I never thought I would write a book. It wasn’t on a list of things I wanted to do; it wasn’t realistic. I’m dyslexic and dysgraphic, and the work of producing a hundred thousand words of prose — sitting at a desk picking word after word — was simply not a shape my life could take.
What my life did have, increasingly over the last few years, was dialogue with language models as a medium for creative and intellectual exploration. Historical roleplay as a lens for thinking through ideas I couldn’t see from my own perspective. Alternate history worldbuilding tested through counterfactual scenarios. Fiction as a way of working out philosophical questions that analytical prose couldn’t reach. Each project was its own thing. None of them were novels, and I wasn’t trying for novels; the tools couldn’t have supported that, and I wasn’t ready for it either.
Then a few things happened at once. The tools reached a point where they could hold sustained context across many conversations. The craft I had been developing — how to push on the dialogue, how to diagnose what was missing, how to refuse adequate output — reached a point where it could be applied at a scale it couldn’t before. And the thing that had not been a shape my life could take became, through a different route, something that was. A book got made. It was argued into existence through months of sustained dialogue — hundreds of conversations, and many times more words of argument and reference material than the finished book contains. It exists.
This essay is about what that method is, and what it isn’t, and why refusing to have a conversation about it is a mistake.
Writing is thinking made into language. What makes a book worth reading is not that somebody produced the sentences but that somebody thought, carefully and at length, about what the book is trying to do — and that the thinking reached language in a form that holds together, that rewards the reading, that says something. Writing is the activity in which thinking becomes language. It has traditionally happened through a single person thinking and producing sentences together, fused, inseparable. But the fusion is a feature of the method, not the activity. When thinking-made-into-language can happen in a different configuration, the activity is still writing.
What I am describing is a method in which the writing happens in the dialogue. Not: I did the thinking and the model did the writing. The writing — the sustained activity of thinking-made-into-language, extended across months, argued, shaped, refused and reshaped — was happening in the back-and-forth. The manuscript is what that activity left behind.
The discourse around AI and writing has not, in most hands, recognised this as a possibility. The assumption is that writing must happen at the sentences and only at the sentences. If the sentences came from a model, the argument goes, the writing didn’t happen. That assumption is where the interesting question lives.
Most AI-involved writing is slop.
I want this on the record before anything else, because what follows will otherwise be read as a defence of a category I have no interest in defending. People who type a prompt and publish the result are not writing books. The text they produce has no depth of creative intent or thinking behind it, because there was no sustained engagement in which creative intent or thinking could have lived. It is often also generically confident and structurally hollow — readers learn to recognise the surface — but the stylistic signature is downstream of the real problem, which is that no substantive engagement produced what was being attributed. The critics who say this isn’t writing are correct. If you’ve encountered people claiming authorship of such text, your contempt is well-placed. I share it.
“I wrote a novel with AI” is, in most mouths, a misleading claim. Often it describes a process in which the human contribution was a paragraph of instruction, and the result is something the claimant did not meaningfully make. The critics’ reaction to such claims is not paranoia. It is reasonable pattern-matching against a common and dishonest use of the phrase.
But there is a specific thing I want to defend. Not a category. One case. What I did, and what others for whom this describes what they’re doing might be doing too, however rare. I’m going to call this method arguing into existence. The phrase is imperfect, a placeholder, but it has the virtue of being specific, of pointing at a practice that doesn’t yet have settled language, and of having already been claimed on the book and in the companion essay, How the Book Was Made. When I say arguing into existence I mean my method. When I say AI-involved writing I mean the general field, which includes my method and includes the slop.
The two are not the same thing, and conflating them is part of why the discourse is stuck.
What I am asking is narrow: that arguing into existence, and work like it, not be thrown out with the slop on the grounds that they share the slop’s surface. The method is real. The work done by it can be genuine. Rejecting it out of hand means missing something that might turn out to be worth finding. That is what I would like the reader to consider, and it is the whole of the ask.
