The Hollowness of AI

My kids always want to talk to me about anime. So many stories start with a character who's completely overpowered. They arrive in a new world, already invincible, and more often than not, they just succeed at everything they do. The best shows that I've seen, Mob Psycho 100, One Punch Man, flip that somehow. What if a character starts with all the power and then has to figure out how to use the power to make the world better before it's gone?

The more I thought about it, the more I saw in it. Aging. Gentrification. The generational tension between boomers who had everything and millennials rebuilding from what's left. It was a premise with real weight.

I woke up at 5am with the whole thing buzzing in my head, too excited to go back to sleep. Normally, this is where I move on with my day. I studied English in college, focused on creative writing, and that's where I learned that I can recognize good writing, and I don't want to do the work of becoming a good writer. That gap between taste and craft has kept every novel idea safely in the "wouldn't that be cool" category for twenty-plus years.

This time, I didn't let it go. Not because I'd suddenly developed the discipline to write a novel, but because I'd been building AI systems. Agentic pipelines where multiple AIs collaborate, check each other, revise each other's work. The most important technology of the last decade was also, maybe, the bridge across the widest creative gap I'd never been willing to cross.

So I decided to build a machine to write my novel.

Before the novel, there was a different experiment. Bear with me, this is important.

During the pandemic, Midjourney was new and I wanted to see what it could do. I knew I wanted an oil painting in the style of Waterhouse—classical, rich, layered. Midjourney understood the general shape of what I wanted, but it had no ability to tell right from wrong in the details.

I asked for a picture of a pirate ship. It complied, and from across the room, it is obviously, beautifully, a pirate ship on a brilliant teal and orange sunset ocean. Look closer and nothing holds up. There are too many masts and too many sails. The rigging is random and haphazard, spars attached in places that make no structural sense. No individual part of the painting is displaying anything that actually resembles a pirate ship. But, the whole is gorgeous.

It took over fifty iterations of prompt revision to get there, and then I spent hours in Photoshop pulling it the rest of the way. It hangs on my wall now, and I love it. Not despite the wrongness, but because of it. The details being off actually adds to the conversation about the piece.

One of my favorite works of art I've ever seen is a painting with the words "you and I are so good together." I've had deep conversations about who "you and I" refers to. The artist and the art? The artist and the viewer? The piece itself is simple. The conversation after the viewing is the value.

The pirate ship does the same thing. It invites you to look, then look again, then talk about what you're seeing. There is a collaboration between me, the AI, and the viewer in that painting, and because of that, it doesn't feel hollow. It feels like mine.

The novel was a different kind of experiment. Nanowrimo asks for 2-5 hours of daily writing for 30 days. I have no interest in that kind of time investment. With the novel, I wanted to see what I could accomplish with agentic development.

I used Claude as a writing coach to understand what the story could be, then I built the system to write it without me. Thirteen AI agents, each with a specific role. Some guarded the plot. Some maintained the setting and tone. Three separate writers produced different versions of each chapter, sparse, lyrical, and warm, and an editor agent chose the best paragraphs from each. A final reviewer asked every agent to critique instead of create. The whole pipeline ran on its own.

It wrote a 22-chapter book—the first in a trilogy. The first draft was better than most first drafts.

I started to read it.

The prose needs work. The character voices blend together. There's too much telling and not enough showing. But those are craft problems, fixable problems. The thing that stopped me wasn't the quality.

When I read a book by a person, I feel them behind the words. I sense their choices, their obsessions, the way they linger on certain moments because those moments mattered to them. I'm not just reading a story. I'm in a silent relationship with whoever wrote it, and that connection is a huge part of why reading matters.

The Midjourney painting invites you to look closer, to talk about what you see, to find the artist in the imperfections. The novel doesn't invite anything. It performs competence and then sits there, waiting for nothing, because it has nothing to wait for.

I've read most of the book. The problem isn't that I don't want to finish it. The problem is that it needs work, and I'm wondering if the work I put in will ever make it worth it. At what point does my revision, my shaping, my voice pushed into the gaps, turn it into something that has a person behind it? Or, does the foundation being hollow mean the whole thing stays hollow no matter what I do?

And it's not just the novel anymore. I'm finding it difficult to appreciate any work of art that doesn't have a person behind it. AI-generated songs, AI videos on social media, short-form content that looks like a performance of people or animals or nature but is just generated noise. Once you feel the absence, it's hard to unfeel it.

I keep coming back to the history of tools and art. Digital art wasn't "real art" until it was normalized. Digital replaced film. Blogs replaced newsletters. Kindles replaced books. Every time, people said the new tool cheapened the work, and maybe it did. Ultimately, the tool didn't matter. The person using it did.

So, maybe this is just the next version of that argument, and I'm on the wrong side of it, an old man yelling at a cloud. Maybe AI is just another tool, and the discomfort I'm feeling is the same discomfort every generation feels when the means of creation shifts beneath them.

But, I don't think that's quite right. A Wacom tablet is a tool. An iPad is a tool. Photoshop is a tool. I used new tools on the painting, and it feels like mine.

AI generates. And the question I can't get past is: how much of myself do I need to put into the output before it becomes mine instead of an echo of the souls of many people who were stolen to make it?

This problem isn't going away. The tools will only get better. The output will only get more convincing. Eventually, most of our songs, our TV shows, our movies, and our books will likely be written by AI. Much of it will be technically better than what many people produce today.

I've now felt the cost firsthand. I built a sophisticated system, fed it a premise I genuinely cared about, and got back something competent that I can't connect to. The craft problems are solvable. The emptiness isn't, at least not without putting more of myself back in, or leveraging the ignorance that AI owned the prose.

I look at the pirate ship on my wall and I see collaboration. I see fifty rounds of prompting and hours of Photoshop and a conversation that's still happening every time someone walks up to it and notices the rigging is wrong. I look at the novel and I see a question I don't have an answer to yet.

The real danger isn't that AI will replace human creativity. It's that we'll get comfortable with the replacement and stop noticing what's missing. We'll optimize for output and forget that the whole point of art was the person on the other end.