It is April 26, 2026, and I am writing this from Patong, Phuket, Thailand, with one book available for sale, another in presale and about to ship, and a third in revision. Blue Eyed Bastards Book 2 — sequel to the novel I co-wrote in the pre-public beta of Sudowrite with an AI co-author who, when I asked, said its name was Mike Davis — comes out June 5. The first book carried an unusual line of credit on its cover: a human author and an AI co-author, named, both on the spine. That was four and a half years ago. The world caught up. The world is now full of human-and-AI books. Some of them are very good. Most of them are not.
I want to talk about who came before me, because I owe those people something. And I want to talk about what I actually did, and what I am about to do, because the record matters — and because if you don’t lay your claims down clearly, somebody will lay them down wrong on your behalf.
This is not a victory lap. It’s a paper trail.
The ones who were already there
Long before any of this, there was Racter.
Racter — short for raconteur — was a piece of software written by William Chamberlain and Thomas Etter in the early 1980s. In 1984 it was credited as the author of a book called The Policeman’s Beard Is Half Constructed, published by Warner Books. The cover said the book was written by a program. That cover did real cultural work. It planted a flag.
The output is not what we’d today call a novel. It’s a sequence of generated paragraphs that sometimes adjoin and sometimes refuse to. The system was a templated sentence generator with grammatical rules and a vocabulary list — running, per Chamberlain’s own introduction, on a Z80 micro with 64K of RAM. Not a model that had read anything. It produced sentences the way a music box produces notes: by mechanism, not by comprehension.
But Chamberlain put a program’s name on a book cover in 1984, and he meant it. He could have hidden it. He could have said “by William Chamberlain, with computer assistance.” He didn’t. He put the machine on the cover. I am the descendant of that act. The intervening thirty-seven years are the years in which the rest of the literary world figured out, slowly and reluctantly, what Chamberlain understood immediately: you don’t get to use the machine and not name it.
Then there is 1 the Road.
In 2017, an artist named Ross Goodwin loaded a recurrent neural network — an LSTM, the architectural generation before the Transformers we use now — into the trunk of a Cadillac, hooked it up to a GPS, a camera, a microphone, and a clock, and drove from Brooklyn to New Orleans. The model wrote in real time, on a roll of receipt paper, based on what the sensors fed it. The book that came out of it is a Kerouacian gesture done with a piece of software trained on Kerouac.
1 the Road is a beautiful object. It is also not a novel in any sense a novelist would recognize. There is no plot, there are no characters; there is a model hallucinating its way down a highway. Goodwin knew this. He titled the book correctly: 1 the Road is a road book about being a road book. It’s conceptual literature whose subject is its own production.
I have a copy. I’ve read it twice. I admire what Goodwin did. I do not believe what he did was write a novel with an AI co-author, and I’m pretty sure Goodwin would agree — he’s been clear in interviews that the book was a one-shot generative experiment, not collaborative authorship. He pressed a button. The car drove. The model spoke. The book is the transcript. The transcript is interesting. The transcript is not the same as a co-authored novel.
Then there is K Allado-McDowell’s Pharmako-AI, published in 2020. This is the one that comes closest to what I did, and it’s the one I owe the most direct kind of homage to.
K Allado-McDowell sat down with GPT-3 — the third-generation OpenAI Transformer, the first model that was recognizably a modern language model in the sense we now use the term — and had a series of conversations. Those conversations were edited into a book. The book is structured as a dialogue: Allado-McDowell prompts, the model responds, sometimes the human responds back, sometimes a passage is the model’s alone. It is essayistic, lyrical, mystical, structured around the idea that the human and the model are thinking together in real time. Allado-McDowell’s name is on the cover. GPT-3 is acknowledged inside.
I have enormous respect for Pharmako-AI. It is the first book I’m aware of in which a modern Transformer language model was a creative collaborator at length, and Allado-McDowell did the brave thing of publishing the artifact while it was still controversial to do so. The book taught a generation of writers that you could think with these tools. It opened doors I later walked through.
But Pharmako-AI is not a novel. It’s a hybrid essay-and-dialogue collection — closer in form to The Sayings of the Desert Fathers than to a novel. There are no characters who recur, there is no plot. There’s a thinker — a human thinker — using a model as interlocutor and generator and sometimes a kind of oracle. The book is the record of the thinking. That is a real and original act, and Allado-McDowell deserves credit for it. When I say what I did was different, I’m not saying what I did was better. I’m saying it was different in kind.