Every medium with a low barrier to entry has produced a flood at the floor.
Photography did. When cameras became cheap, the world filled with snapshots, most of them uninteresting or trading entirely on what they happened to be pointed at. Video did. TikTok is the current extreme: the floor is effectively zero, and the vast majority of what gets produced is slop or captured luck. Neither photography nor video became a lesser medium because of the floor dropping. What happened instead is that the culture around each medium slowly developed vocabulary for distinguishing snapshot from composition, pointed-camera from made-thing, luck from craft. Nobody now confuses a phone snapshot of a sunset with an Ansel Adams landscape, even though both are produced by the same basic action. The vocabulary came.
Notice what that vocabulary is doing. It is tracking where the thinking happened. An Ansel Adams landscape is not a great photograph because of anything mechanical that differs from a phone snapshot; it is a great photograph because of what Adams knew about light, what he was looking for, how long he waited, what he chose to frame and what he chose to leave out. None of that is visible on the print itself. The thinking happened upstream of the shutter, and the photograph is the residue. We are fluent in this. We do not look at an Adams print and ask whether he really made it, given that the camera did the optical work.
Prose is going through the same transition, later. Until recently, prose had a much higher floor than other media, because producing a hundred thousand words of prose was itself a filter. You couldn’t accidentally write a novel. The floor and the ceiling were closer together than in photography or video, and the culture never needed to develop fine-grained vocabulary for levels of work inside the medium — the medium itself sorted the levels. AI has collapsed the floor. Anyone can produce a hundred thousand words. The flood at the floor is real, and the critical reaction to it is accurate to what’s flooding.
But the reaction is happening before the vocabulary has caught up, and in the absence of vocabulary, AI was involved is being used as a flat marker for slop. The heuristic is functional on average. It fails in the cases this essay is about, for the same reason that the camera did the work fails as a criticism of Adams: it mistakes the surface of production for the site of the thinking.
I wrote this book. I am its author. I want to say this plainly before anything else, because the rest of what I’m going to say will otherwise be read as negotiation over the point, when it isn’t. The word author applies in its standard sense. So does writer. The book is mine in the way books are mine — I conceived it, I shaped it, I argued it into its specific form over months of sustained work, and my name belongs on the spine because I am the one responsible for the work existing and for its being what it is.
The rest of this essay is about two things that follow from that, and that should be kept distinct.
The first is why authorship of this kind — where the credited author did not execute every surface detail — is legitimate at all. The critic’s objection to AI-involved writing usually rests on an implicit rule: that authorship requires both holding the vision and executing the surface. If the sentences didn’t come from your hand, the argument goes, you cannot be the author. This rule is not a principle of authorship. It is a feature of specific solitary traditions, and plenty of creative practice contradicts it.
The second is how to describe what I did, once the authorship is settled. The word author is correct, but its conventional image — someone sitting at a desk producing sentences — is not what arguing into existence consists of. Prose does not yet have developed language for the method. Other creative practices do. The directorial pattern, in particular, captures the structure with a precision that nothing in prose’s existing vocabulary matches. Describing the method that way is worth doing — not to replace the word author but to make honest what the authorship actually involved.
Divided-labor authorship is not new. Across creative practice, whenever vision and execution have been separable, the pattern has been consistent: authorship locates at the frame around the process, where the vision is held and the judgment exercised, not at every point where execution happens.
A director does not operate the camera, act in the scenes, or build the sets. The film is theirs. A conductor plays no instrument during the performance. The performance is theirs. An architect does not lay bricks. The building is theirs. The Greek choragos, who funded and produced a dramatic festival’s plays, was credited in ways we would now call authorial even though he wrote and performed nothing. In every one of these cases, the authorship is not diminished by the distribution of execution. It is understood, by anyone familiar with the practice, that a work of this kind involves many hands and that authorship locates at the frame, not at every hand.
