9. PRESERVING THE DEPOSIT

Our dataset will be the most precious thing we own in the age of intelligence, at least for the duration of the transition. I would think twice before signing over my rights to a publishing house for 25 years or for eternity (I have never understood authors who sign such contracts; apart from the first book I published with a traditional publisher, I have only ever signed 10- or 15-year contracts). In fact, I would hesitate to sign even a 5-year contract right now. The texts you own will become your greatest asset in the new world taking shape. Not only could they allow you to explore — even to elucidate — your vocal imprint, but one can imagine many other uses as well. Whereas a text published with a traditional house remains frozen in place, a proprietary text can be spun into multiple variations by means of AI. It can be expanded, transformed, compressed, or distorted. Two texts can be cross-bred to yield a third (following the formula: 1 + 1 = 3). A text can be “adapted” into AI-generated images or video. The possibilities are virtually limitless — but they suddenly become very narrow when a publishing house restricts what we can do with our own texts.

Scattered notes can be organized, assembled, and reassembled. Embryonic texts can be developed, inflated like those little sponge dinosaurs we had as children: once you dropped them in water, they swelled before your eyes — except that here, the water is the set of constraints you feed the machine, and the resulting shape is never quite predictable. For instance, I have always wanted to write a narrative that might have been called Things to Say Under the Stars. It is an obsession that has followed me for about ten years — or perhaps I have been carrying it, unnamed, since adolescence. Two friends in a park, sitting on the grass, at the foot of the pines, at night, under stars half-erased by the city lights, finally saying the things no one ever says; or saying that they are going to say them, saying how good it would feel to finally say what no one ever says, if only they knew how… I have started this récit several times, but I have never managed to write it — perhaps because it is too fictional. What will happen when I prompt it into Claude Opus, loading hundreds of pages of my prose from other texts into the context window, and pointing toward that intuition, that feeling of Things to Say Under the Stars? I could ask the AI to produce 100 versions of 100 pages each, choose the one that speaks to me most, rework it in collaboration with the machine, expand it if needed, cut it if necessary, dilate the truest passages, and so on.

Those who have over-invested in the legacy literary world and hold few proprietary texts will find themselves impoverished and constrained. Those who, on the contrary, possess a dense and complex data corpus will be the new rich of the world to come. I am of course using the word “rich” in the broadest sense: what matters above all is literary wealth — the creative variety that can be extracted from the data. For that matter, we will also be free to have a field day with public domain texts: those too can be declined, rearranged, deconstructed, and reconstructed. The fact remains that we are not monks — not even copyist monks — and wealth must also translate into hard currency. We are at the beginning of the transition, and in this in-between, we need to be able to work within legacy structures — the only ones that do not have to be invented from scratch. That is why I raised the issue of copyright. The first move of the transition, as I see it, is to reclaim ownership of one’s texts as far as possible and to use that leverage to shift from a copyright logic to a proprietary-data logic, from reproduction rights (the book) to control of the deposit (the dataset, the source).

Among the possible derivations of our texts, there is also translation — a domain over which publishers almost always reserve control by contract. I believe this is where one of the main commercial opportunities lies during the transition years. I have run a few translation tests on my own texts; to read them, click here. I find the results impressive. When you ask a chatbot to write a text from a simple prompt, the pull of the average is ever-present; in the current state of the models, as we have seen, you must dictate a whole series of constraints to pull the AI toward a particular style, and you absolutely must use the most capable model for creative writing. With translation, the situation is different, because the source sentences serve as a guide for the machine, keeping it from straying too far from the direction you want to set. The translation of ordinary texts — what is called “pragmatic,” non-literary translation — has effectively ceased to exist as a human activity (it is now revision of AI output). Today, it is literary translators’ turn to feel the ground giving way beneath their feet. In France, the Association of Literary Translators has introduced a no-AI clause in its standard contract, which speaks volumes about the profession’s anxiety. If one were confident that artificial translation could be reliably distinguished from human translation, such precautions would be unnecessary. I am obviously not opposed to measures protecting literary translators — though I believe they will sooner or later, like authors, have to adapt their practices to the world that is coming.

When it comes to translating our own texts whose rights — and, by extension, proprietary data — we hold, I see things differently. If I want to translate Relief into English using Claude Opus or another model, one paragraph at a time, revising the output myself (human in the loop), that is entirely within my rights. I invite authors who want to enter the age of intelligence to do the same, provided they have working knowledge of another language. I am a conference interpreter and my English is strong, but I am not a native speaker: for my translations, I will therefore not forgo a final review by an anglophone. That said, the cost of a review bears no comparison to the cost of a full translation. We non-anglophone authors thus have the opportunity, in 2026, to translate our texts into English and thereby enter the largest market there is: the English-language book market, publishing in particular through Amazon KDP.¹

This is perhaps the best avenue for — at the very least — replacing the income of the legacy literary world. I have other ideas, but they depend on inventing structures that do not yet exist, and I know how difficult it is to build something from scratch. If we manage to form a small “Literature & AI” exploration group, we can exchange ideas and perhaps trace the outlines of a new economic model for practitioners of the art of writing.

In the meantime, all I can do is encourage authors who want to board the aircraft to turn themselves into machines, as I put it earlier. Does it still make sense to write all one’s texts the old-fashioned way and publish a book every two or three years, when proprietary data makes it possible to deploy a wealth of forms? In any case, given the acceleration of change, what world will we be living in two or three years from now? If we turn our backs on intelligence, we will remain frozen in the legacy world, shut out of intelligent writing — which may well become the new name for writing itself.

There is a mourning to be done. It is a molting. We shed our old parchment skin and accept being raw and exposed while the new skin takes shape.

---

1. Kindle Direct Publishing, Amazon’s self-publishing platform. ↩