I am writing a novel series with an AI. Not prompting ChatGPT for plot ideas. Not using autocomplete to finish sentences. Not sprinkling generated prose into otherwise human-written chapters and hoping nobody notices.
I am running a writing room. Claude is the staff. I am the showrunner.
This is not a step down for writers. It is a step up.
A screenwriter who becomes a showrunner doesn’t stop writing — they stop doing only that. They hold the room. They set the standard. They decide what’s good enough and what gets rewritten. The craft doesn’t disappear; it moves to a higher floor. You’re not playing every instrument anymore. You’re conducting the symphony.
What I gave up was the agonizing over individual word choices in a first pass that was going to be rewritten anyway. What I kept was every decision that actually matters — voice, structure, character, intent, the line that makes a reader stop breathing for a second. The AI doesn’t know when a moment is earned. I do.
What this actually looks like in practice:
Before a single word of chapter one existed, there were weeks of architecture. Character dossiers with psychological profiles. Voice calibration documents — not just how a character speaks, but what they’ll never say, what they reach for under pressure, what they’ve spent twenty years not admitting. A constraint library that encodes every craft failure I catch, so the next session doesn’t repeat it. A canon database holding the entire world — locations, relationships, timeline, lore — queryable in real time.
You don’t walk into the writing room and say write me a story. You walk in and say: here is who this character is at her cellular level. Here is what this scene needs to do to the reader. Here is the quality bar — and here is what happened the last three times you fell short of it. Now draft.
The AI executes. I architect. The distinction is everything.
The project is Vampires of Tucson — a dark urban fantasy series set on the Arizona border, built around human trafficking, theological damnation, and the question of what it costs to remain human when the thing that rejected you is still standing right there. Traumatic Literary fiction. Production scale. Quality gates that would embarrass most traditional publishers.
And every word of it was written in collaboration with an AI.
Here is what I believe:
AI slop is a standards problem. Not a capability problem.
You already know what it sounds like. HFY channels on YouTube have built entire audiences on it — the flat affect, the hollow heroics, the prose that technically describes events without ever making you feel anything. The workflow is fully automated: AI writes it, Eleven Labs narrates it, YouTube publishes it. Some of them have accidentally sent the AI disclaimer straight through the pipeline. “Claude is an AI and can make mistakes. Please verify responses.” Read aloud by a synthetic voice. Into a microphone. Published. Nobody checked.
That is not a model failure. That is an author who was not in the room.
Compare that to this — a passage from Vampires of Tucson, the project that is proving this thesis:
“Did you hunt them?” Abuela asked. “Or did they hunt you?” “They hunted me first.” The words came out defensive. Aggressive. “I was just trying to survive. They were tracking me, had guns, were going to drag me back to…” “Mija.” Abuela’s voice cut through the overwhelming emotions. “I did not ask to justify. I asked which is true. Did you hunt them, or did they hunt you?” Rocket forced herself to breathe. To think past the defensive rage. “They hunted me. I killed them when they found me. Told them to stop. Let one go to tell the others to stop. They kept after me. I planned. Set traps. Waited for them to come close enough that I could…” She stopped. “I wanted them dead. Not just to survive. I wanted them to pay for what they did. For what they were going to keep doing to other girls if I didn’t stop them.”
That last admission didn’t come out of the first draft. The model produced a version where Rocket was more sympathetic — more victimized, easier to root for. I pushed back until she told the truth. Until she admitted she didn’t just survive them. She wanted them dead. That distinction is everything Rocket is as a character, and it only exists because someone was in the room demanding it.
That is what an author in the room looks like.
Every time you read AI-generated content and wince — the flat prose, the hollow characters, the sentences that technically say something but feel like nothing — that is not the model’s ceiling. That is the author’s floor. The model delivered exactly what the author was willing to accept.
I decided early that I would not accept it.
So I built systems. Quality gates with scoring criteria. A constraint library that encodes every craft failure I catch, so future sessions don’t repeat it. Character voice calibration. An MCP server with a canon database. Prompt architecture that treats the AI as a writing room, not a vending machine.
The result is prose that doesn’t read like AI wrote it. Because in the way that actually matters — the standards, the intent, the refusal to ship mediocrity — it didn’t.
This newsletter is the methodology.
Not the hype. Not the breathless “AI will replace writers” takes. Not the defensive “AI can never be creative” counter-takes.
This is about promoting yourself from the person choosing the words to the person running the room. That promotion comes with a cost nobody talks about: you have to be willing to enforce your standards. You have to say no — repeatedly, specifically, on the record — and then go back and rebuild the scene, the character, the chapter until it earns its place. The AI will give you something acceptable on the first pass. Acceptable is the enemy. You push again. You refine. You iterate. Relentless, uncompromising.
The writers who get slop out of AI are the writers who accepted the first draft.
In 1994, we stood at the edge of the Internet Era. Most people looked down and stepped back. The ones who stepped off that cliff built the world we’re living in now. We are standing at an identical precipice. The technology is different. The choice is the same.
I’m not one to “Thelma and Louise” my way into the canyon floor.
Writers can adapt — not by surrendering craft, but by carrying it to a higher floor. Or they can wait and see.
The writing room is already real. The only question is whether you’re running one.
Welcome to Architecting the Writing Room.
— E L Frederick