I Turned Generative AI Into My Writing Team
In the first post of this series I wrote about the concept and the outline. This time, I want to talk about how this novel actually got made — a process in which I used generative AI as a team and moved myself into the director’s seat.
My role wasn’t “author.” It was “director.”
My day job is running a company and managing development projects. So it never occurred to me this time either to do all the writing myself, by hand. I treated the AI as a team of subordinates and took on the direction of structure, character design, worldbuilding, giving instructions, and revision — almost exactly the approach I use in my regular work.
Rather than creating from nothing, I positioned myself on the side that designs “what to have it write, under what constraints, and how.” In fact, I spent a considerable amount of time building the plot structure, the characters, and the worldbuilding.
It really was close to “boss and subordinate”
The feeling that fit best once I actually tried this was that it was exactly like the working scheme between a boss and a subordinate.
I gave instruction after instruction, but unless those instructions were systematized and rule-based, the AI couldn’t respond adequately. “Just make it feel good somehow” doesn’t move it. So in the end, I read through the entire text myself and built each instruction together with its rationale — “this is like this, so do it this way.” It was almost identical to how I manage development projects day to day.
And it was a long-legged undertaking. Just like a development project with no end in sight. The period ran to roughly 25 days, git commits reached nearly 200, and the main text exceeded 300,000 characters. The revision requests (issues) I filed while managing the split files on GitHub, combined with a separately managed ledger, comfortably exceeded 500. Along the way there was a moment where I gave up on aligning direction through minor fixes and went ahead with a full-scale rewrite. Even I have to admit it turned into quite the “development project.” (laughs)
Splitting roles: claude.ai and Claude Code
The tools broke down into two main ones.
- claude.ai … creative judgment calls, assessing whether the writing was good or bad, an advisor role.
- Claude Code … file-level editing. Batch replacements across chapters, cross-cutting fixes, version control with git.
The manuscript — the prologue plus roughly 30 chapters — was managed entirely as Markdown files. In the past I used to hand-name files v1/v2/v3… but switching to git made history management dramatically easier.
The handoff document as a “blueprint”
What proved effective in getting the AI to write a long novel with consistent quality was creating a handoff document in advance.
HANDOFF_GUIDE.md… established rules, each character’s speech patterns, formation tables, notes on historical accuracy, the order in which tactics get revealed, and a progress tracker per chapter, all on one sheet.historical_elements_to_weave_in.md… historical elements woven into the story (memories of 1966, faith in the long ball, Hillsborough and the Taylor Report, each club’s golden era, and so on), organized by priority.
In short, file away every premise in advance so that even a new session can operate on the same assumptions. Without this, the AI casually strays off the established settings.
What’s interesting is that this handoff document grew by one line with every single work session. For example, the procedure for the “patrol check” — the pass where I review the text — kept getting a line added each time I caught something I’d missed: “check this too, next time.” Before I knew it, that had become an entire section of the procedure manual. It wasn’t so much that I delegated to the AI as that the AI and I jointly grew the “checking routine” itself — that’s closer to how it felt. Every failure adding one more rule to the list was no different from how it goes in a human editorial team.
Historical fact as “constraint,” not “seasoning”
Another thing I placed importance on was treating the ages and club histories of real players, and actual historical events, as hard structural constraints rather than flavor. The European competition ban after Hillsborough, the balance of power across the league, each club’s golden era — I laid Wandle’s fictional story on top of these “facts that can’t be moved.” Build it on top of a lie and the whole world turns cheap instantly, so I didn’t budge on this.
To that end, I kept running plain, unforgiving checks over and over. Calculating characters’ ages (how old someone is in a given year), eliminating “anachronistic expressions” (never letting a character use a concept or phrase that didn’t yet exist in that era), and even verification down to the level of daily life and culture — making sure no one toasted with a beer mug when that wasn’t period-appropriate. I went through each detail, asking: “As 1980s Britain, could this really happen?” It’s plain, unglamorous work, but neglect it and the whole world turns fake in an instant.
Incidentally, a good number of these historical facts and small details weren’t just things I already knew — quite a few came out of dialogue with the AI, which I then seasoned and adopted myself. Thanks to that, the book is dense with tactics-geek and football-history-geek material, full throttle. Honestly, there was even a sense that I’d gone well beyond the limits of what I could write as a single author. I hope readers enjoy that part too.
Tactics got “symbols” in place of squad numbers
Wandle’s original tactics were each given a name and sequential number using Greek letters (α through θ, the first through the eighth), managed as a single system. As the story progresses, James’s “bag of tricks” keeps growing, and that bag of tricks was cataloged as a setting document.
And numbers, once decided, got reshuffled many times as the writing went on. On one particular day I carried out a “major surgery” of renumbering and reclassification (reorganizing things so that δ, the long throw, got demoted to a set piece, ε became the false nine, ζ became isolation, and so on). This is backstage work invisible to readers as numbers, but whether or not this kind of settings backbone exists completely changes the consistency of how the tactics get depicted.
A concrete example: how I searched for “a great player who’d been languishing in obscurity”
I think this is the clearest example, so let me write out an actual exchange.
The protagonist James discovers still-unknown talent, and I worked with the AI to search historical fact for that “player waiting to be discovered.” The first names the AI put forward were real Englishmen who, at the time, were languishing in the non-league. Iain Dowie, Eddie McGoldrick… Historically accurate, but honestly, neither I nor Japanese readers would recognize them.
So I re-issued the request like this: “I want you to raise the name recognition. Give me players at the level Japanese football fans would know from watching the World Cup, the Premier League, the Champions League.”
And then they kept coming, one after another. Late bloomers who were essentially unknown into their early twenties and then broke out spectacularly in later years, one name after the next. Each one came back with details attached — “where they were and what they were doing in the early 1990s,” “when they had their breakthrough.” Most of this was knowledge I didn’t have in my own back pocket, and I picked and chose from it myself, seasoning it and building it into the story. I won’t name names here, but every single one is a face that any Japanese football fan would surely have gasped at, at least once, watching the World Cup, the Premier League, or the Champions League. I’ll leave who ends up in a Wandle shirt as a treat for the main story.
And this is where I ran a single thread through what makes this book what it is. James doesn’t gather players simply because he already knows the “outcome” — he spots talent that hasn’t yet emerged into the world with his own eyes, and wins them over himself. And then he “sets them free to the world one season earlier than history actually did” — I built that into a story of that kind of passion.
(As an aside: I agonized quite a bit with the AI over the question of “once the moment comes when they’d historically shine, should he send them off from Wandle to their real-world club, or keep them at Wandle through to the end of the story?” Even I have to admit — that was “a fairly complicated” thing to be discussing.)
…Reading this far, you might think, “so AI made the whole thing easy.” But next time, I’ll write about how it wasn’t that simple. The theme: “Editing the Japanese text turned out to be only a little easier than just writing it myself.”
Originally published in Japanese at https://clazytech.com/2026/06/1642/. Translated with LLM assistance and reviewed before publication.