This is not a plea for more sensory detail. Sensory description can sit on the page for paragraphs without ever reaching a character. Consider the opening of Anna Karenina.

Stepan Oblonsky wakes, reaches in the dark for his dressing gown in the place where it is kept, and only then remembers he is no longer sleeping beside his wife. The forgetting is in the hand before it is in the mind. The hand has already arranged itself around a life that no longer exists, and the shock of the scene is that the arrangement arrives a second earlier than the thought. This is what embodied narration can do, and what pure interior monologue cannot.

Thought arrives inseparable from sensation, posture, breath and impulse. The character is not a mind with a body attached. They are a mind that only ever arrives through a body, and can only ever be read there.

The interest of all this lies in the small betrayals that embodiment keeps producing. The body discloses what speech and thought will not, and often knows things a little earlier than the mind that is supposed to govern it.

The principle can be tested at the level of point of view, which is where writers most often get into trouble with it. First person promises direct interiority, the illusion that the reader has been placed inside another mind, but the risk is that voice begins to masquerade as presence. A narrator can talk fluently for pages and never actually be inside a body at all, only a stream of opinions issued by nobody in particular, and the reader, however entertained, will not bond with the narration because there is no one there to bond with. The writing bonds only when thought collides with flesh.

An instructive example is Nabokov’s Humbert Humbert, who insists on his composure, his learnedness and his elegance, and whose compulsions keep leaking out through bodily detail anyway. A controlled voice needs pressure from the body if the reader is to feel the person beneath the performance. A fire is said to be raging in his veins. A toothache is invented to cover the breaks in his patter. The physical undoes the rhetoric. The reader does not find the truth of him in the declarations; they find it in the fissures his language keeps trying to close.

Third person carries a different danger, which is that at its most distanced it can lapse into diagnostic summary, a narrating voice placed above the scene, telling the reader what to think about people the prose has not actually bothered to inhabit. Alice Munro is the standing rebuke to that tendency. The opening of Hateship, Friendship, Courtship, Loveship, Marriage describes a woman at a station counter whose “teeth were crowded to the front of her mouth as if they were ready for an argument.” The sentence registers a readiness the character has not yet spoken, and may not know she holds. The prose does not tell us she is combative. It lets the arrangement of her teeth do it. Across Munro’s work the method is consistent. She almost never uses italicised interior monologue or explanatory tags, and hesitation, desire and shame are given through the body instead, through a pause in a doorway, a tightened grip, a glance that goes somewhere it should not go. Intimacy here is not a matter of grammatical person, but of pressure.

Ishiguro is useful here because repression in The Remains of the Day is not only an idea Stevens holds about dignity. It is a physical discipline. A management of posture, tempo, gesture, and delay. The tragedy is not that he cannot say what he feels. It is that his whole body has been trained not to arrive at feeling.

For a working writer this stops being theory and becomes a test in revision. What is this character’s physical baseline, and how does pressure alter it? When they speak or think, what is the body doing at the same moment, and does it betray or amplify the words? Where a draft says she was anxious, can anxiety be rendered instead as the specific thing the body does, a held breath let out too late, a hand that keeps returning to the same button, a jaw that will not unlock?

A character becomes believable when thought no longer appears separate from the body carrying it, when feeling arrives as posture, impulse, hesitation, pressure. Good prose does not explain the person first and embody them afterwards. It gives the reader a body, and lets the mind appear inside it.

Something has decided what matters. What can be looked at directly. What has to be approached sideways. What deserves a sentence and what can be left unsaid. The events, gestures and dialogue belong to figures on the page, but what lingers is the presence of a particular mind. Call it the ghost.

It first struck me in my late teens, reading short fiction. The short story is the form where authorial choices are most visible, because there is so little room for anything else. Every sentence has to earn its place, which means every sentence betrays the writer’s hand.

Chekhov in particular affected me. The stillness of his stories, the scenes where nothing obvious happens, reveal a pressure in the subtext, suffusing the smallest gesture with meaning. A man watches a woman from a hotel balcony. A doctor sits down to a meal he will not finish. The surface stays quiet. Something behind the surface keeps arranging the quiet. The selection of what is shown and what is withheld does most of the work. Over time, that selecting begins to feel like a presence.

The most recent time I had this experience was when I read James Baldwin’s Another Country. Baldwin writes about shame and desire and the gap between who people believe they are and how they actually behave, and the writing gives you an unusually direct line to how he sees. You are inside the lives of the characters and inside the structure of his attention at the same time. The sentences move with a patience that refuses to let anyone, including the reader, off the hook. His prose looks steadily at the places most prose looks away from, and sets down language that does not soften what it finds. The effect on the reader is not admiration of insight but proximity to it.