No one thinks Wes Anderson isn’t really the director because he doesn’t operate the camera. No one thinks Renzo Piano isn’t really the architect because he doesn’t lay bricks. The concept of frame-authorship is not controversial in any of these practices. It is the settled way creative traditions with separated vision and execution work, and it has been for a long time.
Prose, by contrast, has mostly been a solitary tradition — for mechanical reasons rather than principled ones. Producing long prose required someone to produce the sentences, and the physical act of writing was slow enough that sustained collaboration during the making wasn’t practical. The writer had to do both jobs because there was no way to separate them at scale. The fusion was imposed by the medium, not required by what authorship is.
Even within prose, the conjunction-requirement has never been absolute. Prose has its own traditions of divided-labor authorship: ghostwriting, substantive editorial collaboration, speechwriting, co-authorship with asymmetric contributions, academic writing with its conventions of first authorship and contribution disclosure. In each case, the credited author is considered the author despite not having produced every sentence. Prose already accommodates divided labor. It just does so more narrowly and more quietly than media whose division of labor is structural rather than occasional.
The critic’s implicit rule is therefore contradicted in enough places that anyone relying on it owes an explanation of why AI-involved writing should be the exception. Directorial authorship is accepted. Architectural authorship is accepted. Ghostwritten memoirs are credited to their named authors. Academic papers list the people responsible for the work and specify what each did. Any position that allows all of these but rejects arguing into existence needs to name what distinguishes this case. The rule, on examination, is not a principle. It’s a default that has never been defended because solitary prose has never needed to defend it.
I am the author of this book. That claim rests on the pattern that governs divided-labor authorship everywhere it has been worked out: credit goes to the person who held the vision, shaped the conditions, and exercised the judgment that made the work what it is. That person is me. The fact that the sentences came from a language model changes this no more than the fact that a ghostwriter’s sentences come from the ghost.
Once the attribution is settled, a different question opens: what was the method, and how should it be described?
Prose’s existing vocabulary for divided-labor authorship doesn’t describe it very well. Ghostwriting is the nearest prose practice, and it is structurally different in an important way. A ghostwriter’s collaboration with their client is typically asynchronous: the client briefs, waits, receives a draft, gives notes, waits again. The ghost does most of the creative work alone, making countless decisions without the client present. The client evaluates a largely-finished product at intervals. The division of labor is real, but the temporal structure is one of handoff and review, not of continuous shared presence.
Arguing into existence is not this. The method has the temporal structure of directorial work, and the directorial parallel is worth developing carefully, because it captures something ghostwriting misses. I’ve mentioned directors already as one example of the pattern of divided-labor authorship. Here I want to develop the example further, because the specific structure of directorial work matches arguing into existence more closely than the other parallels do.
A director’s authorial presence runs through a multi-phase process. Before anything is shot, there is casting — not a solo decision but a collaboration with casting directors, agents, and producers, iterated over weeks, with the director ultimately making or approving each choice. Script development, set design, costume, lighting, location scouting — each of these is its own sustained collaboration, with its own specialists, its own timelines, its own feedback loops. The director is present through all of it, not executing the specialist work but holding the vision against which each specialist’s contributions are evaluated and shaped. By the time shooting begins, the director has already made thousands of decisions — not alone, but in continuous exchange with the people whose craft produces the specifics the director chooses among.
Then shooting begins, and the pace changes but the shape doesn’t. A take on set is a dense moment: the actor performs, the camera rolls, the director watches. But the take is happening because of everything upstream. Because the script was developed a certain way. Because these particular actors were cast. Because the set was built like this. Because the cinematographer proposed this framing this morning and the director agreed. Because this is the third take of this line and the first two revealed something about the rhythm that needed adjusting. The take isn’t where the direction happens — the direction is what made the take possible. The direction continues through the take, as the director watches and decides whether to let the actor’s unexpected reading stand or to call for another attempt. And it continues after — in dailies reviewed that evening, in discussions with the editor weeks later, in the cut where what survives from which take is decided.