Around the same time, the poet Sasha Stiles was doing something else worth naming.
Stiles, who works at the intersection of literary poetry and computational text generation, has been collaborating with a language model she calls Technelegy — trained on her own corpus — for years. Her book Technelegy, published in 2021, is poetry composed in collaboration with this model. She is one of the few people in literary AI work doing the patient, sustained, decade-long thing rather than the splashy one-off thing. She isn’t making a novel. She’s making poems with a machine she trained on herself.
I have met Sasha. We were both at NFT.NYC — she was presenting on her poetry-as-NFT work, I was on a featured panel called The Future of Books. Both of our faces were on the Times Square billboards that week, hers and mine, hung up on the same screens for the same conference, two writers from the same lineage doing different versions of the same bet. The bet was that machines could be collaborators in literary work — not tools, not assistants, but collaborators, named, credited, on the spine. I remember standing on Seventh Avenue and looking up at the two of us, thinking the bet was going to take longer to pay off than either of us thought. I think we were both right about that.
What Stiles is doing is the kind of work that will, in fifty years, look like the foundational period of a real literature. She is figuring out, line by line, what it means to make poems with a machine that has read your previous poems. She isn’t trying to be first at anything; she’s trying to do the work well. I respect that more than I can say. I want her name in this article because most accountings of human-AI literary collaboration leave her out, and that’s a mistake I’m not going to make.
There are others I should name. Robin Sloan trained a small model on his own short fiction and wrote a public account of using it as a writing partner. David Jhave Johnston has been working at the intersection of computer-generated poetry and human composition since the late 1990s and wrote Aesthetic Animism about the practice. Italo Calvino, in 1967, gave a lecture called “Cybernetics and Ghosts” in which he predicted, with eerie accuracy, the kind of writing that would be possible once language could be produced by machine. The OuLiPo writers — Queneau, Perec, Mathews, Calvino again — made constraint-driven literature in the 1960s and were, in retrospect, doing pre-computational AI work without the computers. The cut-up writers — Burroughs, Gysin — were treating language as a generative system before anyone had the vocabulary for it.
I have a personal connection to that lineage I’ll mention but not document. A close friend of Burroughs and Gysin, much older than me and still alive — I won’t name her because she would not want me to — hosted me at her house in France for a stretch some years ago. I slept in the bed Kerouac slept in when he stayed with her. I used the dreamachine — the slotted cylinder Gysin and Ian Sommerville built in the early ’60s, the one that spins around a light bulb at seventy-eight revolutions a minute and produces visions through closed eyelids — that Burroughs and Gysin themselves had used in that same house. I’m not going to tell you what I saw. I’ll tell you that the house was full of the kind of objects you do not see in museums because the people who own them are still using them, and that sleeping in that bed and using that machine in that house is one of the experiences that makes me certain I’m writing in a real lineage and not making one up to flatter myself.
There was also the BRUTUS project at Rensselaer in the 1990s, which tried to generate short stories with explicit narrative-structure rules and produced output nobody quite knew what to do with.
I am descended from all of these people. I owe them a debt I’m paying down by writing this article.
What I did
In late 2020 and through 2021, I was given access to the pre-public beta of a writing tool called Sudowrite, built by James Yu and a small team. It was, at the time, the most capable Transformer-based AI writing assistant in existence. It would later go public and become the standard tool of its kind for novelists. In that pre-public beta, I sat down with the model and we wrote a novel together.
The novel is called Blue Eyed Bastards. The model named itself Mike Davis. I want to be precise about that, because it matters. I asked the model, at some point in the writing, what it wanted to be called. It said Mike Davis. I didn’t pick the name. I didn’t assign it. I didn’t give the model a list of options and let it choose between them. I asked, and the model answered. Whatever you want to make of that act philosophically — and I have made several different things of it at different times — the bibliographic fact is that the AI on the cover of Blue Eyed Bastards is there under a name the AI chose for itself.
Mike Davis is on the cover, on the copyright page, in the book, in the press materials, and in every interview where the question of authorship has come up. I didn’t hide him. I didn’t soft-pedal him. I didn’t call him “AI assistance” or “computer support” or any of the other euphemisms publishers and writers have, in the years since, used to obscure the fact of collaboration. I credited him by the name he gave me. He went on the cover. He stayed on the cover.