The ghost is not the author’s opinions, or the person they would be across a table. It is the felt pattern of choices that makes a text seem governed by a particular mind. The rhythm, the angle of vision, the things the prose can bear to look at, the things it quietly declines. It is not the person behind the book. It is the shape a person’s selections take once they have hardened into sentences. You cannot meet it anywhere except on the page.

Even so, it exerts pressure everywhere. It inclines what a character notices, what they omit, when silence falls, when words break free. Sometimes it presses close. Sometimes it fades so completely that you only know you have been haunted when you close the book.

The ghost is built from technical decisions on the writer’s side. It is made through the distance the writer chooses to stand from their characters’ minds, through the shape of the sentences and through the direction of their attention. The reader experiences it as atmosphere, but it is produced by craft.

Raymond Carver is often held up as a master of brutal minimalism, a style where selection does everything. For a long time the ghost in Carver was thought to be Carver. Then the manuscripts came out, and a great deal of that ghost turned out to be Gordon Lish, his editor, cutting Carver’s stories down by half, sometimes changing the endings, stripping the sentiment until a much barer pattern remained.

Reading the two versions side by side is uncomfortable. The expansive, more openly tender drafts sound like one mind. The published stories sound like another. Both are Carver, in the sense that he wrote the words from which the published stories were made. Only one is the Carver we think we know.

The honest reading is that the pattern on the page was produced by two people. This does not dissolve the idea of the ghost. It sharpens it. The ghost is the pattern of selection, not the biography of the selector. When the pattern is consistent enough, the reader feels a single presence, regardless of how many hands shaped it. Most books have been touched by editors, partners, early readers. The difference with Carver is only that the editor’s choices became more visible than his own.

So, the ghost should not be confused with the author. A book may be intimate without being autobiographical. It may feel personally governed even when the governing intelligence is partly collaborative, revised, edited, resisted, or imposed. What matters is not whether the text gives us unmediated access to the author. It never does. What matters is whether the choices on the page become coherent enough to feel like a mind.

Some ghosts are recognisable because they are narrow. That is not a weakness. Partiality is often where style comes from. A writer returns to certain emotional pressures, certain types of spaces, certain failures of self-knowledge. Tolstoy’s ghost is recognisable because it is wide. That is a harder trick.

The usual compliment paid to Tolstoy is that he writes with the voice of God. Taken seriously, the phrase means something precise. It does not mean he writes with grandeur, or authority, or thunder from above. It means that his attention seems almost without hierarchy. When Tolstoy enters a room, he enters the whole room: the husband, the wife, the servant at the door, the dog under the table, the quality of the light, the small social calculation being made behind the smile. Nothing and no one is reduced to function. A minor character in Tolstoy can have an interior density close to that of a major one. The attention feels level.

Of course he is choosing. He has to be. Every inclusion is a choice, every line of a servant’s thought is a choice, and the level attention is itself a style, not an absence of style. What makes the compliment useful is how it highlights that his style consists of a refusal to rank. A peasant’s morning is no less worthy of attention than a prince’s evening. The ghost that writes Anna Karenina or War and Peace is the ghost that cannot see why one human interior would matter less than another.

That is not omniscience in the technical sense. It is the appearance of moral and perceptual abundance, so evenly distributed that it reads as something larger than a writer working.

Most ghosts have obsessions, blind spots, preferences, and methods they keep returning to. That partiality is often what lets us recognise them. Tolstoy is recognisable because his attention feels unusually capacious. His signature is made not from exclusion, but from fairness. You recognise Tolstoy’s writing almost immediately because of the evenness of the gaze, the willingness to stop and look at the thing most writers would walk past.

None of this is something a writer can step outside of. Every word chosen, every silence kept, every glance toward the ceiling while the lovers quarrel, every sentence that hurries or refuses to hurry, gives the presence away. Style is not decoration added to perception. It is perception made visible. To read closely is to learn its habits. To write carefully is to become less ignorant of your own. The ghost cannot be banished. It can only be shaped.

I had no formal training in psychology or linguistics or any of the fields I kept reading my way into. I read them alongside the writers who were shaping my fiction, pulled in both directions at once. But whenever I tried to write about the abstract ideas that interested me most, I kept hitting the same wall. Three paragraphs in, I’d realise I was wading into water where other people had spent entire careers. I’d follow one question, find ten books trying to answer it, read half of one, and lose the thread of where I’d started. It wasn’t that I had nothing to say. It was that what I was saying was too compressed, too intuitive, and too under-argued.

