November 8, 2026. A day I was supposed to have off.
I had an erotic thriller sitting in front of me, and the cover looked like it knew exactly what kind of book it was selling. There was a woman on it, the lettering promised heat, and nothing about that cover suggested I was about to spend the opening chapter listening to a polite conversation in somebody’s nice house. I’d been through enough books by then to recognize the difference between a cover that makes somebody want to buy and an opening that makes somebody want to keep reading.
I put the chapter on and listened.
A cellist was having a conversation. The house was nice. Everybody seemed to know how to behave, and nobody was doing anything that made me need to hear what happened next. I kept listening because I was waiting for the story to arrive. I figured the surprise might be around the corner, or maybe somebody would reveal something that changed the meaning of the conversation.
Ten minutes went by. Nothing.
Another five minutes of people talking, and I was still waiting for the book to become the book the cover had promised. The writing wasn’t necessarily terrible. That was part of what made the problem so aggravating. The sentences were there. The people were talking. The setting was described. Everything looked like a chapter if you counted the words, but there wasn’t enough happening to make me care.
I stopped the recording.
“CHAPTER 1 HAS TO SELL THE BOOK. So far, I am 10 minutes in and there is no gotcha moment… it is just basic dialogue about nothing.” I wasn’t finished.
“You are not telling a story at all.”
Now, I want you to understand something about those two statements. They weren’t new instructions I had just invented because I was in a bad mood. I had been telling the machine that very thing for weeks, and I’d been careful enough to write the instruction down so I wouldn’t have to keep explaining myself.
Since October 14, I had a written record of what was going wrong with the openings of my books. That record already contained four examples of books that had broken the same rule. I had even written a description of the mistake so specific that it could have been a description of the chapter I was listening to.
“The cover sells a woman with cleavage on an erotic thriller and then chapter one is a conversation.”
There it was, in writing. And there I was, listening to it happen again.
Five books. Five versions of the same problem. I couldn’t blame the machine for failing to understand an instruction I’d never given it, because I had given it the instruction. I couldn’t claim I hadn’t noticed a pattern, because I’d written the pattern down. The evidence was sitting in front of me, complete with examples.
The part I had gotten wrong was believing that writing something down meant I had solved it.
I’d spent so much effort telling the machine what I wanted that I hadn’t built a reliable way to make it prove it had followed the instruction. I had a record of the problem. I had examples of the problem. I had a rule describing the problem. What I didn’t have was anything that would stop another chapter from being accepted when the problem happened again.
That was my mistake.
The machine could produce a perfectly neat chapter that broke the most important promise of the book. Unless somebody stopped it, that chapter could move forward simply because it looked finished. There was no check standing between a smooth piece of writing and my approval.
I was that check, sitting there with my headphones on. On my day off.
Let me give you the number that matters. Five.
By November 8, five different books had broken the same opening chapter rule. Four earlier books had already taught me what to watch for, and the fifth had just demonstrated that I hadn’t done enough with what I’d learned. These weren’t five unrelated mistakes that happened to come along while I was learning a new tool. They were five repetitions of a failure I’d already identified.
The immediate cost was one entire book opening written twice.