There is a moment in songwriting that nobody talks about in interviews, because it sounds too small to be important.
You have a voice and an acoustic guitar. You have a lyric that feels right, or close to right. You have a melody that does the thing a melody is supposed to do, which is to make the words feel inevitable. This is the part people romanticize. The poet with the instrument. The singer-songwriter at the open mic. The idea.
But the idea is not the song. The idea is the seed of a song, and the distance between a seed and a finished recording is where most of the actual work lives, and most of the actual joy.
I.
Consider the bass.
You have your voice and your guitar and your lyric, and now you need a bassline. This is, in a real sense, an engineering problem. The bass has to lock with the kick drum. It has to support the harmonic structure without duplicating the guitar. It has to move through the changes in a way that feels both surprising and inevitable. It has to groove.
You could play root notes. Dun. Dun. Dun. Dun. One note per chord change, landing on the downbeat, following the guitar like a shadow. This works. It is the bassline equivalent of a beige wall. Functional, invisible, and forgettable.
Or you could do what Paul McCartney did on "Something," which was to write a counter-melody so good that it could be its own song, a bass part that converses with George Harrison's vocal line rather than simply accompanying it. Or what Flea did on "Around the World," which was to treat the bass as a percussion instrument, popping and slapping and syncopating against Chad Smith's drums until the two of them sounded like a single organism with eight limbs. Or what Mike Dirnt did on "Longview," starting the entire song with the bass, alone, establishing a riff so iconic that the guitar almost feels like it's joining the bass's band. Or what Kim Deal did on every Pixies record, which was to play with a simplicity so deliberate it bordered on defiance, holding down root notes and fifths with a tone so fat and distorted that the restraint became the statement.
These are all different solutions to the same engineering problem. Each one took time to find. Each one involved a musician sitting with the song, trying things, discarding things, trying more things. Playing the part one way, then another, then going to get a beer and coming back and playing it a third way that was somehow better than the first two.
This is the part that feels good. Not in an abstract, "I love my craft" kind of way. In a physical, immediate, dopamine-releasing way. The moment when the bassline clicks in with the drums and the guitars and suddenly the song is a song, not just an idea with a chord progression. The moment when you look at the other people in the room, or at the screen if you're alone, and something registers in your chest that is impossible to fake.
Countless musicians have felt this at 2 AM on a Saturday, or on a Sunday afternoon when the light is coming through the studio window at a low angle, or in a basement that smells like carpet and old amplifiers. The moment of "we nailed it." The eureka after the toil.
Suno can generate a finished track in eleven seconds.
II.
Let me be precise about what I mean by "finished."
Suno, and Udio, and the other AI music generators that have proliferated since 2024, do not produce a voice and an acoustic guitar and then invite you to figure out the bass. They produce the entire arrangement. Drums, bass, guitars, keyboards, strings if you want strings, horns if you want horns. Mixed. Mastered. Ready to upload.
The output is not always good. But it is always complete. And completeness, it turns out, is the thing that compresses the joy out of the process.
Because the joy was never in the completeness. The joy was in the incompleteness. In the gap between "I have a song" and "I have a finished song." In the hours and days and sometimes weeks of figuring out how to fill that gap. In the wrong turns and the happy accidents and the argument about whether the chorus needs a tambourine.
When you compress that process to eleven seconds, you don't just save time. You eliminate an experience. A specific, embodied, human experience that has no equivalent and no substitute.
You eliminate someone's Saturday night.
III.
But here is where it gets complicated, because the layers of craft in music are not all loved equally by the same person.
The songwriter who lives for the poetry, who spends an afternoon finding exactly the right word for the second line of the bridge, who plays the guitar and sings and feels the song in their throat and their fingers — that person does not necessarily love equalization. They do not necessarily find deep satisfaction in the mathematics of high-pass filters and low-shelf frequencies and the psychoacoustic properties of stereo imaging. They might not care whether the vocal is compressed at a 4:1 ratio or a 6:1 ratio. They might actively resent having to think about it.
But someone does care. The mastering engineer cares. The person who spent years training their ears to hear a 200-hertz buildup in a muddy mix, who can listen to a track and say "the vocal is sitting on top of the guitars instead of in between them" and know exactly which three knobs to turn — that person finds the same satisfaction in solving a mastering problem that the songwriter finds in landing a lyric. Different layer. Same eureka.
Logic Pro now has AI mastering. You press a button. The algorithm analyzes the frequency spectrum, adjusts the dynamic range, applies limiting, and produces a master that sounds, to most ears, professional. The mastering engineer's Saturday night eureka has been compressed to a button press, just as the songwriter's arrangement eureka has been compressed to a text prompt.