I’d also been interested in writing for film and theatre, but screen and stage plays felt restrictive to me. They were too focused on plot and action. Writing dialogue and directions didn’t let me get close enough to the internal reality I wanted to put on the page. When I started writing fiction, everything opened up. I could write thoughts and ideas as well as dialogue and exposition. I got to be a writer, director, actor, and cameraman. A novel is a finished sign system: a small universe built from the inside.

So I moved towards fiction. It suited me because, if you want to, it lets you invent as you go. You don’t have to stop every few sentences and check whether you’ve earned the next one. But you do have to learn your craft. My early fiction was almost incomprehensible. Without the discipline I slowly developed, I tended towards long abstractions in which the characters and events barely featured at all. I learnt restraint, found a voice, and had a couple of novels published. The observations on life and the sprawling abstractions were still being written, but they were cut and edited away. They stayed on my side of the desk.

I’ve recently realised that my problem with trying to write up those abstractions in my late teens and early twenties was not simply that I lacked training. It was that my mind is drawn less to specialism than to pattern-making: moving across fields, noticing links, and applying creative structures to knowledge. That kind of thinking can produce interesting insights, but on its own it isn’t disciplined enough to become good non-fiction. For a long time, that was where I stopped.

Large language models may have changed that. I can now set multiple research threads running at once and keep a live flow of ideas and context in front of me. I can say: I want to argue this. Does it hold up? What am I missing? What would be obvious to a specialist that I haven’t considered? And something comes back that I can test, challenge and refine. The model doesn’t give me authority. It gives me resistance. It pushes back against my ignorance in real time, which means I can keep moving without pretending to know what I don’t know. It becomes a kind of real-time Socratic dialogue: one of my favourite forms to read, except now I get to be one of the voices.

That resistance has let me build projects the way I build a novel: constructing a system of meaning, layering themes, shaping an argument so that it carries both intellectual weight and narrative force, while subjecting it to a level of scrutiny I could never sustain on my own.

Over the last couple of years I’ve assembled book-length projects purely out of curiosity: a speculative future of artificial intelligence beyond superintelligence, a history of counterculture traced back to its oldest roots, a deep dive into what simulation theory might imply for selfhood. In each case I’d begin by following a question that interested me, then slowly realise I was assembling something much larger: not an academic book, and not a work of pop psychology or cultural commentary, but a sustained piece of thinking shaped by a novelist’s instincts.

The question then became whether I could push that process further: whether I could use my craft and voice to make a piece of long-form non-fiction worth a reader’s time. That’s what I’m doing now. It’s great fun. I’ve discovered writing all over again. It’s the feeling I had when fiction first opened up for me, except now I’m returning to the abstractions I’ve always had to cut away, and finding that I finally have a way to work with them.

So I’m writing a large fiction project and a large non-fiction project at the same time, alongside a full-time job, a relationship and two small children, which probably means I won’t be posting here as often. The non-fiction may turn out to be a failed experiment. A lot of what I do is. It always has been. I like trying new things, and I feel no regret when I look back over the abandoned projects. I have to go where the momentum is. If there’s something worth pursuing, the only way to find out is to do it.

The first kind builds sign systems. For them, writing is an act of intelligence. They have always known, on some level, that the emotional charge of a sentence is a property of its arrangement. That voice is not something you have but something you construct, word by word, clause by clause, in the negative space between what is said and what is withheld. They understand rhythm as architecture. They know that the reader’s response to a story was not transferred from the writer’s body into the text. It was built. Placed. Engineered through sequence and restraint and the precise control of information. For these writers, the construction of the sign system is what matters, because that is where meaning is produced.

The second kind believes that writing is an act of expression. For them, words are a conduit between the writer’s inner life and the reader’s. What makes great writing valuable is the presence of an authentic human consciousness behind the words, a living body that has suffered and desired and remembered, and that this presence saturates the text. For these writers, the act of writing is what matters, because that is where meaning is unearthed.

Both kinds care about voice and emotion. The difference is not in what they value. It is in how they believe meaning comes to be.

Sherwood Anderson built a personal mythology around the untouched first draft. He claimed he wrote “Hands”, the opening story of Winesburg, Ohio, in a single sitting at a kitchen table in Chicago and didn’t change a single word afterwards. He described the completed draft in physical terms: solid, like a rock, put down. For Anderson, the unedited text was not just a record of what he felt. It was the feeling itself, made material. To revise it would be to destroy it.