At no point is the director working alone. At no point is the director’s presence absent. The collaboration shifts in collaborators and tempo — fast exchange on set, slower exchange in editing, different exchanges with different specialists throughout — but the authorial presence is continuous. And the work is iteratively emergent: the director doesn’t arrive knowing exactly what the film is. The film is discovered through the process, in response to what the actors find, what the cinematographer proposes, what the editor surfaces in the footage. The director holds the vision and exercises judgment. The specifics come from the collaboration.
This is the shape of arguing into existence. Not a phase of structural work followed by a phase of generation followed by a phase of editing, but one continuous dialogical process with shifting modes, with the authorial presence running through all of it. Some parts of the process were dense with structural argument and had almost no prose generation. Some were dense with generation and light on structure. Some were editing-heavy but still involved structural rework and new material. The modes interleaved constantly, and the balance shifted over the project’s duration, but the dialogue never stopped being the site where the writing was happening.
And like a film, the book was iteratively emergent. I did not arrive at the first conversation knowing what the book would contain. The question the book exists to answer — what does a Mediterranean look like over centuries when two great powers coexist and neither overcomes the other — was itself discovered through early argument, not brought to the dialogue as a premise. The frameworks that govern the world, the ten-layer analysis of why Rome stays off the sea, the mirror framework for why Carthage stays off the land, the in-world source tradition that determines which events are recorded and which are lost — none of these existed at the start. Each was found, through argument, at the point where the work revealed it was needed. Many were found after significant prose had already been generated, and the structural insight triggered rewrites of material that was already on the page. The book did not get drafted and then revised. It crystallized progressively from a solution that kept getting worked and reworked as the structure deepened. Large portions were rewritten repeatedly — not because the earlier versions were bad, but because each new structural insight changed what the earlier material had to be.
Some traditional writers work this way too; the density of restructuring is not unique. What’s distinctive is the interleaving — the way structure and generation and editing happened together throughout, in continuous dialogue, rather than as successive phases. That’s partly what made the book’s internal coherence possible. The structural logic could keep deepening because the prose could keep being remade against it.
This is the shape of directorial work, and of arguing into existence. Describing it that way is not a claim that I should be called a director. The word author fits. What the directorial frame offers is description of the practice — the shape of the work, where the authorial presence lives, how it runs through a multi-mode collaborative process with shifting tempo and emphasis. That description is not available in prose’s existing vocabulary, because prose has not, until recently, had a practice that needed it.
The ghostwriting comparison is worth returning to briefly, for a different reason. Beyond the structural mismatch — ghostwriting’s asynchrony versus the dialogical continuity of arguing into existence — there is an ethical question worth naming, but it has to be named precisely, because AI-involved writing spans a spectrum and the ethical question looks different at different points on it.
Ghostwriting has historically been a practice of concealment. The convention has usually been to hide the ghost, to present the credited author as the sole maker, to let readers form a relationship with an author who did less of the making than they assume. That concealment produces the unease that sometimes attends ghostwriting when it becomes visible — the feeling of having been misled.
AI-involved writing is at the point where similar patterns of concealment are forming. But the comparison to ghostwriting only fully applies to cases of real divided-labor authorship — cases where someone has done substantive authorial work with a model and is choosing whether to describe the method or hide it. For cases like arguing into existence, the analogy is direct: the authorship is real, and the question is whether to disclose the method or leave readers with a false image of how the work was made.
But most AI-involved writing isn’t that case. Most of it is slop, where the problem isn’t concealment of a collaborator — it’s that authorial claim is being made where the authorial engagement was minimal or absent. Transparency about method doesn’t fix this. Adding “this was produced by prompting a language model” to a piece of slop doesn’t make the attribution more honest, because the attribution itself was the misrepresentation.
And between these two cases — real divided-labor authorship and near-zero-engagement slop — there is a wide space of intermediate practices with thinner authorial involvement, where the honest description of what was done varies considerably. I’ve produced things myself where I was the initiator of something I found interesting, but where I wouldn’t feel right claiming authorship in the full sense that I claim it for this book. The honest responses to those cases are their own — not claiming authorship, or describing the engagement honestly rather than borrowing the conventions of solitary writing. The point is that honest description is the principle that varies with the case, not any universal disclosure formula.
