What happens when you run every email through an LLM for a year? One writer found out the hard way: you stop trusting your own sentences.
N. Cailie, writing on LessWrong, describes a pattern that will sound familiar to anyone who's leaned heavily on AI writing tools. It started with reasonable uses - checking emails, polishing drafts. Then it became a dependency. Every piece of writing got routed through Grammarly or an LLM for validation before sending. The result wasn't better writing. It was a slow erosion of confidence in their unedited voice.
The specifics are telling. Cailie's creative work - slam poetry that had previously won festival recognition when composed spontaneously - started sounding "generic" after months of AI-assisted refinement. A piece that was 80% original but polished through LLM tools got rejected for appearing AI-generated. The irony is sharp: using AI to improve your writing can make your writing indistinguishable from AI.
The Dependency Loop
This isn't just a personal anecdote. It describes a feedback loop that anyone using AI writing tools regularly should think about. You use the tool for polish. The tool smooths out your distinctive quirks. You start internalizing the tool's patterns. Your unassisted writing gets blander. You lean on the tool more because your raw output feels worse. Repeat.
Cailie offers no solutions, which is refreshingly honest. The essay is a diagnosis, not a prescription.
But the practical takeaway for people who write for a living is worth stating plainly: AI editing tools are best used for catching errors, not for rewriting voice. The moment you start accepting suggested rewrites of sentences that weren't broken - just different from how an LLM would phrase them - you're training yourself to write like a model. And readers can tell. The flat, agreeable, hedge-everything tone of LLM output is becoming instantly recognizable, and any writer whose voice starts converging on that pattern has a problem.
The best use of these tools is probably the most boring one: grammar checks, typo catches, maybe flagging unclear sentences. The moment they start touching your style, close the tab.