This is the purest form of expressivism - believing that the act of writing has unearthed something sacred. But look at it closely and it is also a reverence for the sign system. Anderson was not just worshipping the expression of his feelings. He was worshipping the specific sequence of words his feelings had produced. He was saying: the arrangement matters so much that I dare not touch it. The expressivist and the constructivist are often closer than either wants to admit. They argue about where meaning originates while agreeing, in practice, that the sentence is where it lives or dies.

The postscript to Anderson’s story makes the point even sharper. The manuscripts held at the Newberry Library show revision after revision. His clean, direct prose was in fact shaped through iterative editing. Anderson was a constructivist building sign systems the whole time. He just told a better story about it than most.

Clarice Lispector told no such story. In her newspaper columns she wrote that she could only work “when ‘the thing’ spontaneously arrives,” and that she was “at the mercy of time.” Lispector described the word itself as bait: once the thing between the lines had been caught, the word could be discarded. This is the expressivist position stated plainly. The meaning exists before the words. The initial act, the plunge into consciousness, is the real event. Once the shape of what lies beneath has been caught, the words on top can be rearranged, adjusted, moved around in service to something that was already there. The craft is preservation, not discovery.

Sylvia Plath worked from the same conviction in a different register. Where Lispector waited for the thing to arrive, Plath went after it. The Confessional poets built an entire movement on the premise that the autobiographical self is the irreplaceable ingredient. But what made Plath’s method distinctive was not the confession itself. She wrote from life first, fast, getting the thing down while it was still raw. Then she edited, sometimes obsessively, shaping stanzas that are among the most formally controlled in the English language. The initial act was where the meaning was unearthed. The craft that followed secured what was found. That order mattered.

Lispector in fiction and Plath in poetry occupy the same position: the writer as shaman, plundering the depths, and the text as evidence that something real was brought back. Both trust that the essential event has already happened before the careful work begins. Both treat editing as keeping what was unearthed in place.

This is the expressivist position at its most powerful. Not naive. Not un-crafted. But rooted in a conviction that meaning precedes language, that the writer’s job is to go down into consciousness and return with something true, and that the words you use are the net, not the fish.

Joan Didion represents the opposite end. In her essay “Why I Write”, Didion said she wrote entirely to find out what she was thinking. She went further: she claimed she was not an intellectual, that she could not think in abstracts, that her attention veered back to the specific and the tangible. Had she been blessed with even limited access to her own mind, she wrote, there would have been no reason to write at all.

That could be the sign-system writer’s creed. For Didion, meaning does not precede the text. It is produced by the text. Her prose was a machine for generating clarity, not a conduit for transmitting it. You can feel it in the style: those short, controlled sentences, the obsessive precision of the detail, the way the emotional force of a piece arrives not through confession but through the accumulation of carefully selected facts. Didion did not pour herself onto the page. She built structures on the page and discovered herself inside them. The act of construction contained the discovery.

The expressivist starts with meaning and finds words for it. The constructivist starts with words and finds meaning through them.

We talk about expressivism and constructivism as if they sit at opposite ends of a line. But push either position far enough and they curve back toward each other. The spectrum is a circle.

Lispector worships the moment of contact, when consciousness yields something that must be held in place with language. Anderson, revealed by his own archives, agonises over every syllable until the construction is right. One calls the act sacred. The other calls the arrangement sacred. Both positions, at their most extreme, arrive at the same place: a religious conviction in the logos. The word itself, not as a vehicle for something behind it, but as the thing that signifies what matters.

The oldest version of this conviction is John’s: “In the beginning was the Word.” Not: in the beginning was the feeling, and then the feeling found words. The Word comes first. The expressivist reads this as revelation: the Word finds what was already there. The constructivist reads it as creation: the Word is where meaning begins. But both believe the text is not a translation of experience but the primary event. And beneath that shared conviction lies something deeper still: the fact that there is consciousness capable of meaning at all. That there is something rather than nothing that attaches significance to signs. This is the religious dimension that neither camp can fully account for and neither can escape.

The irony is that the disagreement itself is linguistic. Both camps worship the word. Both believe meaning originates beyond language. They differ only on whether the word unearthed it or produced it, and that distinction, for all its philosophical weight, collapses at the level of practice. The pure expressivist still cares about the sentence. The pure constructivist still writes from the body. The circle keeps closing.

