
I wrote about AI music generation when Suno felt like a magic trick. Re-reading it today, some of it holds up. A couple of things I got specifically wrong — including one I ended the original piece with: “After all, aren’t we all, in some form, neural networks ourselves?” Yeah.
The thing I didn’t account for
When I wrote that piece, the questions felt speculative. They aren’t anymore.
In June 2024 the RIAA filed copyright suits against both Suno and Udio — Sony, UMG, Warner, mass infringement of training data. By late 2025 parts of it had settled: UMG reached a licensing deal with Udio in October 2025 establishing a per-generation royalty rate ($0.002–$0.005), Warner settled with Suno — and handed Songkick over to Suno as part of the deal. Suno is still fighting Universal and Sony, arguing fair use. Independent musicians filed their own class actions, correctly pointing out that major-label settlements protect nobody but major labels.
None of that was in my article. I wrote about creative potential without mentioning that the legal ground under these tools was completely unsettled. That was naive.
Adam Neely said the uncomfortable part out loud
The sharpest public thinking on this came from Adam Neely, whose video “Suno, AI Music, and the Bad Future” hit nearly a million views. His core argument is about craft versus taste.
The optimistic framing — AI handles the production, humans provide taste — he finds uninspiring at best:
“There is very little inspiration to be had in another person’s taste, whereas there’s a lot of inspiration and a lot of direction that you can gain from another person’s craft.”
The musicians that shaped people — the ones worth aspiring toward — were compelling because of what they could do. If the next generation has no craft to look up to, only preferences, Neely argues something important dies. He worries about de-skilling the same way early manufacturing de-skilled craftsmen, but this time it’s the art form itself getting hollowed out.
He also ran a survey. Why do people use Suno? Save time. Save money. Replace friends. His summary: “Suno lets you make the same music faster, cheaper, and lonelier.”
And most respondents said the music didn’t feel particularly theirs.
Nick Cave is not being precious
Nick Cave has said this stuff multiple times and he’s not softening it. After trying Suno in 2024, he called it “utterly banal” with “no soul or spirit.” When someone earlier sent him ChatGPT-generated lyrics in his style, he called it “bullshit, a grotesque mockery of what it is to be human”.
But the quote that stuck with me:
“Its intent is to completely sidestep the sort of inconvenience of the artistic struggle, going straight to the commodity… We don’t make things anymore. We just consume stuff. It’s frightening.”
And he’s clear that he’s not worried about his own livelihood:
“I’m not worried about my own job or something like that about being replaced. Just what it’s saying about us as human beings.”
That’s a different argument than most musicians are making. It’s not economic, it’s civilizational. The struggle isn’t an obstacle to the output — the struggle is the point, and we’re choosing to skip it.
Rick Beato can’t bring himself to call it bad
Beato has testified before a Senate committee about AI regulation. He’s worried about who captures the economic value. But when he actually listened to AI-generated music, he said this:
“It’s an incredibly well written melody that the arrangement sounds exactly like what you would expect to hear on a well-produced pop song today.”
He can’t call it terrible when it isn’t. That honesty is worth something. The argument “but it sounds bad” is getting harder to make, and pretending otherwise doesn’t help anyone.
He’s also predicted that in 20 years people will say they prefer AI Rolling Stones to the real ones. I don’t know if that’s right. But I don’t know that it’s wrong either.
Not everyone is fighting it
Not everyone resists. Holly Herndon harmonized with her AI twin “Spawn” on Proto back in 2019 — using it as an instrument rather than a replacement. Grimes opened her AI vocal model to fans via Elf.Tech, letting them release music using her voice in exchange for a 50% royalty split in her favor.
These are minority positions in the discourse but they point at something real: there’s a version of this where the technology expands what a musician can do. I use AI music this way too — more for inspiration than anything else, and sometimes something comes out that’s genuinely interesting enough to share. The ones I thought were worth keeping are on my page.
What actually happened while I wasn’t paying attention
Suno is generating 7 million songs daily. That’s “an entire Spotify catalog worth of music every two weeks,” per their investor materials. Deezer went from 10,000 AI track uploads per day to 50,000 by late 2025 — and 75,000 by spring 2026. iHeartRadio launched a “Guaranteed Human” program banning AI synthetic vocals from airplay. AI tracks entered Billboard charts — Breaking Rust hit #1 on Country Digital Song Sales.
Recording Academy CEO Harvey Mason Jr. said “every” songwriter and producer he knows has now used these tools. Suno ran songwriter camps at Rick Rubin’s Shangri-La studios.
ElevenLabs — primarily known for voice cloning — entered music properly. Sound effects in May 2024, full music generation in 2025, and Music v2 in May 2026 with genre-switching mid-track. Critically, they licensed from Merlin Network and Kobalt before launching, which is why they’re not in court. That contrast is not subtle.
Voice cloning stopped being hypothetical too. The “Heart on My Sleeve” deepfake — fake Drake/Weeknd collab — went viral in 2023. Drake later used AI to fake Tupac’s voice in the Kendrick beef. Tennessee passed the ELVIS Act, the first AI-specific voice protection law. Randy Travis recovered his singing voice through AI after a stroke. These aren’t thought experiments anymore.
I used the phrase “complement, not replace”
Everyone did. It’s technically defensible and practically misleading. For background music, licensed tracks, game audio, jingle-adjacent work — AI has replaced human musicians. Fully. The work was there; it no longer goes to humans. “Complement” is what you say to avoid that sentence.
I was also too quick to dismiss the na jedno kopyto critique. I argued that skilled prompting produces variety. True. But skilled prompting is rare, most output is generic, and the internet got flooded with the modal case — not the exceptional one. My friend was observing the ecosystem. I was defending the technology. We were answering different questions. He was more right than I gave him credit for.
And then I closed the original article with “After all, aren’t we all, in some form, neural networks ourselves?” — a line that deserved to be cut before it was ever published.
The value is real. The market for it is the question.
Musicality is still fundamentally about communicating an emotional state from one person to another. That’s still true.
But for a large portion of music consumption, listeners don’t care about that at all. The question isn’t whether human music has value. Obviously it does. The question is how large the market for that value will be — and whether the infrastructure that sustains musicians survives long enough to matter.
Cave is right that the struggle is the point. He’s also someone who fills concert halls globally. The musicians for whom this is most existential aren’t the ones with platforms to discuss it.
The conversation got sharper in two years. The technology got better. The arguments got harder to dodge.