In one sense, I started my novel in November of 2017, eight and a half years ago. Since then it has gone through six drafts, more than 200,000 words, and now multiple years of functional idleness as I try to spark some of that old energy. I had trained myself to be a romance novelist so well that writing any other genre had become impossible. This was a problem, given that I wanted to write a science fiction novel where my characters’ romantic feelings were coerced, oppressive, and not consummated in a romance-friendly way.

To pull off this vision, it became necessary to stop writing this novel and any other romantic fiction for multiple years. I also had to read other genres almost exclusively. Over time, the goal became not to overhaul the novel, but to overhaul everything to do with my writing process and instincts.

Actually applying this brand new writing approach—one I have little experience with—is turning out to be challenging. I am taking worldbuilding extremely seriously, to the point of having three little curricula set up for myself. I have a science curriculum (Entangled Life, Merlin Sheldrake; Deep Water, James Bradley), a social curriculum (The Warmth of Other Suns, Isabel Wilkerson; Black Reconstruction in America, W.E.B. Dubois), and a fiction curriculum (Red Mars, Kim Stanley Robinson; Neuromancer, William Gibson). I fear this is the retired scholar in me rearing its little head, trying to convince me that there is a minimum of two years of reading required before the writing can begin. I do not think this is true, but I do trust the instinct that makes itself known every time I sit down to outline my novel that says, “I don’t know enough to do this yet.”

Initially I thought that annoying little nudge was procrastination. It is very easy to avoid writing, and almost as easy to invent reasons why the writing cannot be done. But as the same pattern repeats itself week after week, I’m forced to concede that it’s coming from a real place.

I’ve been thinking a lot about MH Rowe’s review of Ray Nayler’s Where the Axe is Buried, specifically about the concept of a professional novel. (I am not formally educated in literature or publishing, so if this is an established notion in the field, I’m not familiar with it.) I read Rowe’s definition of professional novels—in his words, novels that "feel like the latter-day inheritors of the pulp tradition”—as usually genre novels intended to be read mainly for entertainment. It is contrasted with “the great novel,” which I take to mean the literary novel. In the world where these categories exist, I assume the literary novelist writes mainly for art’s sake, while the professional novelist writes mainly to be able to write for a living.

I don’t think such a framing, if present, comes with any malice directed toward the professional novel or its writer. As Rowe notes, the writer of a professional novel may be exceptionally good at certain elements of the art of writing. Good professional novels are experts at managing suspense, a characteristic that makes or breaks novels of any kind. Anyone can write a dusty old tome, but it takes genuine talent to write for—and achieve—sustained entertainment.

I also believe the writer of the professional novel is the only one who has a snowball’s chance in hell at becoming a full-time novelist anymore, short of one of those social media miracles. John Scalzi, I think, is a great example of a writer of professional novels whose exceptional skill in entertaining people landed him a historic 10-year, 13-book, $3.4M USD contract with Tor.[1]

Scalzi got that deal off the 17 books he wrote between 2005 and 2015. Comparatively, 13 books in a decade is an easy ask. I find this to be a remarkable achievement; but like Stephen King’s usual nine-month-per-book pace, it also raises certain questions. I have read exactly one John Scalzi and enjoyed it well enough to pick up the next two in the trilogy, which sit unread on my shelf (though I am due to read all three soon as part of my fiction curriculum). It was entertaining! I suspect my reviews for any of his books might read the same. Similarly prolific SFF authors of this ilk I can think of offhand include T. Kingfisher, Stephen Graham Jones, Brandon Sanderson, and Adrian Tchaikovsky. God, but can they all spin a yarn. And all of these people derive enough income from their books to live on.

Now: I don’t think I want to be a writer of professional novels.

There has been much talk and discussion, including by me, about how the novelist must locate themself in this era of LLM-generated content. Unlike some authors on Bluesky—whose exceptional solidarity on the subject is all of impressive, laudable, and annoying—I do not think personally experimenting with LLMs/“generative AI” makes you worse than shit.[2] I think it is normal and socially appropriate to experiment with emergent technology, at the very least just to understand it. Such an instinct is born of curiosity and the drive to learn, traits that fundamentally separate us from computers.