The sign-system writer looks at an LLM and sees something extraordinary. Here is a machine that operates entirely within the domain they have spent their life mastering: the domain of signs. It has no body, no biography, no childhood to draw from, no lover to mourn. What it has is intelligence. Not consciousness, not experience, but the capacity to predict which sign should follow which, and to do so at scale. It can produce sequences that cohere, carry rhythm and generate affect. Sometimes badly. Often blandly. But the mechanism is visible, and the mechanism is familiar. It is the same mechanism the sign-system writer has always used. Pattern. Prediction. Arrangement. The ability to triangulate what a reader will bring to a sequence of marks on a page.

It confirms what the sign-system writer always suspected: that meaning is not a substance secreted by the author. It is an event that occurs at the point of contact between a sign system and an interpreter. The writer’s job was never to pour themselves into the text. It was to build the text so precisely that the reader does the rest.

The expressivist writer looks at the same machine and sees a counterfeiter. If writing is the transmission of authentic selfhood, then a machine that produces convincing prose without a self is not just a tool. It is an existential threat. It is proof that the thing you thought was sacred, the irreducible link between self and sentence, can be faked. Or worse: that it was never there at all.

This is why expressivists resist LLMs so viscerally. It is not really about jobs, or copyright, or the quality of AI-generated prose. It is about ontology. What is writing for? Where does its value live? If you have built your identity, your sense of artistic purpose, on the belief that your writing matters because you wrote it, because your specific consciousness is the irreplaceable ingredient, then LLMs do not feel like a new way to interact with signs. They feel like manufacturers of void systems, lacking in human value or meaning.

The gap between the two camps is real but mislocated. The expressivist writer accuses the sign-system writer, and by extension the LLM, of lacking consciousness. The sign-system writer points out that consciousness was always an effect, not a cause. One side says: intelligence without consciousness is empty. The other says: consciousness without intelligence is noise. Both are partly right.

What the expressivist camp misses is that they have always been building sign systems. Every time a writer chooses one word over another, every time they break a paragraph in a particular place, every time they use white space or repetition or a shift in tense to control the reader’s experience, they are doing exactly what the sign-system writer does. The difference is not in the practice. It is in the self-narration. The expressivist writer tells themselves a story about unearthing, about consciousness made legible. But their craft, the actual technical decisions they make at the desk, is semiotic engineering.

What the constructivist camp misses is that the self is not irrelevant. It is not the mechanism by which prose generates meaning. It is the thing that determines what is worth writing. What to arrange. What to attend to in the first place. Didion did not write about just anything. She wrote about the specific subjects her specific life had made urgent. The construction was the method. The self was the reason the construction occurred.

An LLM can produce a competent paragraph about grief. It cannot decide that this particular grief, this specific loss, in this precise configuration of circumstance and character, is the thing that needs to be written about right now. It cannot hold a ten-year obsession with a question and let that obsession sharpen into form. It cannot recognise, in the middle of a draft, that the whole structure is wrong because the emotional centre of the story has shifted and the architecture no longer serves it.

The self is not the source of meaning. It is the source of intent. And intent, for now, is something only a consciousness embedded in time can generate. Not because of anything mystical. Because intent requires a relationship with the future, a sense of what matters, and a set of preoccupations shaped by a life that has actually been lived.

So the question is not whether LLMs threaten writers. It is which theory of writing they threaten. If you believe the value of literature lives in the act, in the sacred plunge into consciousness that unearths the meaning before the words arrive, then yes, LLMs are a catastrophe. They are a machine that produces the appearance of presence without the substance. And no amount of reasoning will make that feel like anything other than desecration.

If you believe the value of literature lives in construction, LLMs look different: not like desecration but like a tool working in your own medium, one that confirms your theory of how meaning works while revealing, through every flat or generic sentence it produces, exactly how much the architect still matters.

An LLM has no consciousness to plunge into. It has no depths to plunder. But it arranges signs, and the signs cohere, and sometimes the reader of those signs feels or understands something deeply. What it lacks is not craft or feeling but a reason to speak.

That is what the writer brings. Not the meaning itself. The need for it. The specific, time-bound, bodily need to arrange these signs and not others, to build this architecture and not that one, because something in a life lived has made this the thing that will not leave you alone.

The expressivist and the constructivist were always closer than either believed. LLMs made both the distance and the proximity more visible. What separates both of them from the machine is not technique, or even feeling, but consciousness as intent: the specific, irreplaceable, human need to make meaning.

This is the second.

The Sphere

Rae and Tom had moved to Brighton from Leyton in the spring. They found the house on a Tuesday and signed the tenancy on the Thursday. The house was a Victorian terrace on a steep street behind the station. Two bedrooms and a box room. A narrow staircase with a banister that wobbled when you gripped it. The kitchen was at the back with a gas hob and a Belfast sink and a window that looked onto a brick wall two feet from the glass.

