A writer can feel the moment the sentence has started doing the thinking for them.
It looks finished. It has a subject, a verb, a tidy rhythm, perhaps even the right shade of feeling. Yet something is wrong: the character has made a decision the story has not prepared, the scene is repeating an idea it should complicate, or the beautiful conclusion is trying to resolve an argument that has never quite been made.
That moment is frustrating because it is also useful. The sentence has exposed the gap.
Eric Grunewald’s recent essay, “Why You Should Almost Never Use AI to Write Anything Substantive,” makes a necessary case for protecting that difficulty. His concern is not merely that AI prose has a recognizable texture. It is that drafting words is one of the ways a person discovers what they actually think.
“Writing with AI makes it easier to avoid the necessary thinking.”
— Eric Grunewald, “Why You Should Almost Never Use AI to Write Anything Substantive,” August 6, 2026
He is right.
A writer who gives a model the premise, the outline, the emotional turn, and the job of turning them into scenes may receive readable material. What they lose is the encounter with their own incompleteness: the transition that won’t hold because two ideas don’t belong together; the image that becomes false when placed beside the next one; the character explanation that reveals the author hasn’t yet understood the character.
For fiction, that encounter is not a delay before the real work. It is the real work.
Fluency Can Conceal a Missing Argument
Grunewald is especially sharp about the way AI prose can sound acceptable while carrying vagueness, factual error, or a sequence of small choices that nobody has fully examined. The trouble is not only that a model may be wrong. It is that the wording can make its wrongness easy to glide past.
Stories have their own version of this problem.
A model can invent a credible quarrel, a surprising reversal, or a line of dialogue that seems to illuminate a character. It can make a plot sound busy and emotionally literate. But a reader’s sense that a story has meaning comes from more than local plausibility. It comes from the way the parts of the story gather into an argument.
Dramatica calls that underlying argument the Storyform. Storytelling is the particular expression of it: the world, the voice, the images, the casting, the sequence of scenes, and the words on the page. A Storyform describes how four distinct perspectives on conflict work together: the Objective Story, Main Character, Influence Character, and Relationship Story.
A machine can create a scene that works momentarily while leaving those perspectives in disarray. The Objective Story may be moving toward one conclusion while the Main Character’s personal arc quietly argues for another. The Influence Character may become a source of agreeable advice instead of a meaningful alternative. The Relationship Story may disappear whenever the plot gets interesting.
The prose can still be smooth.
That is why smooth prose is such a poor measure of whether a story has found its meaning.
The Useful Role Is Interlocutor
Grunewald does not reject AI wholesale. He explicitly allows for transcription, data analysis, research, brainstorming, feedback, copy editing, and clarity revisions when a human deliberately accepts or rejects the changes. The line he draws is around asking a model to do the substantive writing itself.
Dramatica’s position begins in the same place. The writer should remain the author of the Storytelling: the person who chooses the words, gives the characters their particular lives, and decides what the work is finally saying.
But AI can be remarkably valuable before and alongside the draft, when it is asked to participate in a different kind of conversation.
A writer might bring a Storyform to an AI and ask: “My Main Character changes at the end, but have I shown the personal inequity clearly enough for that change to feel inevitable rather than convenient?” Or: “This scene advances the Objective Story, but what has it done to the Relationship Story?” Or: “If the Influence Character’s approach is meant to challenge the Main Character’s worldview, where have I made that challenge too easy to dismiss?”
Those are not requests for a substitute scene. They are questions that make a writer return to the story with greater precision.
The model can misread the story. It can offer a banal answer. It can identify a contradiction the writer intentionally wants to preserve. Its value lies in bringing possibilities into view so the writer can judge them. The author still decides which reading is true, which scene belongs, and which line has the exact weight the moment requires.
AI can help a writer see a story’s questions more clearly. It cannot answer them on the writer’s behalf.
A Structural Conversation Changes the Prompt
Most writing-AI workflows begin with a command: “Write this scene.” The author supplies a few facts and asks the system to turn them into prose.
A Dramatica-informed workflow begins elsewhere. It asks the writer to name the structural work at stake before any sentence appears.