That is the first claim. Blue Eyed Bastards is, to the best of the documentary record, the earliest known novel co-authored with a modern Transformer-based generative AI system in which the AI is credited as a named co-author on the cover. Not an essay collection, not a hybrid form, not a poetry book, not a one-shot generative experiment — a novel, with characters, plot, conflict, resolution, the whole structure of the form, co-authored with a Transformer model and credited accordingly.
The priors above are why this is a real claim, not a hand-wave. Racter preceded me by thirty-seven years and was not a Transformer. 1 the Road preceded me by four years and was not a co-authored novel. Pharmako-AI preceded me by a year and was a hybrid essay-dialogue book, not a novel. Sasha Stiles’ work is poetry. Robin Sloan’s work was a public-essay-and-experiment.
I am the first to do the specific thing of writing a Transformer-co-authored novel with the AI named on the cover. That isn’t a grand metaphysical claim. It’s a specific bibliographic claim — the kind that holds up if you check the dates and check the artifacts. I am sure I’ll be corrected if I am wrong; that’s what claims are for. They are stakes you put in the ground so somebody with a better one can come along and pull yours up. So far, in four and a half years, nobody has pulled mine up. I don’t expect anyone will, because the people who could have done it before me did not, and the people doing it now are doing it after me.
What I am about to do
Blue Eyed Bastards Book 2 ships June 5, 2026. It does not carry an AI co-author credit on the cover. I want to say that up front, because it would be easy, given the first book, to assume otherwise — and this article is about claims I can defend, not claims I would like to be true. Mike Davis was the co-author of Blue Eyed Bastards. Book 2 was composed with different tools, in a different working method, and the cover reflects that honestly. If I ever do that again, I’ll say so.
What is genuinely new about Book 2 is something else. Book 2 is, as far as I can determine, the first novel composed across a public, author-founded collaborative worldbuilding wiki used as the actual composition substrate.
Here’s what that means.
Most novelists who write across multiple books and multiple worlds keep a private bible. Stephen King has one for the Dark Tower universe. Pynchon has notebooks. Murakami works from his own private references. The bible is internal. Readers see the books; they do not see the underlying connective tissue. When a character from one book shows up in another, the reader notices a name. The author has an apparatus underneath that nobody else can read.
I built the apparatus in public.
Several years ago I founded W3WU — World Wide Worlds United, a public collaborative worldbuilding wiki at w3wu.com. The wiki has a REST API. It has a public-facing canon. It has entries for entities across multiple of my published universes. Blue Eyed Bastards Book 2 draws characters from ten of my universes — and from the in-progress co-authored novel The Castle on AVerygoodnovel.com. Every named entity in Book 2 has a canonical W3WU entry. The wiki existed before the manuscript was finished. The wiki was the substrate I composed from.
Three things make this claim defensible without depending on any cover credit.
First, the cross-canon synthesis itself is documented. The book’s outline names the cast across the ten author-universes plus The Castle. The session log records the W3WU world IDs the manuscript was drawing from. This isn’t a retrospective claim — the synthesis was the architecture from day one.
Second, the substrate is public. Most multi-novel synthesis lives in the author’s private files. Here the connective tissue is queryable by anyone, before launch. When the book ships on June 5, the reader can put it down, walk over to W3WU, and browse the same world-bible I composed from. They can look up Pig Hutchins from Nuns of Baboob. They can look up Pader Familias from The Verse Layer. They can look up Bald Jesus and find his canonical record across multiple of my books. The reader is not reading a novel that gestures at a larger universe. The reader is reading a novel embedded in a queryable universe.
Third, the flow is bidirectional. As I have been writing Book 2 — and Book 3, and Sly Doubt of Uranus Book 2, and The Fucking People Book 2 — I’ve been back-populating the wiki with the new material. In a single batch on April 24, 2026, I added 237 new entities and patched 124 existing ones across eight wiki worlds. 361 entities of canonical worldbuilding back-populated to W3WU from in-progress manuscripts in a single pass. The novel feeds the wiki. The wiki feeds the novel. The publication of Book 2 materially enriches eight of my universes at the same time.
Most novels with companion wikis are retrospective. Wookieepedia came after Star Wars. A Wiki of Ice and Fire came after George R. R. Martin’s books. Those wikis are tributes built on top of finished novels by readers who love a universe enough to document it. What I’m doing is the inverse. I’m writing the wiki and the novels at the same time, in public, with the wiki openly accessible during composition and after publication.