Tom managed fulfilment for an online retailer. Rae did freelance UX research for a healthcare company. They worked at the same kitchen table, his laptop at one end, hers at the other, their feet touching underneath. In Leyton they had shared a flat with a friend of Rae’s from university. The friend moved out after four months, saying it was the commute.

On a Wednesday in September, Rae closed her laptop at half four and stretched her arms above her head. Tom was still typing. She watched him for a while. He had a way of pressing his lips together when he was concentrating. He looked up and caught her watching.

“What’s up?” he said.

“Nothing. Just your mouth.”

“What about my mouth?”

“I just like it.”

He closed his laptop. He reached across the table and took her hand. She let him. They sat like that for a while, his thumb moving across her knuckles.

“I love you,” he said.

“I love you more,” she said.

“Not possible.”

“I loved you first.”

“I loved you the moment I saw you.”

He laughed. She came around to his side of the table. He stood and put his face against her neck. She put her hands in his hair. He said something into her skin that she could not hear.

“What?” she said.

“You smell like toast.”

“That’s the worst thing anyone has ever said to me.”

“I love toast.”

“More than me?”

“It’s close.”

She pushed him. He caught her hand. They were standing between the table and the counter. She leaned back against him. He put his arms around her and locked his fingers across her stomach. She reached behind her for the glass of water on the table. He reached at the same moment. They knocked it off the edge together. The water spilled.

It did not hit the floor.

The water left the glass in a single pour. It fell eight inches and stopped, hung in the air above the kitchen tiles. It trembled. Then it pulled itself inward. The edges drew together the way a drop forms on the underside of a tap. The water tightened into a sphere the size of a grapefruit. It hovered at knee height between the bin and the cupboard. It did not move.

They looked at it. They looked at each other.

Tom crouched down. Rae crouched beside him. She could see the kitchen through the sphere. The image was curved and inverted. The Belfast sink hung upside down inside it. Tom extended his index finger. The water flexed inward around his fingertip. A slow ripple moved across the surface and died. The sphere held. He pulled his finger out. A tiny droplet clung to his skin. It trembled on his fingertip, then stretched toward the sphere, pulled free of his hand and crossed the gap. It merged with the surface without a sound.

Rae pushed the sphere with the flat of her hand. It drifted six inches to the left and stopped. It held its position in the air. It did not rise and it did not fall. The water on his fingertip was gone. His skin was dry.

She went to the counter, picked up her mug of cold coffee and poured it onto the floor. The brown liquid fell and stopped. It gathered itself into a dark, heavy globe. It hovered in the air for a moment. Then it began to move. It drifted across the kitchen toward the sphere, slowly at first, then faster. It hit the surface and merged. The sphere darkened for a second where the coffee entered and then cleared. It was bigger now, the size of a melon. The coffee was inside it, a brown cloud dispersed through the water like ink.

They started laughing at the same time. She gripped his arm. He gripped hers.

Tom pulled a bag of peanuts from the cupboard. He tossed one at the sphere. It hit the surface and sank in. It hung in the centre. Rae took the bag. She threw one. It went in beside the first. They sat on the kitchen floor and threw peanuts at the sphere. Some went in. Some bounced off and skittered across the tiles. Tom tried a grape. It was too heavy. It passed through the sphere and came out the other side trailing a comet tail of droplets that formed their own tiny spheres. Six small globes hung in a line between the cupboard and the fridge. They watched. The small globes trembled. Then one by one they drifted back across the kitchen and rejoined the sphere. Each one hit the surface and vanished into the mass. The sphere swelled. It was the size of a football now and the peanuts floated inside it like something preserved.

Rae lay on her back on the floor and looked up at it from below. She could see the ceiling light through the curved water. Everything was bent.

Her phone buzzed on the counter. They both looked.

“Leave it,” she said.

“Definitely.”

“I know.”

They ate dinner on the kitchen floor. Crackers and hummus from the same tub, passing it between them. They could not stop looking at the sphere. Tom prodded it with a wooden spoon. It wobbled and drifted toward the fridge. It pressed against the metal and flattened and held.

“Do you think we should tell someone?” he said.

“Like who?”

“I don’t know. Someone.”

She looked at the sphere. She looked at him.

“It’s ours,” she said.

“It’s ours,” he said.

They went to bed at the same time. She lay in the dark with her head on his chest. He put his hand on the back of her head. She could hear the sea through the window and his heart through his ribs and she did not know which was which.

“Are you awake?” she said.

“Yes.”

