Both launch-day runs of my book planner needed a redo
I built WriteMyBookAI, and both launch-day production runs failed my code's structure checks on their first attempt. Both passed after an automatic redo, taking 62.2 and 65.6 seconds overall. I got accepted results, and a retry hidden inside each roughly minute-long wait.
What I built
I launched WriteMyBookAI as a solo maker on October 4, 2026. It turns a paragraph about a book idea into a chapter-by-chapter plan with word targets, plus a first draft of chapter 1.
You can choose a book type (novel/fiction, nonfiction or a children's book) and a target length. The result has a title and subtitle, an intended audience, and concrete beats for each chapter. The opening draft gives you somewhere to start writing. It does not generate the whole book.
A real run
I entered:
A practical guide for first-time managers at small startups: the mistakes people make in their first 90 days, and what to do instead.
I picked Nonfiction and "about 30,000 words."
The result was titled "The First 90 Days: A Practical Guide for New Startup Managers," with the subtitle "Avoid the Most Common Mistakes and Build a Strong Foundation for Leadership." Its audience: "First-time managers at small startups, especially those promoted from within."
The plan had 10 chapters of 3,000 words each, adding up to exactly 30,000, with 4 beats per chapter. Chapter titles included "Mistake #1: Managing Like You're Still an Individual Contributor" and "Mistake #5: Trying to Change Everything at Once," so the outline stayed tied to the idea: mistakes a new manager makes, not a broad survey of leadership.
The opening draft, "The First 90 Days: Setting the Stage," was 696 words. Keep the two numbers apart when you read a result: 3,000 is the chapter's target in the finished book, 696 is the draft you actually get to start from.
I exported the result to Markdown and Word.
The checks before the result
An AI model accessed through OpenRouter writes the plan and the opening chapter. Before a plan is shown, my code checks it:
- 8 to 14 chapters.
- 3 to 6 beats per chapter.
- Chapter word targets add up to the book total within 5%.
- The chapter 1 draft is 600 to 900 words (at least 400 for a children's book).
- If the visitor picked a length, the book total is within 25% of it.
A plan that fails is regenerated once automatically. If the second one fails too, the visitor sees an error and the attempt is not counted.
These checks are narrow. They don't judge whether the writing is good. But they give the app a reason to reject an answer that only looks finished, and the result page says what was checked: "Structure checked: 10 chapters, 3–6 beats each, chapter targets add up to the total, chapter 1 is 696 words."
What the redo cost in time
The recorded managers-book run took 65.6 seconds of server time, redo included. The launch-day deploy smoke test took 62.2 seconds and returned 10 chapters with a 641-word chapter 1, also on the second attempt. My logs don't say which check failed first, so I won't guess.
Before launch I ran local probes for a cozy mystery, a guide for first-time startup managers and a children's chapter book. They returned 10, 8 and 10 chapters, with opening drafts of 784 to 804 words, and took 110 to 124 seconds each.
The checks kept the first outputs off the result page. The price was a longer wait while the app tried again.
Waiting about a minute, honestly
Your book plan and its opening chapter are generated in one request. While it runs, the page shows a progress bar, the elapsed seconds and a stage label such as "Outlining chapters and beats..."
With a redo in the middle, a visitor waits about a minute and only ever sees the accepted plan. The elapsed counter keeps that wait visible instead of pretending the result was instant. In the launch video I sped that part up and put the real 65.6 seconds on screen.
Taking the draft out
The result page has Markdown (.md) and Word (.docx) export buttons. Book ideas and plans are not stored in a database, so the export is your copy to keep working on.
Pricing and limits
The free allowance is 1 book plan per day, with no sign-up. Monthly is $9.90 per month for 60 book plans. The Pack is $4.90 once for 15 book plans, valid for 12 months. Payments go through Stripe, and failed attempts are not counted.
Every plan has the same scope: plan the book and draft its first chapter. It does not write the whole book.