I Wrote a Book With AI. Here's What Actually Happened

On AI, memoir, and the importance of humans who will tell you when it's bad

I want to tell you about a book I’ve been working on, and I want to be honest about how I finished it — because the how is going to raise some eyebrows, and I’d rather get ahead of that.

The short version: I used AI to help me write my memoir. The longer version is more interesting, and it’s the one I actually want to tell.

The rough material for this book has existed since around 2017. Blog posts, journal entries, stories from the mission field — two and a half years in the Netherlands, going on seventeen in Peru — stitched together into something that looked vaguely like a manuscript if you squinted.

I didn’t know what to do with it. It was clay. Not nothing, but also not a sculpture. So I set it down, and the years went by, and occasionally I would open the file and stare at it and then go do something else.

Earlier this year I picked it back up and decided to actually finish it. I fed what I had into Claude — Anthropic’s AI — and asked a simple question: what could I do with this?

Not “write it for me.”

Not “make it sound better.”

Just: here is this pile of material. What do you see?

What came back was an outline, a set of questions I hadn’t thought to ask myself, and a series of observations about where the gaps were — what was missing, what was repeated three times without me noticing, what threads were there but hadn’t been pulled. It was like having a very patient editorial assistant read your entire draft at two in the morning and actually pay attention.

From there, the process got interesting.

I have twenty plus years’ worth of emails. Newsletters to supporters, personal correspondence with mission leaders, messages I wrote to friends when things went sideways on the field — all of it unfiltered, all of it written when the events were fresh rather than reconstructed from memory two decades later once I decided I wanted to write about it.

I used Claude to search through that archive, to find the threads that corresponded to specific events I was writing about. What that gave me wasn’t AI writing my story. It was my own words, from the time, handed back to me. The difference between “I think I felt this way” and “I wrote about this to this person three days after it happened” is not a small difference. It’s the difference between memoir and mythology.

Then there were the voice recordings.

I would talk through a scene or a chapter — what happened, what I thought then, what I think now — record it, copy the transcript into Claude, and say: here’s the raw material. Help me shape it. What came back needed work. It always needs work. But it was shaped from my content, my voice, my specific embarrassing stories about mispronouncing Spanish words in front of a congregation in Lima or to our ministry school students in Chorrillos. Nobody else has those stories. AI didn’t invent them. It helped me get them out of my head and onto the page in a form that a human editor can actually do something with.

Here’s the part I want to be transparent about, because I think it matters

AI systems are trained in ways that create a pull toward approval. There’s a real risk that when you’re excited about a scene, Claude or your LLM of choice finds things to praise rather than things to cut. I did my best to build in guardrails — specific instructions about voice, about what the book was and wasn’t supposed to be, about flagging repetition and structural problems rather than just validating what I handed it. And to Claude’s credit, there were moments where it pushed back: identifying passages I’d repeated across multiple chapters, flagging content that belonged in editorial notes rather than the published book, holding the line on things I’d established as rules for the manuscript.

But I’m also self-aware enough to know that I can’t fully audit whether an AI has been straight with me or has been quietly buttering my bread the whole time. That’s not a reason to abandon the process. It’s a reason to not let the process end there.

This manuscript is going to a human editor. Then human proofreaders. Probably even beta readers before either of those stages. It does not go to market without human eye balls first. Not as a formality — as an actual check. A human editor whose professional reputation depends on honest feedback, and who has no particular incentive to make me feel good about two decades’ worth of missionary stories, will catch what Claude won’t. That’s the point.

The AI helped me get the draft across the finish line. The humans will tell me whether the draft is actually any good.

The book is called Just Your Average Revolutionary — working title, may or may not survive contact with the cover design process. It’s a missions memoir, which means it will find its natural audience among people who have been on the foreign mission field. But I’ve written it for anyone who organized their life around a vision that turned out to look nothing like the brochure. That’s a larger category than just missionaries.

It’s funny in places — at least I think it is. It’s honest about things that are genuinely hard to be honest about. It is emphatically not bitter, which I say not to be defensive about it but because the line between honest and bitter is one I care about, and I’ve worked to stay on the right side of it.

There are stories in here I’ve never told publicly — because the people involved were still my supervisors when I started writing, or because I wasn’t sure the world needed them, or because I hadn’t processed them enough to be fair about them. Some of that has changed.

The thing I was most surprised by in this process was how much having AI involved helped me see my own material clearly. When you’re knee-deep in your own writing, you lose the ability to read it. You know what you meant, so you stop seeing what’s actually on the page. Having something that could analyze the whole manuscript and say you’ve told this story in three different chapters, I’d suggest removing it from two of them — that’s not cheating. That’s editing. The underlying thought is still mine. The decision about what to do with the feedback is still mine. The embarrassing stories are definitely still mine.

What I ended up with is a manuscript I’m genuinely proud of, which is not something I expected to be able to say about this project after it sat dormant for the better part of a decade. Whether it’s actually as good as I think it is — that’s what the humans are for.

If you’re curious about the book, or about this process, or want to tell me I’ve completely sold out and it shows, the comments are open.

Just Your Average Revolutionary is a memoir about twenty years of being somewhere between a missionary and a person, with a lot of cultural misunderstandings, a Peruvian wife who is patient with me, and the slow realization that the revolution was real.

It just looked nothing like the brochure.