AI-Assisted Workflow and A Cold Civil War
Since October, I’ve been writing a serialized political thriller called A Cold Civil War. The project mixes conventional narrative chapters with fictional congressional reports, legal analyses, internal memos, future university course materials, magazine retrospectives, and other documentary-style material.
At its core, the project is an attempt to persuade people to think differently about the direction the country may be heading before political division hardens into something irreversible. I’m trying to use fiction to make institutional and constitutional risks feel personal and human rather than abstract. You rarely change hearts with logic alone. It usually takes a story.
One thing I realized early was that I did not have time to write this project the way novels are written. If the story was going to influence how people think about where the country may be heading, it had to respond while events were still unfolding and people were still persuadable, before a cold civil conflict hardened into something worse.
I’m neurodivergent. In elementary school, they used the word dyslexic. I learned early to use technology to fill the gaps. Before the Apple II was released, I built a word processor with spell check on an HP2000 in HP-BASIC. I solved that problem with technology, and I’m doing the same thing now. I need to move fast. Very fast. How does someone with my skill set write a novel in months?
To move this fast, I built a workflow designed around rapid iteration.
It’s not “AI writes the book for me.” It’s more like:
AI-assisted editing
continuity review
structural critique
factual stress-testing
procedural review
serialization support
I still write the drafts myself. But I use AI heavily as part of the editorial process, along with a growing set of prompts, Markdown workflows, Obsidian organization, and small Python tools built with Claude.
What surprised me was that the biggest gains came after the writing. They came from the editing workflow I built.
For example:
style passes
beta-reader passes
fact-check passes
line-edit passes
continuity passes after structural rewrites
The process feels much closer to running an evolving production pipeline than asking an AI to “write a scene.”
I also don’t see technology and serious writing as contradictions. Writers have always absorbed the tools available to them. What matters is whether the final work communicates something truthful, emotionally coherent, and worth reading. At least that’s what I hope I’m doing.
I organize everything in Obsidian using Markdown, though the same workflow could probably work in Google Docs or similar tools. The Python program would need modification.
My process currently looks like this:
Human Draft
I write the initial draft myself and revise it until it says what I want to say. Sometimes I use AI to brainstorm plot directions or scene structures. For example, in a scene called “Corruption Is Real,” I was trying to show how corruption works in the modern world while still making the scene feel human and readable. ChatGPT gave me five possible approaches. I didn't use any of them directly, but the process helped me think through the scene and sharpen what I was trying to do. Here is the scene:
In future posts, I hope to dive into how I use AI during the drafting process. This post is focused on editing.
Step 0: Setup
I upload the scene to an AI. I use Claude and ChatGPT. I also upload character notes and running summaries when needed. I sometimes upload prior scenes because they help keep the AI on course.
At the same time, I run a small Python tool built with Claude. It helps manage file paths, scene names, and side-by-side review of edits.
Python Program to Manage Document Edits
Step 1: Style Pass (iterative)
I run a dedicated style-check prompt on the AI.
I cut and paste the AI's output into the Python program and preview the edits. For each edit or note produced in the Python screen, I click:
accept
reject
I keep Obsidian open and may jump to the edit or note to make changes by hand.
After edits are applied, I reload the updated text into the AI before running another pass. This turned out to be critical. Otherwise, the AI keeps editing outdated text. I then tell the AI, “Re-review with the attached text.”
The notes are the most important. They often lead to major rewrites.
I repeat step 1 until the suggestions become minor or repetitive.
Steps 2–4
Then I repeat the same process using different prompts:
Fact Check
Beta Reader
Line Edit
The Beta Reader step has probably been the most valuable. It often catches pacing or structural problems I missed completely, especially in serialized fiction where scenes move around after major rewrites.
And if you want to see the project itself, the best starting point is probably the preface:
Preface — A Cold Civil War
Master Index
Substack Archive
All my posts related to AI are in this archive:
AI Writing Archive
I’d also be curious how other people here are approaching long-form AI-assisted fiction without falling into generic “AI voice.”