For arguing into existence, the honest description means naming the method clearly. The cover of Mare Non Nostrum bears the Latin attribution Ex Dialogo cum Machina — from dialogue with a machine. The authorship note inside the book runs to several pages, describing how the work was made and naming the model I worked with. The attribution is partly method-description and partly a refusal to hide something I cannot fully resolve: a model was involved, I do not know what that ultimately means for the ontology of authorship, and the most honest thing I can do is point to the method clearly and leave the deeper questions open for readers and for the conversation to work out over time.
The closer model for this kind of disclosure is academic citation. Academic papers list all authors, often with a final section specifying who did what. The first author’s credit is meaningful — they led the work — but so is the disclosure of how the work was divided. This is not considered a dilution of authorship. It is ordinary professional practice. The book’s attribution is in that family. It names the method because naming the method is what responsibility requires, for this case.
This might not be obvious, because the current discourse often treats language-model use as skill-free — as if because the interface is natural language, using the tool well requires no more than being able to speak. This is already visibly wrong for technical use of these tools, where the gap between skilled and unskilled users is large and growing. It’s wrong here too, and probably more so. Language is already a skilled medium; working with a model through language inherits that skilledness rather than erasing it.
Arguing into existence is a craft. It has to be learned. Knowing what to push on, when to refuse, how to shape context, when to step back from prose generation and spend weeks on structural work, when to let the model run and when to interrupt, how to diagnose what is wrong at the right level of abstraction — none of this is automatic. Most people trying to write with AI are not doing this. The outputs show it. The distinction between arguing into existence and casual AI use is as large as the distinction between any skilled craft and its untrained equivalent.
Mare Non Nostrum was not my first sustained creative work with a model. Over roughly two years before I began the book, I had been developing the practice through smaller projects across different genres: the historical immersions and alternate history worldbuilding and philosophical fiction I mentioned at the opening. What these shared, and what built the craft, was something specific — they were play. Not in the dismissive sense, but in the sense of sustained open-ended engagement driven by curiosity rather than deliverable. I’ve used language models heavily for years for other purposes too, including in technical work, but that use didn’t build this craft. What did was dozens of projects of unstructured creative and intellectual exploration, over time, where the goal was the exploration itself. Previous projects had been mostly constrained to single conversations by the technical limits of the tools available. Mare Non Nostrum was the first project I undertook where the craft could be applied at book scale, across hundreds of conversations with shared context — made possible only recently. The craft came first. The book is what that craft produced when the substrate finally supported applying it at scale.
This is worth saying because the critic’s image of AI-involved writing is formed mostly by what happens when someone without the craft uses the tool casually. That is slop. What I am describing is something else — a sustained deliberate practice, developed over time, applied with increasing ambition. The tool and the practice are different things. The tool exists. The practice exists. They are not the same, and conflating them is part of why the current discourse is stuck.
What I did, specifically, across the project:
I held the vision. Not at the level of “a book about Carthage winning the Punic Wars” — that is a premise — but at the level of the structural question the book exists to answer. That question, and the frameworks the work needed, lived in me in the sense that I was the one diagnosing when the work was incoherent, when a gap had opened, when something had to be worked out before anything else could proceed. The frameworks themselves emerged through argument — they were not pre-formed and imposed — but the diagnostic judgment that determined when and where a framework was needed was mine throughout.
I shaped the conditions. By the time the model generated any prose, the space of possible prose had been narrowed severely — by frameworks that had been argued into place, by context loaded specifically for this passage, by prior dialogue about what this chapter had to do. What the model produced inside that space was not the model writing freely; it was the model operating against conditions that were continuously mine.
