Joel Borgen had kids and a part-time job, and a novel he might have written eventually. What AI changed, he said, was the "activation energy" needed to get started. The result is The Receipt Horizon, a novel set in a world shaped by advanced AI that he wrote with AI models. On a weekly news episode of The Cognitive Revolution, published October 1, 2026, Borgen described what he brought to the book himself, where the models fell short, and why he thinks the human role may not last.
Host Nathan Labenz offered his own verdict in his introduction: the text shows a few AI tics, but the book is legitimately good.
The author draws up the plan
Borgen said he began the project "well over a year ago," knowing more or less what he wanted to make. He sketched the book himself, and the first half or so was "pretty well set" before he brought in any model.
The models are good at some things and getting better at everything, he said. Prose is the exception. "Even if you tell it exactly what you want and you have a plan for a chapter, you often get something that's unreadable unless you know how to scaffold it and then edit it yourself."
By scaffolding, Borgen meant the structure he hands the model. He usually writes out a large-scale architecture for the story and many of its beats, the small turns of plot inside each scene. Then he goes back and forth with the leading AI models, drafting chapter by chapter. He mostly used whichever Claude model was Anthropic's flagship at the time, along with OpenAI's ChatGPT Pro model from GPT-5 onward. He called ChatGPT Pro "the workhorse" because it did the most detailed work. Claude, he said, "used to be a lot lazier" about how much it would follow up on. He works mostly in the ordinary chat interface, which he said probably has both advantages and limitations for him.
Longer memory changed the work
In the early days, the biggest obstacle was context length: how much text a model can hold in view at once. While he worked on a chapter, earlier material kept dropping out of the conversation, and he had to manage that carefully. As context windows grew, he called it "a huge unlock": he could give a model the entire book. Each time a new model came out, he used it to try to improve the novel.
He said working alongside AI made the book "a lot better," but the process "also had weaknesses as well that I didn't entirely account for." He also expects AI writing to be controversial for many readers.
Editing out the over-explanation
Much of Borgen's craft came in revision. He cut the book by about 14%, mostly by hand, with AI helping him decide what to cut. A so-so chapter, he found, can often get better simply by removing what AI does especially badly: over-explanation.
He has since built his own tools for that work. They take his stated preferences, read the whole novel, and return an analysis and a list of things he might consider. Going through a manuscript manually is extremely time-consuming, he said, so the model acts like an assistant pointing things out while he decides "point by point." He called it "a fairly satisfying way to work."
When the draft was finished, Borgen took the advice to hire professionals for copy editing, art and layout. He said he learned a lot from them, but the work was "very expensive and time-consuming and kind of slow," and frustrating at times. He has tried to capture what he learned and fed it to ChatGPT's Pro model, building workflows for a follow-up book that he says is already being written quite differently as the models improve.
The book's AI and production note says AI took part in drafting and revision, and that Borgen remains responsible for the finished novel and credits himself with its story, world, characters, themes, structure and final prose choices. The note also says the audiobook was made with Labenz, using generated character voices with human direction, editing and review.
A utopia still needs conflict
Labenz asked how Borgen built a positive vision of the future into a story that still needs conflict. Borgen said utopias "often want to become dystopias in narrative form" because fiction needs friction, and it is dull when everything works out and everyone is happy.
His answer was to set the book after a major AI war, with humans largely locked into what they are allowed to do. Readers see "incredible flourishing" in technology and choices, along with "some really major downsides." According to the publisher's synopsis, the world is overseen by a superintelligence called the Steward, which maintains prosperity and safety while limiting what people can choose for themselves.
Borgen said he did not want any character to be "a mouthpiece" for him. Much of what he would like to see in the world appears in the story, but he also tried to steelman the people who would disagree, presenting their views in their strongest form.
Five notes on a viola
Borgen is also a musician. For well over a year, "maybe two years," he kept a private test for reasoning models: a sharp, high-resolution image of five notes written for viola. He asked each model for the clef, the notes, their lengths and the time signature. None got it right, he said, until OpenAI's Astra model. His guess is that OpenAI "presumably" added real musical scores to its training data.
Then he tried a complex 20th-century score, which he called "almost painful to look at," full of double sharps, accidentals and ties. Working through Codex, OpenAI's coding agent, the model did "an almost perfect job" of converting it into a machine-readable form. A score that had existed only as a PDF could now be edited and analyzed. "It went from kind of zero to 99 overnight," he said.
Borgen sees this as a way to turn old printed scores into material for training AI systems. He would like to see Suno, the AI music generator, add much more classical music. He argued that classical music requires a deeper grasp of form because it works at a much larger scale: a three-hour opera with internal structure, or a 20- or 30-minute single movement with far more going on than a pop song. Suno's Studio 2.0, announced August 13, 2026, centered on editing tools such as MIDI import and timeline editing.
Art made for one person
Borgen said he found the book worth writing because there was "a lot of me in it." He recalled hearing on a podcast about someone publishing mostly automated, AI-written romance novels on Amazon, a couple of hundred a year. As he remembered it, a few of them would make a few dollars, and the operation as a whole was profitable.
He expects models may soon produce customized work on demand that is as good as or better than the novel he spent many months on. That raised a question he said he cannot answer. As a viewer, he imagined being offered endless new films in the style of Ingmar Bergman or Stanley Kubrick, made for him alone. Culture is already fragmenting, he said, with less shared experience. "Is that art still valuable at that point, if it's not something that we're sharing with other humans?"
For now, Borgen thinks we are in "a kind of a sweet spot": making a work you are proud of with AI "still requires a great deal of human judgment and work." He also believes someone could already design a system with the right feedback loops to produce "a pretty decent book." Even for music made only for himself, he said he would want to work alongside a future model so that much of his own effort and taste went into it. He does not know how long that period of human-AI collaboration will last. Until it ends, he said, "we use it as a tool, but it still requires a lot of you and your judgment."