AI can already write the perfect pop song
So what can we do now?
I’ve seen the music business change in ways that made it a lot more about the latter — friends who felt burned out selling music they didn’t want to make, others who became veritable merch machines (but really wanted more time to practice the craft), and the “evolution” of pop songwriting attaining the efficiency of Chinese sweatshops. Every time I saw incredulity over “It takes that many people to write one song!?”, it reminded me in an odd way of the mythical man-month.
So here we are, on the cusp of 2026. AI denialism is regularly debunked by technology’s actual capabilities: with all that training, multiple platforms can output the perfect pop song in mere microseconds. Like what PC Music forecast, it can be catchy to the max, pure saccharine earworminess. People really grasping for something will cliché-latch on to its “soullnessness”, failing to acknowledge that across sight + sound, the vast majority can’t tell. Only when disclosure is done does someone say “I don’t like it”, and that’s the worst kind of hindsight bias.
Certain kinds of music, in outre parlance, have become solved problems. That means there’s a finite solution set, it’s well-established territory. We already knew the same chord progressions get recycled — like those savants solving Rubik’s Cubes, mainstream music became a commoditized optimization problem of maximizing the biggest revenue from the shortest streamable hits. Which sounds precisely like what machines are for.
You know what the absolutely great thing is about all this?
It frees musicians up to follow their own quirky tastes instead of trying to outdo the machines — driving more of a decisive divider betwixt said motivations for making music, as in “Do you really love the craft or are doing it for the money?” — and while I don’t think “imperfect” is quite the right contrast (since AI can convincingly simulate wear ‘n’ tear foibles too), it can definitely reflect individual taste, rather than a prompted hive mind.
All this continues to be a moving target, so it requires adaptation, which taps into the human capability for continuous growth. Be curious: even if you hate AI, give the tools a shake, firsthand understand what they can/can’t do. E.g., I’ve tried a variety of the AI-powered musicmaking tools, and still find it frustrating I can’t even specify key or chord progression (basics!) at the song level, and it isn’t outputting exactly what I hear in my mind. I want more playful ways to shape the totality in realtime. However, ElevenLabs does churn out occasionally good “Persian wailers falling down a flight of stairs after chain-smoking 55 packs of cigarettes in D minor at 130 BPM”, so I’ll use those pieces like Legos, raw sample ingredients for a hearty stew.
I don’t need to hear another clone of something that’s already solved, but I do want to be surprised and curious about what I’m listening to. That output can come from a machine or a human — or a hybrid of both! — but it needs to be memorable and add value to my life.
* This image is so very obviously AI-generated, and there’s parts I take issue with: odd note smears, chart gobbledegook, cash stacks cut in two (which would make them unspendable). But then I look at what it did get right: transforming a paragraph of an abstracted concept into a shareable image that roughly visually communicates the point I want to make, which is the net win. It’s crazy how many have become so desensitized and dismissive of mindblowing advances. I recall where we were a few months to a year ago, and as these flaws increasingly fade away without intervention, I’ll miss ‘em like I do Dall-E 2.