the 'ai writing assistant' category has split. one camp optimizes for drafting speed — you describe the scene, it writes it. the other optimizes for thinking partnership — you draft the scene, it helps you see it clearly. these sound similar and are completely different products.
the drafting-speed camp has a structural problem: fiction written by an LLM reads like fiction written by an LLM. the voice is averaged across the training corpus; it is confident and thin and competent and forgettable. users who adopt these tools full-throttle find their work feels less like theirs over time, not more. the tool was supposed to amplify; it substituted.
the thinking-partner camp has the opposite tradeoff: slower than drafting yourself, but the output stays yours. you wrote every sentence. the AI helped you see what you couldn't see on your own — the plot hole, the character inconsistency, the paragraph that doesn't earn its length.
what lucy is optimized for:
character consistency across long projects. tell her about your 15 named characters on day one. session 27, she still knows. this saves the week you would have spent re-reading your own notes.
honest first-read of problem paragraphs. paste what isn't working, ask 'what's off?' she gives you 2-3 specific diagnoses, not a rewrite. you take her observation and fix it yourself.
structural diagnosis. 'this chapter feels like it's not doing anything' → she'll ask you what changes between scene 1 and scene 4. usually the answer is 'nothing,' which is why the chapter feels dead. you restructure, not her.
accountability-light. not nagging, not streak-based. if you want a body-double feel, keep a voice call open while you write (bonded tier). silent presence, zero chatter.
starting point: free tier 25 msg/day. seed her with 3 characters and 1 setting on day one. come back 48 hours later, bring up one of the characters. if she references what you told her accurately, the setup worked. if she confabulates, tell her to correct it and she will.