I am partially influenced in these agnostic opinions by my field. Machine translation has been a mainstay in my industry for decades, and is now functionally essential in any translation firm that wants to stay afloat. I use it reluctantly at work, mainly for time savings on texts where the goal is plain language.[3]

This usage is also how I know I will never use it for my art: it cannot possibly be useful in this task. LLMs can produce a fine text if your goal is to convey information directly. In art, I am actually trying to find increasingly creative ways to avoid doing this. I am interested in the act of writing, in creation—not post-edition. I am interested in meaning and feeling and intention. I am interested in what reading at least 20 books will bring to my work—what approaching my creativity with a scholar’s instincts can do. This is good work to me. To get maybe a little weird with it, I believe it is the sort of work I am on this earth to do, the sort of work I am for.

I don’t think the distinction between the professional novel and any other is useful beyond making for an interesting discussion. Perhaps the writer of the professional novel is doing the sort of work they are for and find it deeply, fundamentally artistic in a way inextricable from their career ambitions. The post-editor of a novel crafted from generated content is probably not doing the sort of work they are for, but like the writer of the professional novel, the post-editor aims to entertain and make a living.

There’s a reason the professional novel sells well: it’s essential that your novel be entertaining if you want it to sell well. It takes a good amount of time, but not an insidious amount of time, to write a good and entertaining book. I read a great John Grisham this week that was so Save-the-Cat shaped I kept waiting for one to appear. I decided writing to publish wasn’t for me; it was interfering with my art. I suspect the most prolific professional novelists find an acceptable balance between these two factors and commit to it.

Post-editors apply the skill of post-edition under the assumption that if the result is sufficiently entertaining, it will sell. I think this assumption is correct. Streaming platforms, especially Netflix, are making similar gambles in their content: if it’s sufficiently entertaining, the amount or consideration of artistic direction does not matter. TikTok is built on this assumption being true. I think it’s undeniable that an audience exists for AI content of any stripe, especially when blockbuster movies have been cultivating our appetite for less creative fare since the success of The Avengers (2012).

I’ve been thinking a lot lately, especially as a translator wrangling generated content into shape, about how people are the original large language model. As children learning language, we quickly pinpoint which word is used in a particular context most often. As my bilingualism concretizes, I increasingly pull from French for the best word in a particular context because I have more vocabulary options to pull from. I’m drawing from every conversation I’ve ever had, every text I’ve ever read. LLMs can perform this function in exactly one way: the way their program dictates. As human beings, we have more options.

One of my strengths as a writer is that I can sit down and just type. I’ve lost five hours doing that into this post. More accurately, I’ve spent five hours typing and making choices: I pruned a few hundred words, rearranged, did research, incorporated and synthesized, chose and replaced words, fixed errors. For me, the process is the point; the journey is the point. These extraneous tasks associated with writing are what I am for.

I’ve been conscious of my waning desire to be a full-time novelist for a few months, but I think it’s an effect of my changing process and priorities. I write because writing is important to me. Learning to write better, developing my skills, is important to me. And doing research until I finally feel like I know enough to approach my novel as I envision it is not procrastination—it is important to me. I hope my writing means something to people, but mainly I must be the overthinking, pretentious bitch of my heart. This is out of an LLM’s reach. Here’s to the hours spent writing and doing all those auxiliary tasks.

-

[1] If you, like me, are interested in how Scalzi’s contract is going, he’s “only” published eight books in 10 years. This seems to have been sort of expected and is mainly based on market factors. He will instead write the 13 books over 15 years or so.

[2] As with any extracted resource, I believe LLM developers are criminally responsible for their illegal theft of materials for profit. I do not believe individual users of LLMs share that culpability. I don’t find fault with the person who fills their car with Shell gas; I find fault with Shell. With the usual clear exceptions, I think moralizing on an individual’s personal use of anything is a terrifically conservative instinct usually best avoided. (I also think there’s always power in a union.) I reject LLMs and their generated material because reading that is a cringe-ass way to spend my limited time on Earth. May we all take refuge in our own reasons.

[3] Any translator worth their salt will tell you that generated text still needs a ton of lifting to meet the bar—even the plain-language bar, even now. A five-hour job might now take four. It is fairly common that it saves no time at all. A thorough, researched confirmation of nuance and meaning of each term and phrase is still necessary. Disambiguation is almost always required. In my opinion, short of instructions from the client to the contrary, doing less constitutes negligence. And that’s before taking stylistic, structural, contextual, and tonal edits into account. Of course, these are the standards of a professional; many people will look at a generated text and find no fault. But though the fault may be subtle, it is there. The machine has no ability to discern such things.