Charlie Stross on Why AI Stays Out of His Writing Process

Charlie Stross, author of the Laundry Files series and other technically literate SF, published a post explaining his explicit non-use of AI in his writing process. The post is notable less for being contrarian and more for being precise about the reasoning.

The core argument is not aesthetic aversion to technology. Stross has written extensively about AI, surveillance capitalism, and institutional dysfunction for decades, so reflexive technophobia is not the issue. The objection appears to be more specific: AI language models trained on existing text reproduce statistical patterns from that corpus, which means output trends toward the average of prior work. For a writer whose value is in distinctive voice and novel extrapolation, regressing toward the mean is a direct cost, not a neutral trade-off.

There is also a craft argument embedded here. Writing, for a working novelist, is not primarily a text-generation problem. It is a thinking problem. The process of drafting forces clarification of ideas, plot logic, and character motivation in ways that delegating output to a model would short-circuit. If the bottleneck is thinking, not typing, then a faster typing proxy solves the wrong problem.

For builders and technical founders, this is a useful framing to hold alongside the pro-AI-tooling default. LLMs accelerate output in contexts where output volume is the constraint and where mean-quality is acceptable or even desirable — boilerplate, test scaffolding, first-draft documentation. They degrade output in contexts where differentiation is the value and where the author's reasoning process is itself the product.

The post does not argue others should avoid AI. It argues that the decision should be deliberate, with a clear-eyed account of what you are actually optimizing for. That is the transferable point for any technical practitioner thinking about where to insert generation into their workflow.

Source