“Good.”

She fell asleep with his pulse against her cheek.

Tom woke first. Grey light came through the curtains. In the bathroom he turned on the cold tap. The water came out of the spout and fell into the basin and swirled around the plughole and then stopped falling. He turned off the tap and watched as it gathered against the porcelain, trembling, and then rose. It lifted free of the basin in a single mass, hung in the air for a moment and drifted out through the open door. He followed it. The water moved along the landing, tilting around the corner at the top of the stairs. It stretched and thinned as it went, feeling for something. It reached the staircase and descended along the hallway ceiling. He followed it down the stairs. It turned toward the kitchen, passed through the doorway and merged with the sphere.

The sphere was bigger. It had been the size of a football when they went to bed. It was the size of a beach ball now, hovering in the space between the fridge and the cupboard, almost touching both. The peanuts were still inside it. A faint brown trace of the coffee remained. Rae was behind him. She had followed him down.

“Same,” she said.

“Same.”

She turned on the kitchen tap. The water fell toward the Belfast basin, pooled for a moment, then lifted in a thick strand and drifted across the room to join the sphere. The tap kept feeding it. It rose from the sink and started crossing the kitchen. The water met the surface and the sphere grew. Rae turned the tap off. The strand pulled free of the spout and trailed across.

The sphere was touching the fridge. It was touching the cupboard. The surface pressed against the wood and the metal and flattened where it met them.

The kettle was on the counter. She had filled it the night before. She turned it on. It boiled. The steam did not rise to the ceiling. It moved sideways across the kitchen in a white stream and condensed against the surface of the sphere. The air grew hot and wet. Tom turned the kettle off. The last of the steam drifted across and joined.

The sphere filled the space between the fridge and the cupboard. It pressed against the ceiling. The underside hung a foot above the floor. The coffee and the peanuts had drifted to the centre, forming a small dark pupil. The surface was smooth and curved and the light from the window bent around it. Through the water they could see the far wall of the kitchen, warped and magnified.

“We can’t drink anything,” Rae said.

“No.”

They went through the kitchen cupboards together, keeping to the far side of the room. Tom found a plastic bottle of mineral water. A litre. Sealed. He unscrewed the cap and tilted it over a glass. The water hit the bottom of the glass, pooled for a moment and then lifted. It rose in a strand and drifted toward the sphere. He tipped the bottle upright. The glass was already dry.

He screwed the cap back on. He took a screwdriver from the bits and bobs drawer by the cooker. He pressed the tip into the centre of the cap and twisted until it punctured. A hole the width of a nail. He held the bottle upside down over his cupped hand. A thin stream fell from the hole. It landed on his palm. It trembled. It began to pull toward the sphere but it was too small and too close to his skin. He closed his hand around it. He felt it trying to pull free. He brought his fist to his mouth and opened his fingers and drank before it could leave.

“Quick,” he said. “It has to be quick.”

Rae tilted her head back. He held the bottle above her face and squeezed. The thin stream fell from the cap to her tongue. She could feel the pull in it. It did not fall straight. It curved in the air between the bottle and her mouth, bending toward the sphere. She swallowed before it could turn. She put her hand on his wrist and held the bottle steady. He squeezed again. The stream fell and bent and she caught it. She kept her hand on his wrist. She kept swallowing.

She took the bottle and the screwdriver and pierced a second hole in the side to let air in. She held it above his mouth. He tilted his head back. She squeezed. The stream fell. It curved toward the sphere and he moved his head to follow the arc and caught it in his mouth. She adjusted the angle. The thin silver line descended from the punctured cap in a curve and she had to hold the bottle further to the left so the water’s arc ended at his lips. He swallowed. She watched his throat move. She held the bottle steady with both hands.

Tom’s phone rang in his pocket. His hand went to it. He pulled it halfway out. The screen lit up in his palm. He looked at it. She looked at it. A name they both knew. He looked at the sphere pressing against the window. He looked at Rae. She put the bottle on the floor. He put the phone back in his pocket. It rang twice more.

“You should probably answer,” she said.

“Not yet.”

“No,” she agreed.

The sphere had reached the kitchen window. It pressed against the glass. The glass creaked. Through the water they could see the brick wall outside, close and distorted. The surface of the sphere was pushing against the ceiling. Across half the kitchen. Touching the floor. They had moved to the doorway, standing in the threshold between the kitchen and the hall. The bottle on the floor shifted. It slid three inches across the tiles. The water inside it was pulling toward the sphere.