The songwriter might celebrate this. Finally, I don't have to worry about the technical stuff. I can just write songs.
The mastering engineer does not celebrate this.
And this is the pattern that repeats across every domain. Every layer of craft has its devotees, its people who find deep satisfaction in solving that particular kind of problem. When you automate a layer, you liberate everyone who found that layer tedious. And you dispossess everyone who found it fulfilling.
The liberation and the dispossession are the same event.
IV.
This extends far beyond music.
A software engineer spends three days debugging a race condition. The code is failing intermittently, in a way that depends on timing, on the order in which threads execute, on conditions that are different every time you run the program. The engineer reads stack traces. The engineer adds logging. The engineer stares at the ceiling. The engineer goes for a walk. The engineer wakes up at 3 AM with an idea, opens a laptop, writes six lines of code, and the bug is fixed.
Those six lines took three days to find. An AI coding assistant can now find them in forty seconds.
The engineer's three days were not wasted time. They were the process by which the engineer understood the system deeply enough to fix it. The understanding was the work, and the understanding was the reward. The eureka was the culmination of a process that had value in itself, not just instrumental value but experiential value. The engineer felt something when those six lines worked. Something that cannot be replicated by reading the AI's output and saying "oh, that makes sense."
But here is the tension: nobody is paying the engineer to have experiences. They are paying the engineer to fix the bug. And if the bug can be fixed in forty seconds instead of three days, the economic logic is obvious.
The eureka is a tax. A beautiful, fulfilling, deeply human tax on the process of getting from problem to solution. And like all taxes, there are powerful forces aligned to eliminate it.
V.
The large language models are, in many ways, oblivious to the texture of how work actually happens now.
Ask an LLM to plan a software project and it will produce something like this: Week 1, set up the repository and establish the architecture. Week 2, implement the authentication system. Week 3, build the core features. Week 4, testing and refinement.
This is how projects used to work. This is not how projects work anymore.
A developer with the right tools can set up the repository, establish the architecture, implement authentication, and build the core features in a single afternoon session. There is no Week 2. There is no Week 3. There is a person sitting at a desk for four hours on a Tuesday, prompting and coding and prompting again, and by dinner the thing exists.
The models are trained on the old world. They have ingested thousands of project plans written by people who assumed that work takes time, because it used to. The models reproduce that assumption faithfully, proposing timelines that bear no relationship to the actual pace of AI-assisted development. They are planning documents for a world that no longer exists, generated by systems that created the world that replaced it.
This is a small absurdity, but it points to a larger one. The cadence of work — the rhythms and rituals that organize how humans spend their productive hours — was never just about efficiency. It was about being human in time.
VI.
Monday morning has a feeling. You know the feeling. The coffee is slightly more deliberate. The inbox has accumulated over the weekend. There is a particular energy, caffeinated and anxious and oddly optimistic, that comes from re-engaging with problems after two days away. People talk about "hitting the ground running," which is a phrase that only makes sense if you've been off the ground.
Friday afternoon has a different feeling. The wrapping-up feeling. The "let's not start anything new" feeling. The quiet calculation of whether a task can wait until Monday, and the quiet relief when it can.
Wednesday has the feeling of being deep in it. Thursday has the feeling of seeing the weekend from a distance, like a coastline approaching. Tuesday is the most productive day, statistically, because it carries Monday's momentum without Monday's friction.
AI has no relationship with any of this.
It does not experience the caffeinated optimism of Monday. It does not feel the deceleration of Friday. It does not have the mid-week immersion of Wednesday or the horizon-scanning of Thursday. It does not distinguish between 9 AM and 3 AM. It does not get tired after lunch. It does not have a circadian rhythm.
And yet it does have a rhythm, of a sort. In March 2026, Anthropic doubled Claude's usage limits during off-peak hours — anytime outside of 8 AM to 2 PM Eastern on weekdays, and all day on weekends. The message was clear, if unintentional: the humans are all showing up at the same time, creating congestion, because humans organize their work around a social schedule that has nothing to do with optimal compute utilization.
The machines would prefer you work at 2 AM on a Sunday. The machines have more capacity then. The entire infrastructure of artificial intelligence is quietly suggesting that the workweek, as a concept, is a human inefficiency.
VII.
There is a well-documented phenomenon in creative and engineering work where you struggle with a problem, give up, go to sleep, and wake up with the solution.
Neuroscientists have theories about why this happens. Something about the default mode network, about the brain consolidating information during sleep, about unconscious processing that continues after conscious effort has stopped. The details matter less than the experience, which is universal: you cannot force insight. You can only prepare the conditions for it and then wait.
This is the rhythm that shareholders do not love.
