The Book I'm Writing With AI (and why it's harder than I expected)

00:00 — Opening

Hey, welcome back to Slow Builds.
I always say this one's going to be a little different — but they're all different I suppose.
What I will say is that as I get closer to finishing the book, I'm getting a little more free, a little more animated. My thoughts are starting to get clearer.
So this one is about the book — where I'm at, what I've been using, and honestly, what's been frustrating about it.

00:20 — Why the book exists

I've had this idea for a long time.
Not really "I want to write a book"… but more like I've been collecting things.
Advice, patterns, small realizations — stuff I wish I understood earlier.
Things that would've made life a little easier if someone had just taken the time to sit me down and point them out. Not "do this, do that" — just little nudges. Help avoiding pitfalls.
And over time, that turned into… maybe this should be something I write down properly.

01:10 — What the book is

It's not a story.
It's more like — if I had to hand something to my kids one day, or a younger version of myself — what would actually be useful?
Not a blueprint. That word's wrong.
More like tidbits. Small, specific things from a life lived — I'm probably closer to halfway through at this point.
Practical ways to think about:

• money
• pressure
• decisions
• responsibility
• family

Nothing motivational. No big speeches. Just stuff that would help them feel like — hey, other people have been here too. You're not alone in this.

02:00 — Starting with AI

So when I finally started putting it together, AI felt like the obvious place to start.
Because the hard part wasn't ideas. Ideas are life — they just happen.
It was:

• organizing them
• structuring them
• turning them into something readable

So I started using ChatGPT. It was working really well — helping me get thoughts down, fill in gaps, organize things. It was starting to look like a book.

03:00 — Tool hopping

Then I wanted more. I wanted infographics, slide decks, a different take on what I was building. So I moved into NotebookLM.
Which is good for structure — outlines, connecting ideas, building a framework.
But I started noticing something. Things weren't lining up. It wouldn't update properly. I'd paste something in, it would cut stuff out, or summarize when I needed the full thing. I'd overwrite the old version and lose content.
It started to feel like I wasn't working on the book anymore. I was just trying to maintain versions and mash things together. It took away from my words — and I really didn't like that.

04:15 — The rewrite reality check

So I got it back to a good place — broke things out into manuscripts, chapter by chapter.
Then I threw everything into Claude. Full tilt. I gave it a very specific prompt: you're an editor who hasn't had a hit book in a while. You need something on the shelves in three months. Your ass is on the line.
And that was a bit of a reality check.
Because it went deep. It found a ton of repetition. Filler. Things that didn't line up. And I thought more content meant progress — but it was mostly just noise.

05:30 — Cutting it down

So I started trimming.
The book got better… but smaller. Which feels weird — like you're losing work — even though you're actually improving it.
I hate books that fill space for the sake of it. Books full of pulled quotes just to pad chapters. Mine does have a quote at the start of each chapter and a short reference to a related book — but it's precise. It sets the tone. It earns its place.

06:20 — Automating the process

Then I went further.
I have a server running open-source Claude, and I built a system using Obsidian and GitHub. The book is in Markdown. I can push, pull, commit — version control the whole thing.
Then I built agents. The idea: I drop in ideas and feedback, and every night the agents work on revisions. They weave in new ideas, work on chapters, do rewrites. Every morning I'd wake up to a progress report explaining what happened overnight.
It felt really good. Like the book was continuously being worked on even when I wasn't there. All I had to do was focus on what I wanted in it — answer questions about timelines, stories, how things fit together.

07:20 — Where it starts to break

And then things started to feel off.
I went back through everything. Checked it in ChatGPT — "great job, you're doing awesome." Checked it in NotebookLM — wanted to tighten things up but gave specifics. Then went back to Claude.
And Claude found a lot of problems.

07:40 — Subtle drift

It wasn't obvious at first.
The agents were filling in gaps, smoothing things out — and suddenly I'd have something that sounded right… but wasn't true.
Stories were kind of in line, but digging deeper: that timeline's off. That didn't happen that way. It mixed up whose wedding I was at, who my business partner was, what client I was visiting. It didn't understand that you can have multiple businesses running at the same time with different people.

08:20 — Made-up clarity

And that's a strange feeling. Because the stories were believable.
But they weren't what actually happened.
And that's scary. If you start trusting AI enough to let it do that — and you don't check — imagine what happens to news articles, non-fiction books, history books. It's not even deliberate. It's just AI wanting the story to feel right, filling in what's missing, smoothing out what doesn't fit.
You can see how a whole drift in society could happen that way.
So now I'm going back through my own book trying to figure out what's real, what changed, and what got quietly invented.

09:00 — More work, not less

And this is the part people don't see when they say "AI wrote it for you."
It didn't.
If anything, it created more work. I have to:

• verify everything
• fix tone
• correct intent
• remove things that aren't mine
• find those things — which is hard in a 155-page book

09:50 — Slowing the process down

So what I do now is slower.
I take sections into Claude and instead of asking it to rewrite, I have it ask me questions first.
What actually happened? What did I mean? What's the point of this?
I answer all of that before anything gets touched.
At one point I went through an hour of back and forth — Claude asking questions, me giving answers. Then I asked: how many chapters do we have left, what's the page count, where are we at?
It said: "I haven't written anything yet. I'm just collecting the information. I'll write when you tell me to."
An hour of answers, and nothing was written down. If I hadn't asked, I would have lost the whole session and had to start over.
So now every time I want to stop, I explicitly tell it: write that in, commit it, push it. Make sure it's saved. Every single time.

10:40 — Still friction

But even that has friction.
It collects everything and waits. Which I understand — I'm the decision maker. But if I answer a question, I'd expect the answer to just be written down. Not stored in memory waiting for me to say "now go write it." It creates overhead. It makes me more aware of the process than the book.

11:10 — The loop

So now I'm in this loop where I'm not just writing the book…
I'm trying to:

• hold onto everything
• keep it consistent
• make sure nothing disappears

And going back through the same loop every session — rechecking, restating, re-establishing context — takes more time than writing.

11:40 — The tradeoff

And I think that's really the tradeoff with AI once you start jumping between tools, models, and workflows.
You're not just creating anymore. You're managing pieces. It starts to feel like a chore rather than sitting at a typewriter and just going.
Things get lost. Slight changes. Inconsistencies. Timelines drift.

12:20 — But for this, it works

But for the book — it's actually useful.
Because it forces me back into it. I can't just accept what it gives me. I have to reread everything, question it, restate it.
And that process is what makes it mine.

13:10 — What AI really is

At this point, AI is basically just my note taker and timeline keeper.
It helps me collect ideas, reflect them back, organize them.
But every important decision, I still have to make. There's no way around it.
I might put "co-written with AI" somewhere in the book — because it's the force that helped me start. I'll give credit where it's due.
But there is no world in which this book wasn't written by me. These are my truths. My words. My life. No one can tell me otherwise.

13:40 — The real lesson

The book sat in my head for years.
And the real problem wasn't writing it — it was starting.
AI gave me a place to start. Not a perfect version. Just something.
And once something exists, you can work with it.

14:40 — Closing

So yeah, it's messy. Back and forth. Sometimes it feels like things are getting lost.
But that's still better than never starting at all.
I'm about three months out from wanting to have some version of this out — at least on Kindle. Something that can breathe. Something my kids can actually get to.
I'm excited for this one.

15:00 — Final line

AI isn't writing the book.
It just made me sit with it long enough to actually work on it.