The most interesting moment in Core Memory’s OpenAI conversation is not the familiar complaint that AI writing feels flat.
Everyone has felt that by now. You ask for something sharp, intimate, funny, tragic, or alive, and what comes back is often competent in the way a hotel lobby is competent: polished, lit, and arranged by someone who has never lived there.
What makes the exchange useful is where Ashlee Vance pushes the question. He is not merely asking why the model cannot produce nicer sentences. He is asking why, after all this visible intelligence, the writing still fails to feel like it came from a mind with a point of view.
“The writing feels there’s no soul.”
Ashlee Vance, Core Memory, April 21, 2026
That line lands because it names the reader’s experience without pretending to diagnose the machinery underneath it. “Soul” is not a technical term, but it points toward a technical failure. The model may understand grammar, tone, and surface convention. It may even imitate a recognizable style. What it usually does not understand is the specific pressure that makes a particular sentence necessary.
A writer does not merely want prettier language. A writer wants language that belongs to the argument of the story.
Personalization Is Deeper Than Preference
Sam Altman’s answer moves toward personalization. A model built for a billion people has to produce language that a billion people can tolerate, which is very different from producing language that feels inevitable to one writer, inside one story, with one particular sense of what “good” means.
That distinction matters. When people talk about AI personalization, they usually mean taste. Make it sound more like me. Use my voice. Avoid the phrases I hate. Remember that I prefer restraint over enthusiasm, clarity over flourish, implication over explanation.
All of that matters, but it barely reaches the floor of what writers actually need. A writer’s taste is tied to the kind of meaning they are trying to create. Their preferences arise from pressure: the pressure of a Main Character trapped in a particular blind spot, an Influence Character forcing another way of seeing, an Objective Story conflict moving everyone toward consequence, and a Relationship Story changing the emotional weather between people.
This is why generic writing advice often fails. Two scenes can both be clear, emotional, and well-paced, while only one belongs to the story being told. The difference is structural. One scene expresses the Storyform; the other decorates the page.
AI writing feels empty when it treats writing as a surface-generation problem. It can generate sentences that resemble finished prose, but resemblance is cheap. The harder task is knowing which sentence advances the story’s argument, which image sharpens the conflict, which silence carries more meaning than dialogue, and which beautiful paragraph should be cut because it belongs to a different story.
That is the part writers recognize immediately, even when they cannot name it.
The Billion-Person Model Has a Story Problem
The scale problem is real. A single model has to serve almost everyone. The thing one person calls great writing may feel sentimental, cold, overwritten, underwritten, evasive, or lifeless to someone else. A model trying to satisfy everyone learns the safest center of language.
Altman frames the next step as making the model “so good at personalization” that different people, with different needs, can each experience it as a great writer. Read casually, that sounds like a style problem. Read from inside a serious writing process, it becomes something much more demanding.
For Dramatica, “what you think is great writing” cannot be separated from the structure underneath the writing. A comedy and a tragedy can share a tone and still require opposite movements of meaning. A Steadfast Main Character does not resolve pressure the same way a Change Main Character does. An Objective Story built around external activity does not generate the same kind of conflict as one built around fixed attitudes or psychological manipulation.
The writing that feels right is the writing that honors those differences.
This is where most AI systems flatten the work. They can remember that you like spare prose. They can remember that your protagonist is angry, your setting is coastal, your villain is charming, and your chapter needs more tension. But unless the system understands what kind of conflict the story is arguing through, it will keep improving the wallpaper while leaving the architecture vague.
A paragraph can be vivid and still be wrong.
Soul Comes From Structural Necessity
When writers say a passage has soul, they are often responding to necessity. The line feels like it could only come from this character, at this moment, under this pressure, in this story. It does not merely sound good. It reveals the shape of the conflict.
That is why voice alone is an inadequate target. Voice without structure becomes mannerism. The prose may sound distinctive, but the story will drift because nothing underneath is deciding what the language is for.
Dramatica gives us a more concrete way to talk about the missing element. A story is more than a sequence of events or a container for expressive prose. It is an argument built from Perspectives: Objective Story, Main Character, Influence Character, and Relationship Story. Each Throughline carries its own kind of conflict, and the writer’s job is to make those pressures felt through Storybeats, choices, reversals, and scenes.
AI writing improves dramatically when it stops asking only, “What should this sound like?” and starts asking, “What pressure is this moment supposed to make visible?” That question changes everything. It turns prose from decoration into diagnosis. It gives the model a reason to choose one detail over another, one emotional turn over another, one ending over another.
The future of AI writing will not be won by the model with the most pleasant default style. It will be won by systems that can understand the writer’s intended meaning well enough to protect it.
The Real Personal AI for Writers
The dream of a personal AI writer is not a machine that writes “like you” in the shallow sense. Writers do not need an echo. They need a collaborator that can recognize the story they are trying to tell before the draft fully knows how to tell it.
That kind of system would remember your taste, but it would also understand your Storyform. It would know when a scene is emotionally attractive but structurally evasive. It would know when a Main Character moment is being accidentally written from the Objective Story Perspective. It would know when the Relationship Story has gone quiet for too long, or when an Influence Character has become advice instead of pressure.
That is where the soul comes back in. Not because the machine possesses one, but because it finally has enough respect for the writer’s meaning to stop sanding everything down into acceptable prose.
The question raised by Vance and Altman is bigger than whether AI can become a great writer. A billion-person model may always struggle to know what greatness means for one story. The real breakthrough comes when the system can learn the writer’s standard of meaning, the story’s structure of conflict, and the difference between a sentence that sounds good and a sentence that belongs.
That is the work. That is where AI writing starts to move from fluent to alive.