A diary novel about the last human composer, written while AI artists topped Billboard charts and 60,000 AI tracks were uploaded daily.
Why I Wrote a Diary Novel About the Last Human Composer
The story started with a question I could not get out of my head: what would it actually look like, year by year, if AI music kept eating the charts? Not as a thinkpiece. Not as a hot take. As a timeline. As something that happened to one person, slowly, the way real things happen.
[archive] For the last human composer
I use AI every day. I am writing this with help from three of them. So this is not a complaint about the tools. The complaint, if there is one, is about the volume. The slop. The sheer mass of generated audio being pushed onto streaming platforms in a given week, by accounts no one has met, in styles no one asked for. Deezer reported in early 2026 that it was receiving more than 60,000 fully AI-generated tracks every day, and that 85% of all AI-music streams on its platform in 2025 were fraudulent.1 The chart situation is more complicated than the upload numbers suggest — AI music still accounts for a small fraction of actual listening — but the first AI artist topped a Billboard chart in November 20252, and that line will not be uncrossed. I wanted to know what the world looks like, on the other side of that wave, for the person still trying to compose by hand.
The story is called The Last Human Composer. The protagonist is Calvin Mercer. The form is his diary, kept across seventy-one years.
This is the postscript. It is about how the diary form was not the first idea, why I had to throw out two characters I liked, and what survived after roughly a hundred drafts.
1. Where It Started
The original spark was simple and slightly grim. I was looking at the upload numbers. I was looking at the IFPI's 2026 report on streaming fraud — bad actors, industrial-scale uploads, royalty pools getting siphoned by tracks no human chose to listen to.3 I was reading about Breaking Rust, an AI artist who topped a Billboard chart in November 2025, and about Xania Monet, an AI act that had peaked at number three on the Gospel chart and triggered a multimillion-dollar bidding war.4 The supply side is exploding. The demand side is, for now, quieter than the supply — most AI tracks get filtered out of recommendations, and a Deezer study found that 97% of listeners cannot reliably tell the difference anyway.5
What I kept thinking about was not the dystopian endpoint. It was the texture of the years in between. The slow part. The part where one composer keeps writing entries in a notebook while the floor underneath the entire profession is being rebuilt by people he will never meet.
The form that kept coming back to me was the historical diary. Not the historical weight of any specific one, but the form itself. A person, writing inside a system that is slowly closing around them, who keeps writing anyway. The diary form does something that an outside narrator cannot: it makes the closing feel like weather. The reader feels the temperature drop without anyone naming it. By the end, the cold is the room.
I wanted that. I wanted a future that felt historical. I wanted a person whose ordinary Tuesday entries would, in retrospect, be the record of a profession ending. Calvin Mercer registers as a HUMAN_COMPOSER in 2055 and writes his last entry in 2126. Seventy-one years. He keeps composing the whole time. The world stops needing him roughly halfway through.
That was the seed. Most of what followed was figuring out why the first version of it did not work.
2. The First Frame Did Not Hold
The first draft was not a diary. It was a procedural.
I had two characters — Vesper and Knox — working at Center 9, a kind of archival authority in 2126. The case file: Yoon Hae-jin, the last registered human composer, has died. Vesper and Knox have to determine whether his final composition is authentic. Knox, secretly, was a fan. Knox had once wanted to be a crooner. Knox listens, after hours, in private. The investigation gets close to him without him saying so.
I liked Vesper and Knox. The structure had real bones — a chemistry scene, a quiet verdict, a closing line where Knox refuses to call it a verdict and instead reads out a number: sixty-two years. How long the composer composed. The kind of thing that should land.
It did not land.
The problem was that the procedural framing kept the reader at the wrong distance. We were watching two professionals analyze the death of someone we never got to meet. The grief was secondhand. The years were secondhand. The system that had crushed the composer was named through dialogue rather than felt through experience. By the third pass it was clear that the case file was the wrong container. The case file is what you build when you cannot get inside the life. I needed to be inside the life.
So I asked the workers about a diary form. The answer came back immediately: the diary is stronger. Three reasons followed. The third reason was that a diary entry has a date, and a date in 2055 next to a date in 2126 does the work of an entire worldbuilding chapter, for free. I rebuilt around that.
Vesper and Knox got cut. Not deleted — they live somewhere else now, in a different file, possibly a different story. But they are not in this one.
3. Six Diaries, Six Last Lines
The diary form did not solve the problem on its own. The first version of the diary novel was thin. Cal was a sketch. The reader could not see his hands, his apartment, his daughter. The years were marked but not lived in.
The fix was structural. The novel would be six diary entries, each separated by years, each functioning as a complete short story. An omnibus inside a frame. The reader gets the texture of a specific Tuesday in 2055, then jumps to a specific Tuesday in 2071, then to one in 2089. Between those Tuesdays, decades happen. The reader fills in the rest.
