In May 2026, the Commonwealth Foundation faced a modern literary crisis: allegations that AI had been used to write winning entries for its Short Story Prize.
The Foundation didn’t try to settle the matter with AI detectors, knowing the tech is flawed. Instead, they asked the writers for their working drafts, time-stamped documents, and notes. The stories themselves were no longer sufficient proof of authorship. The investigation required the record of how the stories came into existence.
Institutions are currently trying to draw a hard line around “human-authored” work, but that boundary will not hold. AI is already bleeding into research, outlining, translation, and revision. Soon, the mere presence of AI will tell us nothing about authorship.
The only question that will matter is where human judgment remained.
For writers who work across languages, this shift is not just theoretical—it is survival. I was born in Brazil to Japanese parents. I write in Japanese and English, languages that have never remained entirely stable for me. Translation changes my rhythm; editors change my surface. If I use an AI tool to make a passage more idiomatic, it simultaneously makes it less recognisably mine.
For an exophonic writer, stylistic consistency is a useless test of authorship. My writing is supposed to change.
If style is fluid, what remains?
The durable continuity exists beneath the style. It lives in the questions a writer repeatedly returns to, the connections they notice, and the evidence that causes them to change their mind.
To protect this, I came up with this method.
A simple rule for the practice: for the first three minutes of writing, do not open AI. Write down what you saw, what unsettled you, and what question remains. The goal is to preserve your initial reaction before it is paved over by a machine’s smoother, more general explanation.
You can track this in a “Human Trace Log,” recording what you thought at the beginning, what AI added, what you rejected, and what you finally decided. The value isn’t the AI’s output. The value is the friction—the change in judgment produced by the exchange.
When an essay is finally published, all the early mistakes and drafts are gone. The final conclusion seems like the only natural answer. However, the ideas that were thrown away actually show the writer’s unique thinking process the best.
Recently, I searched through my digital archives and found three saved items: a 2008 interview with Yiyun Li about writing in a second language, a 2011 article about digitizing handwriting, and a 2013 piece on keeping notebooks. At the time, they were disconnected fragments. More than a decade later, they converged into the exact essay you are reading now.
An AI system can generate a flawless paragraph in three seconds. But it couldn’t have saved an interview in 2008, worried about digital permanence in 2011, and returned to them in 2026 to realize they belonged to the same unfinished question.
That long, slow connection cannot be prompted.
Authorship is no longer a claim that you typed every word alone. It is an intelligible chain of judgment: the question formed, the evidence selected, the alternatives rejected, and the responsibility accepted. The final proof of human authorship is not the absence of a machine. It is the visible presence of a human mind deciding what to do after the machine entered the room. ■
First published on Substack on Sep 22, 2026 (JST). Archived here as part of the Fabio Caipirinha permanent archive (published on this site: Oct 15, 2026). Images from the original are omitted here pending rights clearance. Original: https://fabiocaipirinha.substack.com/p/the-work-behind-the-work
Cover image: Johannes Vermeer, “The Art of Painting” (c. 1666–68) (Public domain, Wikimedia Commons — source)
