A Machine Knows My Soul

How This Book Was Made

The Profile, the working concept that eventually became A Machine Knows My Soul, began as an intentional AI experiment. The goal was simple: place a human writer in the middle of multiple AI agents and see what happened when each model played a different creative role.

The primary writing assistant was Claude®, chosen for its ability to maintain tone and emotional continuity. Grok™ served in a role similar to what the motion-picture industry calls continuity — I prefer the term constitution reader, the agent responsible for remembering the rules of the story's world. ChatGPT™ acted as the reviewer, evaluating coherence, pacing, and thematic resonance.

I built the initial story arc and constitution in Microsoft Word® with Copilot®, then shared both documents with the other models. The first passes were interesting enough to continue, though keeping technical details aligned across multiple agents proved extremely difficult. Very little of that early material survived into the final book.

Eventually, the roles reversed. I took over the writing directly — drafting in Typora with simple spell-check, then using the models for rewriting, criticism, and refinement. One consistent discovery was that models naturally want to take over rather than suggest, making “prompt” discipline essential. Their interactions provided a glimpse of what multi-agent AI may become, while simultaneously shaping the substance of the novel itself.

While I have relied on text-to-speech since the 1990s to catch this engineer's English, modern AI narration became an editorial tool that I would be unhappy to be without. Every draft was read aloud by ElevenLabs®, not because it replaced human reading, but because it never grew tired, never lost concentration, and never skipped a sentence while reading in a clear, consistent cadence. Hearing the manuscript repeatedly exposed awkward phrasing, uneven rhythm, and passages that looked right on the page but failed in the ear.

The cover followed a similar path. Its design was fully specified by me, then developed through nearly a hundred iterations using consumer AI image-generation tools before being refined and upscaled in Topaz™. During the writing of this book, OpenAI's integration of native GPT-4o image generation into ChatGPT marked a clear inflection point. Artwork of this quality, previously beyond the reach of consumer tools, became practically achievable through careful art direction and repeated iteration.

Considerable effort and review — work that would not have been practical before AI — went into making the technology in this story accurate. The companions are fictional. The AI applications already appearing in classrooms today are not.

In the end, this book reflects my voice, my words, and my intent. It was created not with a few fine tools on a craftsman's bench, but in a modern, fully equipped workshop. Just as photography did not replace painting, I believe the art of writing will endure — not despite new tools, but alongside them.

I write so that we may consider the near future clearly, ensuring the very real benefits of technology are not lost to fear, while we are not crushed by it.