Hands-on with Sudowrite's Story Engine, a generative AI tool for long-form fiction that works better than previous options but still has big limitations
Thoughts are swarming my mind like a nest of cyber-rats. … Last week, generative fiction tool Sudowrite launched a system for writing whole novels.
Context & Ripple Effects
When The Verge last looked at Sudowrite in mid-2022, it was a feature-level assistant: proposing plot twists, rewriting passages in specific tones, and suggesting metaphors inside a draft the human was already writing. Story Engine marks the step up — the same tooling reorganized into a pipeline meant to carry a writer from premise to a complete novel draft.
The hands-on verdict matters because it lands between two poles in the corpus: the New Yorker writer who found an LLM tuned to his style produced mostly hollow approximations of insight, and the later Wired reflection weighing whether these generators are good for literature at all. A tool that works 'better than previous options' but with big limitations sits squarely in that debate.
First-order effects
- Fiction writers gain a structured path to a full manuscript draft rather than piecemeal paragraph help — but the reported limitations mean the drafting burden shifts to revision and repair of generated chapters.
Second-order effects
- Rival AI-writing products are pushed to compete on whole-project workflows rather than single-shot text generation, since Story Engine resets what 'the product' is for long-form fiction.
Third-order effects
- If whole-novel generation becomes standard practice, the downstream effect visible in later experiments — where GPT-4 access boosted individual creativity but reduced the collective diversity of novel content — points toward a more homogenized fiction supply, with human value concentrated in curation and voice rather than first-draft production.
The trend: Generative fiction tools are climbing from assistive features to whole-manuscript pipelines, relocating the writer's work from drafting toward heavy editorial correction of machine output.