Have you wanted something to exist so bad that you decided to make it happen yourself?
I just did that with my novel, Rajah Versus Conquistador. This is my fourth book, but the first one I'm looking forward to reading. By the time I finished my previous books, I was so fed up with all the rewriting that I wanted a break from the finished products after I shipped them.
This novel is different. I'm excited to read it. Why so? Here's my guess:
I wrote this to be enjoyed rather than to solve a problem. My previous books were non-fiction. They solved problems. I wrote them because I had experience solving the problem (eg, project management) or the writing itself solved the problem—ie, they restructured my mind. I thought the books would help others with the same problem—that's why I wrote them—but since I no longer had the problem after writing the book, there was no reason for me to read them.
I used AI to help me with the writing. I realized that the exhausting part of writing a novel isn't the imagining of a good story, but its translation into words on paper. It took me an average of four weeks to write the fully manual chapters, but only around a week for the AI-assisted ones.
It still took me 14 months and 682 pomodoros (~340 hours) to complete the entire thing (89k words). The language is also better than I could produce manually. It's not better out of the box. Still, because I could easily get it to produce multiple versions of entire sections, paragraphs, sentences, and down to that perfect single word, my job becomes more of a curator rather than a producer.
How much better? Let's find out.
Let me ask ChatGPT to rate first drafts of sample chapters. Here's the conversation.
Here are the results:
The manual chapters have an average score of 9.65 (I’m no Nick Joaquin, but I’m no slouch either), while the AI-assisted chapters have an average score of 9.95. This is surprising for me. I thought the difference would be higher.
The difference in the time it took to write these chapters is massive. The manual chapters took an average of 41 pomodoros or around 20.5 hours, while the AI-assisted chapters took an average of 7 pomodoros or around 3.5 hours. Based on this data, this novel would have taken around six years if I wrote it without AI!
We are still in the early days of using AI for writing works of fiction, so I thought of recording a video to share my process while it is still fresh in my mind. It is mainly about my tech stack:
Readwise for organizing highlights from my references
Roam Research for early-stage writing and work tracking. Here’s the Roam graph I show in the video.
Google docs for later-stage writing and getting feedback from my editor and alpha readers. Here’s the “novel map” I share in the video.
Claude for translating story outlines into nice-sounding chapters
ChatGPT for research and for developmental editing
Apple Notes for capturing messages from the diwata (muses)
I open the video by talking about my "neo-animist" approach to creative work as hinted by the last point above. I also talk about getting a professional novelist as a coach. Are all these cheating? I answer this question in the video.1
I experimented with using AI to edit the video. I got it to half of its original length, but it became quite jumpy. Unlike written text, the tradeoffs with using AI for video editing are still substantial at this point. Or this could simply be a consequence of my being much better at writing than at speaking.
Share this post