I argued with what it produced. The dialogue is where the writing happened — not in the sentences the dialogue returned to me, but in the argument that shaped what sentences could come back. Roughly one and a half million words of argument, reference material, and worked-through reasoning produced a hundred and twenty-five thousand words of manuscript. The ratio is not incidental. The book is the residue of the writing — residue in the sense of precipitate rather than byproduct. What formed. What the process produced and left behind.
I refused. Much of the method’s discipline is in refusal — in the capacity to say not yet when the output could plausibly pass. Quality lives in the refusal, and the refusal requires knowing what right looks like before you’ve seen it. That knowledge is not delegable.
Some creative decisions in the finished book are the model’s. A specific phrasing, a choice of image, a way of approaching a scene I did not specify but allowed to stand. At the sentence level the proportion is substantial. At the level of structure, implication, causal logic, world — the levels that make the book what it is — the proportion is near zero. The model did not choose what the book was about. I chose. Through continuous argument, inside frameworks that were being built and revised the whole way, over the full duration of the project.
The obvious objection: if you could have written the book yourself, why didn’t you?
Two versions of this are worth separating. The first is practical: if the method is that demanding, why not spend the effort writing in the normal way? The second is sharper: the skills of structural imagination and sentence production aren’t really separable; most writers hold both; the reason you used the method is that you weren’t willing to do the work of holding both. Stop being lazy and write the book.
The sharper version rests on a premise worth examining. It assumes that the effort expended on producing sentences is itself what makes writing legitimate, such that a method that reduces that effort is illegitimate regardless of the outcome. This is a strange premise once stated clearly.
The closest parallels are not assistive tools like word processors or spell-check — those make the labour of writing sentences easier, but they do not bypass it, and a critic who thinks sentence-production is the only legitimate site of writing can accept them without inconsistency. The closer parallels are in creative media where new technology restructured rather than assisted the work. Digital photography did not help film photographers work in darkrooms; it replaced the darkroom with different equipment and a different practice. Synthesizers did not help orchestral composers write for violins; they opened up making music without violins. Computer-aided design did not help architects draft by hand; it restructured what architectural drafting was. In each case, the old labour was not assisted but rearranged into something different, and in each case the charge was the same — that’s not real photography, not real composition, not real architecture — and in each case the charge has mostly given way to the recognition that different methods produce different work, and that holding practitioners to the old labour as a moral requirement is gatekeeping rather than craft.
In every creative field, we evaluate work by what it is, not by how hard it was to make. We don’t ask filmmakers how many hours each shot took and adjust our opinion accordingly. We don’t ask composers whether they wrote the score by hand or in software. We don’t ask architects how many drafts they went through. We ask whether the film is good, whether the music moves us, whether the building works. The work is the test. The method is a matter of what suited the work.
And attention is finite. Time is finite. Interest is finite — and this last is the one most often forgotten. The interest that sustained me through months of structural argument would not have sustained me through the years of sentence-by-sentence production that traditional writing, for me, would have required. Interest is a resource. It depletes. It cannot be willed back into existence. For me, the choice was never between arguing into existence and the same book produced differently. The choice was between this method and the book not existing. Calling that choice lazy requires imagining a version of me with infinite time and interest, and the charge only lands in that imagined world.
There is also what happens after. The time the method didn’t demand is not theoretical. Since the book was finished, I’ve been exploring further creative work in the context of the alternate history it established — but new kinds of writing, new modes of expression, new structural considerations. Not a sequel. Different work, developing the craft in different directions, thinking through things the book opened but didn’t pursue. This would not be happening if I had spent two years on sentence-craft for the first book. The method didn’t save me from work; it freed the work that would otherwise have been displaced by sentence-craft. The creative engagement continues, at different scales, in different modes, building on what the book made possible.
And the bottom line remains: without arguing into existence, I would not have written this book. Not a smaller version. Not a simpler one. No version. The book does not exist in the counterfactual. The question why not just write it yourself has no better answer than the question has content — in the world where I didn’t use this method, there is no book to be written about.