The business does not want you to sleep on it. The business wants the solution now. The project timeline does not have a line item for "stare at ceiling, go for walk, sleep, wake up with answer." The standup meeting does not have a status category for "my unconscious mind is working on it."
And yet this rhythm produced the transistor, and the double helix, and general relativity, and "Yesterday" by the Beatles, which Paul McCartney famously dreamed in its entirety and woke up convinced he must have heard somewhere before because it came too easily.
AI does not need to sleep on it. AI does not need to struggle with a problem for three days before the solution crystallizes. AI produces answers immediately, without incubation, without frustration, without the specific human experience of earning an insight through suffering.
This is, from the perspective of productivity, an unambiguous improvement.
From the perspective of what it feels like to be a person who makes things, it is something else entirely.
VIII.
Before you conclude that this is an argument against progress, or a lament for a lost world, or a suggestion that we should slow down for the sake of human feelings, consider the grocery store.
You are buying groceries. You want the shelves to be stocked. You want the checkout to be efficient. You want the food to be good quality at a fair price. You want the parking lot to be plowed in winter and the carts to have four working wheels and the organic section to have that specific high-end yogurt your partner likes.
You do not, at any point, think about the warehouse worker's eureka moment. You do not consider whether the logistics coordinator felt a deep sense of satisfaction when she optimized the delivery routes. You do not care whether the supply chain involved human toil or algorithmic efficiency, as long as the yogurt is there.
This is not because you are callous. This is because you are a consumer, and consumers want outcomes, not processes. The entire economy is built on the principle that the customer does not care how the sausage gets made, as long as the sausage is good and the price is right.
So when AI compresses the work, when it eliminates the toil that preceded the eureka, it is doing exactly what consumers have always wanted. Faster, cheaper, better. The Saturday night in the studio, the three days debugging the race condition, the mastering engineer's trained ears — these are costs, from the consumer's perspective. Beautiful, human costs, but costs nonetheless.
And costs get optimized away. That is what costs do.
IX.
So what remains?
Not the toil itself. The toil was never the point, even if it felt like the point while you were in it. The point was the resolution. The moment of clicking into place. The eureka.
But the eureka required the toil. You cannot have the "we nailed it" without the hours of "we haven't nailed it yet." The satisfaction is proportional to the struggle. A bassline that took three minutes to write does not produce the same feeling as a bassline that took three days. Even if they are the same bassline. Even if the notes are identical.
This is irrational. This is also true.
We are going to have to adapt to a world where the struggle is optional. Where the eureka is available on demand, pre-packaged, instant. Where the bassline arrives fully formed and the mastering is a button press and the code writes itself and the project plan is not four weeks but four hours.
Some people will adapt by finding new struggles. Harder problems, deeper layers, domains where AI cannot yet reach. This has been the pattern with every previous wave of automation: the eliminated jobs are mourned, and then new jobs emerge that nobody anticipated, and those new jobs have their own satisfactions and their own 2 AM eureka moments.
Some people will adapt by redefining what they value. The handmade will become precious because it is handmade, not because it is better. The slow process will become a luxury, like the furniture maker in the Cowichan Valley who takes three weeks to build a table that a CNC machine could produce in an afternoon. You are not paying for the table. You are paying for the three weeks.
Some people will not adapt at all, and they will be angry, and their anger will be legitimate, because something real was taken from them. Not just a job, but a source of meaning. A way of spending a Saturday night that made them feel like they were contributing something that mattered.
The machines do not care about any of this. The machines have more bandwidth on weekends.
X.
I keep thinking about a musician in a studio at 2 AM. The song has been almost right for hours. The drums are good. The guitars are good. The vocal is good. But the bass is wrong, and everyone knows it, and nobody can figure out why.
They try it one way. They try it another way. The bass player puts down the instrument and listens to the track without playing, trying to hear what's missing. Someone suggests they change the key. Someone suggests they drop the bass out entirely for the first verse. Someone suggests they order pizza.
They order pizza.
They eat pizza. They talk about something else entirely. They come back to the song. The bass player picks up the instrument and plays something different from anything they've tried before, something that nobody suggested, something that came from the particular alchemy of frustration and pizza and 2 AM and the specific way the bass player's fingers happened to land on the fretboard.
And it works.
Everyone in the room knows immediately that it works. There is a moment of silence, and then someone says a word that is not printable in this essay, and then they play it again from the top, and it works again, and the room fills with something that is not quite joy and not quite relief but something in between that only happens when a creative problem resolves after extended resistance.
This moment is worth nothing to the market. It produces no additional revenue. The consumer who streams the song will not know it happened and would not pay more if they did. It is economically invisible.
It is also one of the best things about being alive.
I don't know how to reconcile these two facts. I don't think anyone does. But I think we should at least name what we're losing before we finish losing it.