What is the Objective Story trying to resolve? What does the Main Character believe or do that keeps them bound to their personal conflict? What alternative does the Influence Character embody? How is the Relationship Story changing because these two people cannot remain exactly as they were?
These questions can be difficult. They should be. They force a writer to choose among interpretations that may all sound workable on a synopsis.
An AI assistant grounded in the Storyform can help compare those interpretations without flattening them. It can trace where a proposed revision might change a Throughline. It can point out that a scene is being asked to carry two incompatible Storypoints. It can help a writer distinguish a new Storytelling choice from a change in the argument beneath it.
The benefit is not speed for its own sake. The benefit is continuity of thought.
A writer is still free to discover that the Storyform itself needs to change. They may realize the ending they wanted belongs to a different story, or that a character they assumed was the Main Character is actually functioning somewhere else. Dramatica does not turn those discoveries into errors. It gives them names, consequences, and a way to be considered deliberately.
The Context Must Travel With the Work
This is one reason the Narrative Context Protocol matters.
NCP is an open format for carrying a story’s working context across tools, agents, and interfaces. It can preserve structural choices such as Storyform, Throughlines, Storypoints, Storybeats, Perspectives, and the authored material connected to them. It keeps the underlying argument distinct from the many forms of Storytelling that may emerge while a work is developed.
Without that context, every new AI conversation begins with a partial summary. The writer explains the premise again, pastes in some notes, and hopes the model has understood which character matters and why. The result is often a fluent approximation of the project: a version that remembers the names but loses the meaning.
With shared narrative context, a writer can ask a precise question in one environment and continue the work in another. An editor can see the intended Storypoint behind a revision. A new agent can understand that an outline is provisional while the Relationship Story is central. The writer has more opportunities to challenge the work without making each tool a fresh co-author of its premise.
NCP is not a license to delegate authorship. It is a way to preserve authorial intent while using tools that would otherwise lose it.
Attention Is Part of the Exchange
Grunewald also names a social obligation that deserves more attention. A reader gives time to a piece of writing with the reasonable expectation that it reflects someone’s considered mind.
“The reader offers their attention and the writer repays that with something of value.”
— Eric Grunewald, “Why You Should Almost Never Use AI to Write Anything Substantive,” August 6, 2026
That contract applies to fiction in a particularly intimate way.
A reader does not open a novel only to receive a sequence of competent sentences. They come to inhabit a way of seeing: the author’s particular arrangement of event, feeling, memory, consequence, and silence. They want to encounter the thought that made this image follow that action, and this ending answer the beginning.
When AI-generated prose is offered as though it were the author’s own work, the reader cannot know where that arrangement came from. The relationship becomes cloudy. Disclosure matters, and authors should follow the requirements of their publishers, contracts, and applicable registration guidance.
The strongest use of AI does not depend on hiding it. A writer can say, honestly, that they used a system to interrogate an outline, test competing readings, track continuity, or surface structural questions. Those uses may make the human work more accountable, because they leave the author facing the choices that form the story rather than accepting a polished substitute for them.
The Sentence Still Belongs to the Writer
There is a strange anxiety beneath many conversations about AI and writing: the fear that if a model can make sentences, the sentence was never very important.
But the sentence is important because it is where the story becomes particular.
A Storyform may clarify the argument. An AI interlocutor may help a writer see a contradiction or recover a forgotten possibility. NCP may keep the work’s structural context intact as it moves between tools. None of them can decide how a character hears the rain outside a locked door, which detail arrives too late, or what a final image must leave unsaid.
Those choices belong to the author because they are the author’s meaning made visible.
Grunewald’s warning is worth taking seriously: don’t hand over the work of thinking merely because a machine can offer language faster than you can. Use the machine where it helps you look harder. Then return to the page, where the story still has to become something only a human being could mean.
Sources
- Eric Grunewald, “Why You Should Almost Never Use AI to Write Anything Substantive,” August 6, 2026.
- “Narrative Context Protocol,” Dramatica.
- “Of Stories and Storyforms,” Dramatica Narrative Platform documentation.