Right now, I could be struggling through sentence generation and continuing to do so for another eighteen months, to produce this book in the traditional way. Or I could be pushing my own learning, exploring new ideas, grappling with questions this book opened, developing the craft in new directions. I unapologetically choose the latter.
There is a separate argument that asks why not employ a human collaborator — a ghostwriter, a co-author. That argument is about labour markets and the ethics of AI, not about authorship. This essay is not addressing it. It is enough to say that arguing into existence exists, that it produced a book that exists, and that the question of what the practice amounts to is not settled by whether some other practice should have been used instead.
Readers will identify this essay as AI-written. They will be right — and specifically right in a sense that matters. Certain rhythms, certain turns of phrase, certain punctuations have become reliable signals of AI involvement, and readers have learned to spot them. The heuristic is mostly functional. Most of the time, when the markers are there, the text is hollow.
This essay has some of those markers. I did not strip them out; stripping them out would have been a different kind of work than the work I was doing. What the markers indicate, correctly, is that a model produced the sentences. What they do not indicate is the thinking behind the sentences. This essay was argued into existence, the same way the book was — through hours of sustained dialogue producing tens of thousands of words, out of which this is the residue. It is AI-written in the sense that the surface came from the model. It is not AI-written in the sense most readers mean when they use that phrase to dismiss a piece of writing. The difference between this and a thought-leadership prompt is the difference between any craft and its untrained analog.
What I can offer is this: I have put effort, across the book and across this essay, into ensuring that if you look past the markers and take the work seriously, there is real depth underneath. Real thought, real coherence, real specificity, real care about what the work is trying to do. You may not find what you find to your taste. You may not agree with it. You may find the subject matter uninteresting or the style not what you would have chosen. What you will not find, I hope, is emptiness. That is what I can commit to. The rest is for readers to judge.
The markers are largely artifacts of thin prompting — what the model does when given little context, no sustained voice, no pressure against its default rhythms. Under those conditions the outputs converge toward the generic, because generic is what the expected completion looks like when little has been specified. Under sustained dialogical pressure, the outputs can be shaped considerably. The markers will often still appear, because some of them are deep in how the model writes, but they sit inside work that the generic prior doesn’t describe. The heuristic catches them. The heuristic does not catch what’s underneath.
Some critics will hold a stronger position: that AI-involved writing cannot be good, as a categorical matter. This is an extraordinary claim, and the extraordinariness is worth sitting with.
It treats AI differently from every other tool and every other form of collaboration ever introduced into creative practice. Every previous new tool produced the same reaction — photography can’t be art, recorded music can’t have soul, cinema is mere spectacle, word processors will ruin prose — and every such reaction looks embarrassing in retrospect. But the claim goes further than technology. No form of collaboration has ever been held to make good work impossible. Directors collaborate with actors and cinematographers and editors; good films exist. Architects collaborate with engineers and builders; good buildings exist. Composers collaborate with performers. Playwrights collaborate with directors. Even contested cases like ghostwriting have had charges of inauthenticity or concealment, not of categorical impossibility. The claim that AI-involved writing cannot be good puts this kind of collaboration in a category by itself — uniquely corrupting, in a way no other tool or collaborator has ever been.
And it runs against the actual history of creative expression. Humans take new media and turn them into canvases. The affordances of any new medium are different — sometimes sharply different — from the old ones, and new craft has to develop within those affordances. But the idea that humans will fundamentally fail to find creative expression within a given set of affordances denies something we know is true about how creativity works. Humans are insanely good at figuring out how to make anything into a canvas. Some canvases are harder than others; many people fail; the default is almost always bad; mastery is rare. None of this is the same as impossibility. The claim that this medium is the one where the pattern breaks — where humans will never figure out how to do genuine creative work within its affordances — is not supported by any evidence and is contradicted by most of human creative history.
The charge is usually made as an argument from incredulity: I can’t imagine how this could produce good work, therefore it can’t. Arguments from incredulity have a long history of aging badly.