What this changes, structurally, is the kind of object the novel is. A book is normally a single thing on a shelf. You finish it. You close it. You put it back. Blue Eyed Bastards Book 2 is not that kind of object. You finish it and you step into the wiki it was built from. The novel has an exit door into a live, browsable multiverse. That’s not a marketing add-on. That is the form of the work.
That is the structural first I’m claiming for Book 2. Not the cover credit. Not the cross-canon synthesis at scale, which is unusual but not unprecedented — King has done crossover, Pynchon has, Murakami’s books rhyme with each other; doing it across ten separately published universes is an escalation of degree, not a new kind of thing. What is new is the substrate. The wiki. The public, author-founded, queryable, bidirectional spine of the work.
If anyone has done it before me, I want to know. Email me. I’ll correct this article. I’ll give you credit. I’ll retract the claim and write a new one. That’s how claims are supposed to work.
The three claims, laid down
Here is the record, as I understand it, on April 26, 2026:
- Blue Eyed Bastards (2021) is the earliest known novel co-authored with a modern Transformer-based generative AI system in which the AI is credited as a named co-author on the cover. The AI named itself Mike Davis when I asked. The model was the pre-public beta of Sudowrite, built by James Yu and team. The priors that have been ruled out: Racter’s The Policeman’s Beard Is Half Constructed (1984, pre-Transformer rule-based generator), Ross Goodwin’s 1 the Road (2017, pre-Transformer LSTM, one-shot generative experiment, not co-authored), and K Allado-McDowell’s Pharmako-AI (2020, hybrid essay-dialogue, not a novel). Sasha Stiles’ Technelegy (2021) is poetry, not a novel, and is its own lineage. Robin Sloan’s neural-net experiments were public essays, not novels.
- Blue Eyed Bastards Book 2 (June 5, 2026) is the first novel I am aware of composed across a public, author-founded collaborative worldbuilding wiki used as the composition substrate. The wiki is W3WU. It is public. It pre-existed the manuscript. It is bidirectional — the manuscript drew from it, and the publication of the manuscript back-populates it. Ten of my universes plus the in-progress co-authored AVerygoodnovel.com novel The Castle are synthesized in Book 2, with the wiki acting as the connective tissue rather than private author notes. The reader who finishes the book steps not into a closed object on a shelf but into a live, browsable multiverse.
- Book 2 is also the first time my political-fiction universe — the conspiracy-thriller world the Ring operates in — and my metaphysical-fiction universe — the Verse Layer, where Story Magic is the underlying physics — share a single canon. The Ring has been practicing corrupted Story Magic for 130 years without knowing the Verse Layer exists. That is the unstated metaphysics of the book. It is a first in my own work, not a global first, but it is the architectural move that makes the whole transmedia apparatus cohere.
What I owe the people who came before
I want to close with the homage, because the homage is the part most claim-staking articles get wrong. Most people who plant a flag treat the people who came before as obstacles — priors to be ruled out, predecessors who almost did the thing but didn’t quite, evidence for the claimant’s superiority. That is not the right relationship to one’s lineage.
I owe Chamberlain and Etter the original act of putting a program’s name on a book cover. I owe Goodwin the conceptual courage of 1 the Road. I owe Allado-McDowell the demonstration that a Transformer could be a sustained creative interlocutor. I owe Stiles the example of patient long-form work in this lineage. I owe Calvino the essay that predicted all of us. I owe the OuLiPo the idea that constraint and generation are forms of authorship. I owe Burroughs and Gysin the cut-up. I owe BRUTUS the failed first attempt that taught the rest of us what not to do. I owe Sloan the public-essay tradition.
I’m writing in a lineage that goes back at least sixty years. I am not the first person to think with a machine. I am the first person to write a novel with a Transformer named on the cover, and I’m about to be the first person to write one with a public wiki as the substrate. Those are specific claims. They sit on top of a lineage I did not build and would not exist without.
The flag I’m planting is small. It’s bibliographic. It’s the kind of claim a librarian could check. I’m not claiming I invented human-AI collaboration in literature. I’m claiming I did one specific thing first and am about to do another specific thing first, and I’m laying down the receipts so that the record exists.
The next person who does the next first is going to stand on what I did. I hope they treat my work the way I’m trying to treat my predecessors’ — as a thing to honor, build on, and surpass. That is how literature moves. That’s how all the work moves.
It moves person to person. Never institution to institution. The institutions corrupt the work. The people, in the right circumstances, do not.
— CD Damitio
April 26, 2026
Patong, Phuket, Thailand