A Hugo-Winning Novelist Just Drew a Line: No Generative AI in His Process

Charles Stross is a British science fiction writer with three Hugo Awards and a back catalog that includes the Laundry Files and the Merchant Princes series. He’s also a former programmer — he wrote code professionally before he wrote novels. Over the past several weeks, on his blog and in interviews, he’s stated flatly that generative AI will not be part of his writing process.

That combination is what makes it worth reading. This isn’t a Luddite reflex from someone who doesn’t understand the tooling. It’s a refusal from someone who has shipped software and knows exactly what automation can do.

Why “let the AI write your first draft” is the wrong pitch

The standard sales pitch for AI writing tools goes like this: you focus on the ideas, let the machine handle the tedious first draft. Sounds reasonable. It’s the same division-of-labor logic that sold every productivity tool of the last thirty years.

Stross attacks exactly that split. Writing the first draft is the thinking. A novelist is not a stenographer transcribing a finished book from their head. You write a sentence and discover who the character is. You get to chapter three and realize chapter one was wrong. Before you write, you don’t actually know what story you’re telling — not precisely, not in the way that matters.

So “machine does the draft, you supply the ideas” doesn’t hold up, because the ideas are born inside the draft. Outsourcing the first pass isn’t shedding the boring part. It’s handing over the only place where the novel actually gets made.

This is a different argument than the copyright fight

Writer opposition to AI has mostly run along two tracks. One is copyright: my books were scraped into training data without permission or payment. The other is labor: this technology is being used to eliminate paid work. Both were central to the 2023 WGA strike, and both are being litigated right now in cases like the Authors Guild suit against OpenAI.

Stross is opening a third track. His point is that even if you solved both of the first two, the objection survives. Imagine a model trained entirely on properly licensed data, with royalties flowing cleanly back to authors. He still wouldn’t use it. The act of writing is itself the creative work, and the moment you delegate it, there’s nothing left to have done.

That distinction matters more than it first appears. The copyright and labor arguments are the kind of thing institutions eventually absorb. Licensing deals get signed. Royalty structures get built. Union contracts get negotiated. Give it a decade and most of that will be plumbing. The third argument doesn’t work that way. No contract resolves it.

The programmer background is doing real work here

Developers reading this will have an obvious objection ready: I use coding agents every day and they’re great. Why is prose different?

Stross has an answer, and it’s the answer of someone who has written both. Code has an external arbiter. Tests pass or they don’t. It compiles or it doesn’t. When an agent writes a function, a machine can tell you whether it’s correct. Prose has no such oracle. Whether a sentence is good is decided by the writer’s judgment alone — and that judgment is built by writing tens of thousands of sentences badly first.

Here’s the uncomfortable part. If your job becomes picking from AI-generated options, the picking ability itself atrophies. It’s a muscle. At first you’re editing AI drafts competently, because you built that skill the hard way. Five years in, you may not be able to tell what needs editing. Similar worries circulate in software — there’s a growing literature on skill erosion among developers who lean heavily on agents. The difference is that code has a test suite as a backstop. Fiction has nothing.

Is this a position anyone can afford?

Let’s be honest about the economics. Stross can make this declaration because he’s already established. His books sell whether or not he ships faster. Tell a debut novelist facing a deadline, or a freelancer being quoted three cents a word, that the act of writing is sacred and they should refuse the tool, and it lands as a rich man’s principle.

Stross largely concedes this. He frames it as a rule for his own desk, not a norm he’s imposing on the field. But the question it generates isn’t personal at all: if a reader can’t distinguish AI-assisted prose from the other kind, does the method still matter?

His answer is yes, and it’s sharper than it sounds. The reader may not be able to tell. The writer is a different person either way. A novelist who published twenty AI-assisted books over ten years and one who wrote five by hand are not the same human being at the end of it. The difference isn’t in the output. It’s in who did the decade.

What the argument actually leaves behind

One caveat worth stating plainly: this hasn’t become a movement. Searching the last month of community discussion turns up no real groundswell — a handful of writers, some blog posts, no organized position. It’s still a minority conversation.

It’s notable anyway, because it moves the axis of the debate. Almost every argument about AI and creative work has been about output quality: is it good enough yet, can you tell the difference, when will it be indistinguishable. Stross declines to discuss output at all. He asks what disappears from the process.

That question isn’t confined to fiction. Whether you’re writing code or a strategy doc, you’re producing an artifact and simultaneously becoming someone capable of producing it. Outsource the second thing and the first keeps arriving, right up until it doesn’t. Worth asking what fraction of your own work is like that — where the process was never the overhead, but the point.