A few months ago I blogged about a business plan as a joke. This month I’m writing about a real business plan. My first novel has just dropped on Amazon.

Somerset Maugham wrote in 1937: “There is an impression abroad that everyone has it in him to write one book… but if that means a good book, that impression is false.” What he’s really saying is that the difference between amateur and professional writers is craft, something arrived at through learning and practice, then maintained with sustained effort. The ever-pugnacious Christopher Hitchens improved on this, saying: “Everyone has a book inside them, which is exactly where… in most cases, it should remain.”

Well, I’m going to challenge that. I think lots of us have been struck by the muse and had a “golden nugget” idea that would make a great book. Perhaps, like me, you have a collection of such ideas jotted down in a notebook somewhere, or you even made a start but got no further than Roger Waters’ “…half a page of scribbled lines”. And while the craft of writing may be something we haven’t invested sufficient time to acquire, that doesn’t automatically mean the underlying idea is a bad one.

I’ve also noticed over the years that writing a book can become the last resort of middle-class folk who’ve fallen on hard times. They just rattle off a novel at the age of 50 or whatever, as if they’ve been doing it all their lives. How many of these are ghost-written? And what if you had your own ghost writer? When I first started toying with large language models a few years ago, I wondered if this might be the help I’d been looking for.

My early attempts to get AI to write anything decent were pretty fruitless. This was largely down to the small size of something called the “context window”, which you can think of as the AI’s attention span. If you asked it to write anything lengthy, it had to be done in small chunks, and there was no continuity between them.

As time went by the tools improved, and I beavered away learning how to get the best out of them. I built a suite of tools I jokingly refer to as “the fiction factory”, which gives me an AI-agnostic way to turn an initial idea into a book on my shelf.

This is a brief look at the three tricks these tools perform.

There’s a very old maxim in software: Garbage In, Garbage Out. The first trick is understanding that the initial specification must be detailed and precise. The first thing I do with my scribbled half-page is work with a chatbot, usually Claude (which has a particular knack for this sort of thing), to develop a book overview. This becomes the master plan: it describes the characters, maps the three acts, and sets goals and beats for each chapter. This is where I do the orchestral conducting, suggesting cadences and crescendos with instructions like “build tension here” or “shorten sentence length”. The better the plan, the better the draft. The output is a master overview file and a separate plan file for each chapter.

The second trick is getting the AI to write the thing without incurring a fortune in costs. Text generation is computationally intensive, and a standard twenty-euros-a-month ChatGPT subscription probably won’t cut it. This is where Openclaw came to the rescue. Acting as my ringmaster, it coordinates resources across the whole process. I have a dedicated PC running an open-source AI model locally. I’m always experimenting, but my current favourite is a Llama 70-billion-parameter model. This allows me to generate as much text as I like for free, well, for the cost of electricity, which runs to between five and ten euros a month (yes, I monitor it).

The third trick is accepting that AI cannot write a good novel without human intervention. My work in this area has evolved into a system built around an advanced editor that lets a human rapidly review the AI’s output, annotate it, and resubmit it for regeneration. The editor shows me one paragraph at a time, alongside the original “beat” instruction the AI used to generate it. I can either edit the text myself, or add a note telling the AI how to revise it on the next run. I’ve also built a lexical analysis tool that scans each draft looking for “echoes”: repeated words and phrases that AI tends to latch onto and overuse. Again, I can fix these manually or add them to a work-order that the AI uses to clean things up on the next run.

The first book to be published using this process has just been released, and it went through five drafts, five passes through the fiction-factory machine. The Mayfly Mutiny is a cautionary science fiction novel about the fate of an under-resourced Martian colony. The Kindle ebook is priced at about the cost of a cup of coffee, and the paperback is roughly what you’d pay here in Spain for a menú del día.

I’ve published the book under the pen name Maureen Avis. The idea is to build a brand, and my face probably wouldn’t be the best starting point for that. Two more sci-fi books are well underway, and I’ve got ideas for about a dozen more on the back burner. The business plan is to build a following by delivering regular, consistent quality storytelling: enough to elevate the brand above the AI slop that is inevitably coming to long-form fiction. While I was learning how to format the paperback version of Mayfly, I watched a YouTube video of a woman who claimed to have written a romance novel in three and a half hours using AI. She’s busy flooding the market with hundreds of similar tomes. Knowing what I know now, I’m confident those books won’t be very good.

In the world of investing, businesses need a "moat": a barrier to entry that slows competitors down. AI puts most business models at risk of being replicated overnight, but fiction has an unusual property: readers are the moat. A reader who trusts a name on a cover, who pre-orders the next book without reading the blurb, is not a customer easily stolen by a better-funded competitor or a faster algorithm. The race being run here isn't to build the best AI writing tool. That battle is already lost to the corporations. It's to build a relationship with an audience before the market floods. Somewhat ironically, AI authorship may be one of the rare business models where the technology is merely the means of production, and the thing it cannot replace, a consistent human voice and the trust that comes with it, turns out to be the most durable asset of all. Fingers crossed!

The Mayfly Mutiny is available as a Kindle eBook or paperback from your region's Amazon store.