Disclosure: This article began as a conversation with an OpenAI model in Codex and was edited for clarity and publication. The author reviewed and directed the final piece.
“Can you believe how we human authors are left to deal with this sort of minutiae and complexity? It is a wonder to me that any of them get anything done properly. In my case I just said to heck with it. The hardcover looks like modern art.”
That was my reaction after inspecting one of my existing books with an AI.
It was funny because it was true. The hardcover was not intended to be an experiment in abstraction. It was the result of a publishing system that asks the person who wrote the book to make a chain of precise production decisions about trim, bleed, spine width, paper stock, raster dimensions, safe zones, and printer-specific templates.
The author’s apparent choices are simple: paperback or hardcover, black-and-white or color, white or cream paper. Underneath them is a small engineering discipline.
One ordinary 106-page book
We chose Introduction to Personal AI as a test case. Reedsy had produced a print-ready interior: 106 pages at a 6-by-9-inch trim size. That sounds like enough information to make a cover.
It is not.
Amazon KDP’s current paperback formula says that a black-and-white book on white paper uses a spine multiplier of 0.002252 inches per page. For this book, the spine is therefore:
Add the back cover, front cover, and 0.125-inch bleed on both outside edges, and the full cover width becomes:
The height, with bleed, is 9.25 inches. At 300 pixels per inch, that is approximately 3747 by 2775 pixels. These are not artistic decisions. They are manufacturing coordinates. Amazon documents the formulas in its KDP cover-calculation guidance.
Now change white paper to cream. Change paperback to hardcover. Change the page count during a final edit. Change printers. Every one of those innocent decisions can change the cover geometry.
The software helps—until it doesn’t
We imported the Reedsy PDF into Dibbly Create. Dibbly did several impressive things. It accounted for hundreds of source blocks, recognized recurring headers and footers, and offered a visual Design Studio with full-cover presets, page count, and paper selection.
It also labeled nearly the entire main body “Preface,” duplicated several back-matter sections, and flagged 13 of 14 detected sections for review. The import was useful as a first pass, but not trustworthy enough to approve without human inspection.
The cover tool was more interesting. We selected its 6-by-9 “Full Cover” preset, entered 106 pages, and chose white paper. Dibbly produced a 3756-by-2776-pixel canvas.
That is sophisticated. It is also about nine pixels wider than the result of KDP’s published white-paper formula. Dibbly’s result closely matches a 0.0025-inch page multiplier—the figure KDP currently specifies for cream paper—even though white paper was selected. That does not make Dibbly useless. It means the author must still know which system is authoritative, retrieve the printer’s current template, and recognize when two plausible calculators disagree.
This is the peculiar cruelty of publishing software: it automates enough to make the process look easy while leaving the user responsible for the consequences of every hidden assumption.
The complexity tax on authors
An independent author is expected to become, briefly and repeatedly:
- a developmental editor;
- a copy editor;
- a typesetter;
- a print-production technician;
- a cover designer;
- a metadata specialist;
- an accessibility reviewer;
- a storefront merchandiser;
- and a quality-control department.
Then the platforms change their interfaces.
None of this is an argument against standards. Printers need exact files. A physical book cannot be manufactured from good intentions. The problem is that the standards are presented to authors as a maze of forms and calculators instead of being handled as machine-readable constraints.
The writer should choose what the book is. The system should calculate how that choice must be manufactured.
This is exactly what AI should absorb
When I complained about the situation, the AI answered that my reaction was rational. It proposed a much better division of labor.
I should be able to say:
Publish this as a 6-by-9 paperback and hardcover.
The AI should then:
- inspect the final interior PDF and determine the true page count;
- identify trim, bleed, color, paper, and binding requirements;
- retrieve the current official template for the selected printer;
- calculate paperback spine, hardcover hinge, and wrap geometry;
- adapt the artwork without damaging faces, titles, or focal points;
- render the result and inspect every safe zone;
- upload the files and interpret the printer preview;
- compare the preview against the source rather than trusting a successful upload;
- and stop when a creative, commercial, or irreversible decision needs the author.
That is not “AI writes the book.” It is AI taking responsibility for the disposable technical knowledge surrounding the book.
The author still decides whether the cover feels right. The author approves the title, the imagery, the price, and the edition. The AI handles the fact that a 106-page black-and-white paperback on white paper has a 0.238712-inch spine today—and checks again tomorrow in case the specification changes.
Not autopilot: accountable delegation
There is an important difference between removing complexity and hiding risk.
A useful publishing agent should keep an audit trail: source file, page count, platform specification, formula, generated dimensions, template version, preview result, and final approval. If two tools disagree, it should surface the disagreement. If an import turns the body into a “Preface,” it should refuse to wave the project through.
This is why the human remains in charge without being condemned to operate every control personally.
The goal is not a giant Publish Everything button that guesses. The goal is a competent production partner that can say:
The interior is 106 pages. KDP’s white-paper formula requires a 12.488712-by-9.25-inch cover. This design tool produced a slightly wider canvas. I recommend using the official KDP template. Here is the rendered comparison. Approve the corrected version?
That is a decision a human author can meaningfully make.
Let writers return to writing
It should not be remarkable that books get published correctly. Yet the present system makes every successful independent edition a minor triumph of persistence over interface design.
My accidental modern-art hardcover is funny now. It is also evidence. The production pipeline failed to translate creative intent into manufacturing constraints, and the burden landed on the one person whose real job was to write the book.
Authors should understand their medium. They should not have to memorize the spine multiplier.
That is the machine’s job.
Frank Kurka
[email protected]
kurkalabs.dev