Kaizō: The Hard Way Out

Because AI and trad pub are the same damn thing.

AI-generated writing does not interest me. Authors who replace their artistic process with the outputs of large language models produce tinny pastiche that seldom holds up against serious reader scrutiny. Still, evidence increasingly suggests that traditional publishing’s gatekeepers prefer this product. Hachette was forced to rescind a book deal (Shy Girl) after readers clocked alleged AI slop. More recently, “The Serpent in the Grove” was published by Granta, and readers quickly discovered evidence of LLM provenance. It’s not hard to see why this has happened and keeps happening. A truth about publishing, privately admitted but rarely said in public, is that its own preferences and AI’s apparent tastes coincide.

It’s hard to make AI write a compelling story, because it resolves or forgets conflicts and, in any case, prefers small ones, as it’s been RLHF-trained to be agreeable. Large language models are most useful when their writing is clear; thus, AI avoids surprise and ambiguity. It takes the easy way out; in the context of an AI system, this is a desirable trait, not a bug. We don’t want, when we ask an LLM how to configure emacs or fix a toaster, to read a short story about a small town’s annual stochastic murder ritual. In 2026, traditional publishing also invariably takes the easy way out. If you’re struggling with a complex scene, a trade editor will advise you to just cut it, because paper’s expensive. If you’re a known author in a creative dry spell, your agent will send you a few dozen query letters to mine for ideas. If the story you’re writing isn’t easy for you to write, traditional publishing’s view is that the typical reader will be too TikTok-addled to easily read it, so you ought to take a break and focus on what’s really important: your author platform.

It’s hard to forecast the future of literature. I’ve written about traditional publishing’s intentions, circa 2028, to silently kill literary fiction, as it has done with epic fantasy under claims of word-count discipline, but I couldn’t possibly say whether it will successfully carry this plan out. To be honest, I care less and less every day about what the book industry does, as it seems doomed to the same soft plagiarism and safe self-repetition we see in AI-generated prose, for which I cannot muster excitement.

If you want to be traditionally published, you will obey word-count guidelines, you will install commercial hooks on the first page, you will accept the cover design and title chosen by the marketing department, you will eat the bugs, and you will be happy. You will take the easy route; if you cannot find one, one will be provided for you.

What would a refusal of this look like? I’m not sure. At least, I should not go forward without confessing my own personal interests, as a novelist but also as a reader. I can’t give a closed-form answer; the best I can do is propose one way out, never excluding other possibilities.

The term kaizō is used to describe fan-made modifications to video games designed for longevity and challenges. These games are hard, but they’re always fair. In short, kaizō is not about making the game impossible, because that’s an easy and cheap thing to do, but about craftsmanship. These are craftsmanlike games that reward craftsmanlike players.

The kaizō novel avoids difficulty for difficulty’s sake, but it rewards attention. It’s as accessible as it can be without watering itself down, but it refuses to condescend. It is often maximalist, but not bloated; it’s not afraid to run long if the story merits the weight, nor to shorten itself for the sake of compression. And while I would prefer to use the term kaizō without definition, as it feels more true to form to leave it for the world to figure out what I mean, it feels more useful to elaborate some of the principles I’ve observed in the kaizō design philosophy. So here goes:

1: You can do anything, as long as you do it well.

The distinction between literary and genre fiction, largely invented by an industry that has forfeited its cultural authority, is outmoded. If you want to include fantastical or future-technological elements in your story, do it. If you want to write a scene where people talk about baseball for twenty pages, subtly foreshadowing the back half of the story, go for it. You can even write a story that begins with a character waking up. But you must put in the work to do it well.

2: Never take the easy way out.

AI takes the easy way out. Traditional publishing increasingly rewards the same. Don’t succumb to the temptation. Write characters who fluster you. Write plots you might not be able to resolve—if you truly can’t, you can fix that in the next round of revision. Check your sentences five or six times until you’ve exhausted the obvious improvements. Run toward technical challenges, not away from them. Your story should be hard enough to write that you don’t have anything left to put into marketing; it should be self-destructively excellent.

3: Punch someone in the taint.

This one, not literally. Don’t do harm to real people. Attack systems and hidden traditions. Punch up, not down. Eviscerate bourgeois values and tendencies. Skewer traditional publishing. Number the assumptions on which your society’s worst traits rest, and invalidate them one by one. Never let the bad things in the world outrun you.

4: Unflinching realism.

The reader should feel like the places in your story truly exist. It was the highest compliment when someone said of Farisa, “I feel like I know her.” You don’t have to write a five-page history for every tree, but you should know whether it’s a mixed forest or a coniferous one, and you should write for the reader who wants to know this too. If your story features violence, include the horror and loss and regret. You’re not writing the summary; you’re writing the experience.

5: Every sentence does something.

There are people in publishing whose connections suffice to give them the benefit of the doubt if they insert sentences that don’t mean anything and don’t do anything, but you are probably not one of them and, even if you are, you should still write like an exile, because they’re the only people who write well. Your ordinary connective sentences should do at least one thing—reveal character, establish themes, advance plot—while your landmark sentences will do several at once. Your dialogue should be spoken by characters who believe they are the protagonist. Your language and style must serve the scene depicted; refine features that do, and remove those that do not.

6: Moral seriousness.

Fights hurt. Crime is corrosive. “Casual” sex does not exist. Good and evil are real, and they matter. You can write characters who choose evil—you probably will need at least one. You can include malice, violence, capitalism, and self-indulgence; as a general policy, you should take care never to glorify them, though you may write characters (viz. Nabokov’s most famous villain) who would try.

7: The prose can be painful.

If a character is bored, a boring paragraph is OK. A reader who closes a book after three sentences about the sounds of prison is missing the point. In disgusting circumstances, feel free to disgust the reader. When POV is held by an inarticulate character, blunt narration and plain language are best. I have a scene in Farisa’s Crossing whose opening sentence is, “Summer sat on the city like a fat girl.” This would be awful writing in 99 out of 100 contexts, but in the POV of a self-loathing obese male misogynist who works for the Global Company, it fits.

8: Readers want exercise—not mindless entertainment.

Reward attention. Use short sentences when they fit; ditto long ones. Trust the reader to catch subtext. Never pad or repeat yourself for the sake of skimmers. I don't advise using sources of difficulty that would kill flow—if the reader must refer to footnotes to get through your work, you’re probably overdoing it—but there’s nothing wrong with a story that tells more on the second read. You can’t provide jump scares, a laugh track, or the visual specificity of film; focus instead on what text does well. Don’t dumb your writing down to trad-pub levels, because if you’re scared of losing readers, they can tell, and you’ve probably already lost them.