Fifteen books took me most of two decades. Sixty-two more showed up in about four months, and I didn't write a single one of them.
That's not a trick sentence. The fifteen are mine, in my name, in my voice: seven on professional performance, five children's books about three real dogs, three that don't fit anywhere tidy. Every one of them cost me the normal price of a book, which is a long time and a lot of rewriting. The other sixty-two books aren't mine at all. They're large-print puzzle books, built by an agent I designed, published under a different name, for a reader my own catalog doesn't reach: people over sixty who want a word search they can actually see.
Why build an agent instead of just writing more books
I didn't start this because I ran out of things to say in my own voice. I started it because I noticed a category with real demand and zero automation anywhere in it. Puzzle books for seniors and nostalgic readers are a real, steady market, and nobody had built a pipeline for them the way software gets built for everything else. Every other corner of publishing I touch, I'm the one doing the writing. This corner, I wanted to see if the same discipline I use to build software, Lean Six Sigma process thinking applied to content instead of manufacturing, could stand up a catalog without me writing a single clue.
So the agent does the thing I'd otherwise have to do by hand: pick a theme, lay out a puzzle format, generate the content, run it through a quality judge, and package it into something ready for KDP. I touch the system that makes the puzzles instead of touching the puzzles themselves, the same way I touch the systems behind my apps instead of hand-coding every button in them.
The part that actually worked
Sixty-two titles went through the judge. All sixty-two passed. Twenty are published, across four series, which is the part of this I'm proudest of, not because the number is huge but because the quality bar held at that volume. A system that only works for the first five titles and falls apart at fifty is a lucky streak with a few extra steps, and this one held instead.
The part that didn't
Knowing a book is good and knowing a book will sell are two different problems, and I built the measurement half of this badly the first time. The table that was supposed to track sales sat empty for the first wave of titles, not because nothing sold but because the data pipeline feeding it hadn't started reporting yet. My kill-switch logic couldn't tell an empty row from a true zero, which meant titles with simply no data yet were about to get labeled failures and shut down. I caught it before anything actually fired, fixed the logic to tell the difference between no data and zero, and reran the whole check clean. Twenty titles came back correctly flagged as not yet published. Twenty-two came back correctly flagged as too early to judge. None of them got killed for a bug in how I was watching them.
One more fix, smaller but real
The early titles also stepped on each other's keywords, four of them aimed at nearly the same search terms, splitting traffic that should have gone to one listing instead of three. Wave 2 gave every title its own lane: one aimed at a gift for mom, one at the holidays, one at someone recovering from something, one at a retiree picking up a new hobby. Zero keyword overlap now across all twenty published titles. It's the kind of fix you only find by actually reading twenty listings side by side instead of trusting that four different themes meant four different audiences.
What this has to do with my own fifteen books
It has nothing to do with them directly, and that's sort of the point. Dorothy Fennimore, the name on the puzzle books, isn't a pen name for me hiding in plain sight. She's a separate catalog with a separate reader, and the two never share a shelf. What they share is the same belief I've carried through every book I've written under my own name: most of what looks like a quality problem is actually a process problem wearing a disguise. Fix the system, and the output takes care of itself far more often than people expect.
What I'd tell someone trying this
Build the judge before you build the volume. It's tempting to generate first and sort later, but a pipeline that produces sixty-two titles before anyone checks a single one is a pipeline that can produce sixty-two bad ones just as easily. The judge has to exist before the scale does. Watch your own instrumentation as closely as you watch the product, too. The bug that almost cost me thirteen titles lived in the dashboard that was supposed to tell me the truth and quietly didn't, not in a single puzzle anywhere in the catalog.