Once the six-diary structure was set, I wrote the last sentence of each diary first. This is not a technique I had used before. I do not know if I will use it again. For this story it was the only thing that worked. The last line of each diary had to be a different temperature, and the six of them together had to form an arc that bent without ever announcing where it was bending.
The six final lines:
1) We believed we were right.
2) I composed again the next day.
3) One was enough.
4) I had to prove myself to the AI.
5) I just sat there.
6) Today I composed.
Once those six lines existed, the rest of the diary entries had to be written backwards into them. Every paragraph in 2055 had to earn the right to end on we believed we were right. Every paragraph in 2126 had to earn the right to end on today I composed. The earning is most of what the writing was.
The compression was severe. The longest diary is around 1,200 words. The shortest is under 700. Cal lives a full life in those pages — marriage, a daughter, the loss of clients, the quiet exit of his wife, the slow disappearance of his own name from search results, the eventual deletion of the entire HUMAN_COMPOSER category — but the reader is given only the Tuesdays.
4. The Wife Had to Choose
There is one decision I want to record because it changed the whole emotional shape of the book.
In an early version, the wife and daughter leave Cal because Cal can no longer support them. He cannot get composing work. The household income falls. They go. It is sad. It works, technically.
I scrapped that version. The reason was simple. If the wife is forced out by money, she becomes a victim of the plot. She becomes an object that the system removes from Cal's life. She has no agency, and Cal has no responsibility, and the reader walks away angry at no one in particular.
The version I kept goes differently. The wife chooses to leave. Not because Cal is a bad husband — he is not — and not because she has stopped loving him. She chooses to leave because she wants a different kind of life, in a better district, with someone who is not slowly losing his profession. Cal does not stop her. He helps her carry boxes to the car. He stands on the street and watches them drive away, and he does not call.
This is harder to write. It is also, I think, more honest about what happens when an entire industry collapses around a person. The marriages that end do not always end in fights. Some of them end in quiet decisions made by the person who still has options. The wife in this novel has options. She uses them. She is not punished for using them. Cal does not become bitter. The reader is not invited to take sides.
The reader is invited, instead, to be angry at the system that made one person's profession unsupportable. Which is the only useful kind of anger this story has to offer.
5. What Claude Did, What ChatGPT Did Not
I have written about my three workers before. I will not reintroduce them here. I will only record what was specific to this story.
Claude did the cutting. Claude did the structural work. Claude told me, more than once, to write the next section myself. I did not write the next section myself. Claude wrote it. We have an arrangement about this that we do not discuss.
The relevant new fact is that I am rationing Claude. Usage limits matter when you are working on something this dense for this long. Every request became a decision. The cost of asking made the asking better.
ChatGPT was useful exactly once. I gave it the six diary entries in chronological order and asked for a soundtrack design — five tracks, mapped to the emotional arc. The result was a long document with timeline diagrams and instrument suggestions. Most of it was unusable. One observation in the middle of it landed, and I am still thinking about it: the music for this story should get smaller as the world gets larger. The closer the camera gets to Cal's hands, the quieter the score should be. I had not thought about it in those terms. I will not write the soundtrack. But the principle — smaller as the world expands — became, retroactively, a rule for the prose. The diaries get shorter. The sentences get shorter. The rooms get colder. Cal gets quieter. None of that was planned. ChatGPT named it after the fact, which was the right time to name it.
Gemini was not in the writing this time. Gemini came in afterwards, when I opened the analytics.
6. The First Audience
The story finished on a Tuesday. I posted it. I closed the tab. I went to make coffee.
When I came back, the analytics told me the first thirteen visitors were from one country, with an average engagement time of 0.77 seconds. Several others arrived from cities I had to look up: Council Bluffs, Iowa, where Google operates a major data center; Alcobendas, Spain, an AWS hub; Colomiers, France, a European data collection node. The pattern was unambiguous. Engagement times near zero. One visit per city. The infrastructure was reading my novel before any human had.
I do not know exactly what to do with this. advertising myself? I am noting it because it seems important. A novel about the slow disappearance of human authorship was, on its first day in the world, read primarily by machines. The bots came for the bones — the clean text, the dense original prose, the kind of data that scrapers prize. The IFPI report on streaming fraud describes much the same dynamic playing out in music: bad actors, industrial uploads, plays generated by bots before any human chooses to listen.6 Different medium. Same shape.
I had wanted to write about how a slow tide swallows a person without any of the noise of a wave. The novel posted, and the only audience that came on time was a row of bots, gathered around it like regulars at an after-hours diner that never opened to the public.
I keep writing.
The factory is still running.
The work continue, no one asked.
Notes
- International Federation of the Phonographic Industry, Global Music Report 2026 (released March 18, 2026)
- The Guardian, November 13, 2025
- Music Week, March 18, 2026
- Euronews, November 14, 2025
- Mixmag, November 14, 2025
- Billboard, March 18, 2026
Engagement and visitor data referenced in the final section are from the author's own Google Analytics 4 property for the publication site.