The prologue of my memoir is five sentences long. Two paragraph-sized fragments, separated by a row of dots. A family in a restaurant turns to look at us. I hold the woman’s eyes for a moment and look away. I wish I could tell them our story.

Then a section break. Then: That is the story I want to tell the family in the restaurant.

I had written that sentence a hundred times without noticing that the “that” had nothing to point at. Between the wish and the gesture toward it, I had put nothing. The middle beat — the compressed summary of everything the book holds — was simply absent. I had written around it for years.

It took an AI to say: the “that” reaches backward into empty space.

I spent an evening and night recently doing what I have started calling a reconciliation. My memoir draft in one window. A scaffold document — chapter structure, archival research, blog summaries, character notes, factual anchors — in another. I asked Claude to read both and tell me where they diverged.

What it found: structural duplications I had missed. A character narrated twice in the same chapter. An arranged marriage scene with nearly identical language in two different chapters. Objects that had drifted into the wrong place. Seven threads from my own archive that belonged in the book and weren’t there.

I am not describing this to tell you AI is useful for writers. I am describing it because of what came after.

Every gap the AI named required me to go back to my own life. The missing middle beat of the prologue: what does this book actually contain? The character narrated twice: which version earns its place and why? The threads from the archive: do you want this in the book, and if so, where does it hurt?

These are not structural questions. They are questions about what you know and what you are willing to say.

The AI can hold a manuscript and a research archive simultaneously and find where they diverge. It cannot know that the moment I was most afraid to write was the one about a cousin, a front yard, a mourning period, a skipping rope. It cannot know why that scene had been sitting in my archive for twenty years without finding its way into the book. It cannot know what it cost to write it, or what it would cost to leave it out.

What I found, working this way, is that the machine is very good at the problems I did not know I had. The duplications. The missing threads. The structural failures invisible to the person who has been inside the manuscript too long to see it fresh.

What it cannot do is supply the reason any of it matters.

There is a version of this essay that ends with a tidy observation about tools and writers and who is responsible for what. I am not going to write that essay. The honest version is more uncomfortable.

I have been writing this memoir for years. I have scrapped drafts, rebuilt structures, accumulated an archive that spans twenty years of blogging and email and handwritten letters. The book exists because of all of that. The machine read it in minutes and found the gaps.

I sat with that for a while.

The gaps are not a failure of the machine. They are a record of where I had been looking away. The prologue without its middle beat. The two accounts of the same woman on the same New Year’s Eve. The birthday I had noted in the archive and never put in the prose.

The machine noticed. I had to decide what to do with each one. It can only hold the mirror still long enough for you to see what you have been avoiding.

What you do with that seeing is yours alone.