I got duped into writing a novel. My native tongue is the essay, and my next book was going to be a playbook for "operating among psychopaths and psychic megafauna." I thought it would be best delivered with a parable, which I would then explain with essays. And I thought it would be cool if the parable were the encounter between Rajah Humabon and Ferdinand Magellan in 1521.
That story eventually demanded to be treated as its own thing, and not a mere palette for some personal development ideas. I said okay. I wrote the first chapter as a kind of prototype for the entire thing, made a deal with a published novelist to guide my novice ass (salamat, Mitya!), and plotted ten chapters.
As I wrote the succeeding chapters, new characters emerged, each demanding their own backstory and personality. I said okay. Now, it looks like the book will end up with 20-plus chapters, each of which will have 3,000 to 4,000 words. Dammit, I realized very recently, the thing wants to be a novel! I wouldn't have said yes to this book if it did not gradually ease me into the idea of writing a novel, as I had this self-image of an explainer more than a storyteller, and a scholar more than an artist.
Anyhow, the new angle I'm bringing to the story—Humabon as a psychopathic Filipino Cacique / Austronesian Big Man / Southeast Asian Orang Besar—came to me some time before the pandemic (his moves suddenly makes so much sense from this lens!). However, my first recorded writing pomodoro1 related to this book was in April of 2023. That's twenty months as of this writing. Since then, I've recorded 296 pomodoros related to the book. That's around 148 hours.
The first four chapters I wrote took two to six weeks each:
Here are the pomodoro counts for each of these chapters:
Conquistador's Gambit: 48
Red Harvest: 29
Boy Versus Serpent: 27
The Bendahara Versus the Baylan: 47
It would probably take me another 20 months to draft all the chapters if I continue at the same pace. Thankfully, LLMs have emerged in the past couple of years, and I'll probably complete the book (I hope!) in six months. I've been using ChatGPT in this book's research since it came out, replacing Google. I also upload highlights of scholarly works I've read to have a sense of this novel's world. This allows me to ask about them rather than scan/search them, saving a lot of time.
I don't recall exactly, but I must have tried letting ChatGPT write some sentences or paragraphs for the chapters. What I do recall was how cringe its output was, or at least not in the voice I wanted for the book. In the past two months, however, I've been experimenting with Claude.ai for this book, and it has been producing good enough paragraphs and sentences, occasionally superb, to be honest, or at least in a voice I like. This is mostly because of a feature called "Projects," which allows you to upload text that it can reference for your conversations. You can do this with ChatGPT for a single conversation. With Claude, you can have unlimited conversations with the same set of text. Claude can write in a voice consistent with the chapters I've written and within the world of the story, considering the text I uploaded: sample chapters and a couple of references (see image below).
Aside from uploading samples of other chapters, providing the first sentences or paragraphs and then asking Claude to complete them with specific instructions has produced the best results. Here’s an example. This screenshot is only around a third of the entire prompt, which contains the entirety of the completed part of this chapter.
The latest chapter only took me three days and 11 pomodoros to write. That's 336% faster than the 48 pomodoros it too me to write Chapter 1.
Building castles with a brick-making machine
There are two main ways Claude is helping me to write faster.
1) I spend less time making bricks.
Writing a story or an essay is like building a castle. In the past, I was the architect, the mason, and the guy who made bricks from scratch. I'd have a sense of the overall shape of the piece. Sometimes, the design is clear, and I just needed to arrange the bricks to fit the castle in my head. Sometimes, I needed to experiment brick-by-brick and explore my way towards a castle. In both cases, I had to create the bricks—words, sentences, and paragraphs—from scratch.
Now, I have a brick-making machine. It is so powerful that I can ask it to make ten bricks, discard nine of them, and use the one perfectly shaped for one section of my castle. ("Morning Pages," from The Artist's Way, is like a human-powered brick-making machine, at least in the way I used it: you vomit out 750 words in 30 minutes, some of which might be useful for your castles.)
I spend more time now being an architect and a finishing mason. I use the vision I was given and my ear to guide the brick-making machine's production and to select the right bricks and to position them. Many times, the bricks are not exactly how they are supposed to be (although I can get lucky—and I get lucky more often the better I become at operating the machine), so I need to make finishing touches.
2) I no longer get stuck with gaps in my train of thought.
My biological brain eventually gets tired and can no longer produce bricks to fill gaps in the castle. When this used to happen, I would just put a marker and return to it after a day or so. I'd then work on another part of the castle or take a walk or just call it a day. Now, I can just tell the machine to custom-make bricks for the gap. In fact, I can ask it to create several versions and I can select the one I like most.
I'm still kinda torn
A lot of the opposition I've seen to AI in writing involves 1) artistic integrity, 2) quality ("AI slop").
On #1: integrity here has two aspects: faithfulness to the vision given by the muse and honesty with the reader.
Let's be real. A muse (or at least this muse) does not actually care who actually manifests her in the world. She just wants to be made manifest. This particular one chose me because I'm her only choice. Who else sits in the middle of this Venn diagram?
Grew up in the area that Humabon ruled, shares his language and culture, found Pigafetta's account of him a bit strange, but then suddenly saw how Humabon's moves made so much sense when seen through the prism of his counterparts today
Studies anthropology and has written papers on ritual sacrifice and the figure of the Big Man / Orang Besar
Created a personality model (The Three Epistemologies) which is perfect for this story whose main character is a psychopath (see this model applied to the film The Departed)
(I'm not being a jerk here. Please don't leave me. I'll get this done, I promise. I'm committed to this. I've said no to many other muses. I'm just reminding myself that this book—which I want to read so much, to the point of writing it myself—probably won't get written unless I write it!)
This muse also does not care if the bricks come from my biological brain or from some robot's brain, as long as they are faithful to the vision of the castle she has revealed. The vision for this book, as far as I know, was only revealed to me. (Hey, I don't mind if you try to find other partners for this—I'm still gonna give my it my best, and I'm gonna do it better and faster than all of them. You see - I've spent so much time inside the mind of our megalomaniacal main character that I can effortlessly channel him!)
On honesty with the reader: I'll just mention that I used AI to help write this book (I have a plan on how to do this elegantly). Soon enough, acknowledging AI assistance will be as unremarkable as mentioning we used a word processor. A minority of books will be completely handmade, and it will say so on the cover. These will be valued the way we value handwritten letters.
On #2, AI slop: again, let's be honest. I'm no Nick Joaquin. Claude has produced bricks better than I probably would have for particular sections of the castle. However, you need someone to appreciate these bricks and pick them out of the pile. That someone then needs to know where exactly those bricks fit in the castle. That someone is me, not Nick Joaquin's dearly departed ass. AI slop happens with people who don't use the machine well and who don't use their taste to filter its output. Before AI, humans used to create slop. It's just easier to create slop now just as it is easier to create non-slop.
Still, I'm kinda torn about using AI. On one hand, I want to get this book done as best and as fast as possible. On the other, a part of me believes in the unique magic of words that come straight out of a writer’s soul. For instance, the bricks that made this post were all handmade. Did you feel it?
A pomodoro is a 25-minute work sprint.