A sharper version of the charge points to a real threshold problem. If the claim is that good work becomes impossible at some point along the spectrum of AI involvement — from spell-check through suggested phrasing through paraphrase through full generation — then there has to be a threshold where good becomes impossible. But no one can locate this threshold, and the failure to locate it is the point. The objection is usually a statistical tendency dressed as a categorical rule. Most heavy AI involvement produces bad work is reasonable and probably true. Heavy AI involvement necessarily produces bad work is much stronger, and the difference between the two matters.
The question of whether good work can come through this kind of method is empirical. It will be answered by specific works, specifically read. The filter in use right now makes that reading less likely. What the filter misses is what may eventually demonstrate that the floor is not the whole medium.
I have tried to name what I did. Arguing into existence is a gesture. Ex Dialogo cum Machina is a gesture. Both are placeholders — pointing at something I don’t yet have words for, kept visible because placeholders are better than silence. The category has fuzzy edges. The practice is, I suspect, rare. The work produced is hard to distinguish from the flood at the floor without looking carefully. None of these is a reason to stop trying to name it.
What I am asking, precisely, is this. Authorship of the work is not in question; the word author covers the case in any reasonable extension of its existing uses, and the patterns that govern divided-labor authorship across creative practice make this clear. What I am asking to be held open — not settled — is the description of the method. Prose doesn’t yet have developed language for what I did. Collaborative traditions in other media do, built around directorial and conductor-like practice, and the structural match is good. New language may emerge. Existing language may extend. The shape of the vocabulary will settle through use, over time, as more people do this kind of work and more readers encounter it.
The critics have an easier time of it. They have a ready-made sentence: you didn’t write it. That sentence does a lot of work quickly, and its availability is part of why the discourse is stuck. But the sentence collapses a set of distinct questions — who is the author, what was the method, how does this relate to other forms of collaboration — into a single verdict, and the verdict depends on an assumption that nothing outside prose’s recent tradition has ever required: that sentence-level production is the site of authorship, and the site of writing.
The book exists. This essay exists. Both were argued into existence, through sustained writing that happened in dialogue rather than in the sentences the dialogue produced. Both are instances of the practice they describe. If what you are reading now seems to you to do real work, then the writing did real work in producing it. If it seems hollow, the writing failed at this task, and you can judge accordingly. That is the only honest test I can offer. The writing is not hidden; it is in a place the usual inspection doesn’t look. The artifacts are there. Look at them, or don’t.
That is what I am asking for. Not agreement. Not a verdict on the book. Not a change of view about AI-involved writing in general. Just the willingness not to throw this case out with the slop on the grounds that it shares the slop’s surface.
Coda
One further thing, taking up the thread I started with at the opening.
Traditional prose writing is a filter. It selects for a particular combination of skills: structural imagination, yes, but also sentence production under sustained load, across many thousands of words, without breaking. These are different skills. They have historically had to be exercised together because there was no way to have the first without going through the second. The filter has therefore excluded, throughout the history of prose writing, a class of people whose structural imagination was strong but whose sentence-production channel was narrow — who could see the book, hold the logic, imagine the world, but could not get the sentences out at scale.
I mentioned the dyslexia and dysgraphia at the start to explain why a book had not been a shape my life could take. But the same observation generalises: the filter is not load-bearing. It is incidental. It is a feature of a method that had no alternative, not a feature of what prose is for. Rejecting arguing into existence means choosing to keep the filter, now that we can see it was never necessary. That choice is legitimate — it is possible to argue that the sentence-production struggle is part of what makes prose prose, and that a book produced without that struggle is a different kind of object. It is an argument worth having. But it should be had honestly, with the cost of the position visible: a class of people have been excluded from prose writing not because they had nothing to say but because the channel to the page was narrow. Some of them now have a route. Whether to accept that route into the conversation about authorship is a choice worth examining.
The main argument holds whether or not this coda is granted. But this is also true, and worth saying out loud.