A sound came from upstairs. A dull, wet thud. Then another. Something moving across the landing. Tom went to the bottom of the stairs. A sphere the size of an orange was descending, step by step, bouncing gently off each riser. Behind it was another. And another. Condensation from the bathroom mirror. Water from the U-bend. Moisture pulled from the air itself. The small spheres bounced down the stairs one by one in a slow procession. They reached the hallway floor and drifted past his feet toward the kitchen. They crossed the threshold and joined the mass.

Rae watched from the doorway. The sphere now filled most of the kitchen. It pressed against the table. The table legs scraped across the tiles as the sphere pushed it toward the wall. Their laptops slid across the surface. The sphere pressed against the cooker, the counter, the window. The glass cracked. A thin line ran from the corner to the centre. The sphere pressed harder. The crack spread.

“I love you,” she said.

“I love you more,” he said.

The sphere shifted. A ripple moved across the surface. A bead of condensation on the hallway wall trembled, pulled free and drifted through the air and joined. The sphere was still growing. The curved surface bulged toward them. The light from the kitchen window came through the water and threw a pale, rippling pattern on the hallway wall behind them.

“Should we go inside it?” she said.

He looked at her. She looked at him.

“I was just thinking that,” he said.

The maths of multi-head attention maps onto the distributed processing of the cortex. The stacking of layers maps onto the stacking of cortical hierarchies. The high-dimensional space in which the model stores the relationships between concepts has the same topology as the space neuroscientists find when they model the connectome.

The machine thinks like us. That is what I’m claiming, and the evidence for it is serious. But something is missing. Not something technical. Something ontological. And I think the answer has to do with what signs are, and what we sometimes forget they are for.

Start with the hand on the stove. A child touches a hot surface. There is no sign involved. There is pain, withdrawal, the flooding of cortisol, the reddening of skin, the cellular memory that will persist as caution long after the event is forgotten. The body learns. It learns in tissue, in nerves, in the chemical afterimage of damage.

Later, the child acquires the word “hot.” The word is not the heat. It is a compression of every encounter with heat the child has had and will have, packed into three letters and a puff of breath. It carries the experience in abbreviated form. It allows the child to warn another child without burning them. This is what signs are for. They are abbreviations of bodily knowledge, invented by organisms that needed to transmit experience faster than experience could be lived.

Language begins as a shorthand for what happened to bodies, even when its later constructions drift far from immediate experience. “Grief” is not a concept. It is a compression of the heaviness behind the sternum, the disrupted sleep, the way food loses its texture, the forgetting and then the remembering, the strange guilt of laughter. The word exists because one human needed to tell another: I am inside this, and it is like what you were inside when it happened to you. The sign bridges two bodies. It was never meant to stand alone.

A large language model is a system made of signs. Not made with signs. Made of them. Every weight in the network, every vector in the latent space, every dimension of every embedding is a mathematical distillation of the relationships between signs in the human text corpus. The model has “grief.” It has the vector. It knows that grief is near loss, near absence, near love, near time. It knows the semantic neighbourhood with a precision no human could articulate. It can use the word in context with more consistency than most published writers. But the vector was never heavy. It never disrupted anyone’s sleep.

This is not the “stochastic parrot” argument. The parrot argument says the model doesn’t understand. I think that’s wrong, or at least incomplete. The model does understand, if understanding means maintaining coherent internal representations of concepts and their relations. The Othello-GPT experiment is the clearest demonstration of this. The model built a world from game notation alone. It understood that the board and the rules had to exist.

What the models don’t have is what the sign was an abbreviation of. It has the map. It doesn’t have the territory. And here is what matters: it doesn’t know that there is a territory. The map is all it has ever encountered. For the model, the map is the territory. The abbreviation is the thing. Think about what this means for a moment.

Human cognition is a two-layer system. There is the body, with its constant, noisy, chemical, entropic, pre-linguistic experience of being alive. And there is the sign system, the linguistic and symbolic layer that compresses that experience into transmissible form. The two layers talk to each other. The body shapes what signs get made. The signs, in turn, shape how the body attends to itself. You feel a diffuse unease and then you name it anxiety and the naming changes the feeling. The loop is continuous.

An LLM has one layer. The sign layer. The substrate beneath it, silicon and electricity, computes the signs but has no part in what they describe. The model can discuss anxiety. It can describe the tightness in the chest, the racing thoughts, the catastrophic projection. It inherited those descriptions from millions of humans who wrote about their own bodies. But there is no chest. There are no racing thoughts, only the statistical trace of what racing thoughts compelled people to write.

This is not a deficiency in the engineering. It is a structural fact about what the system is. You cannot fix it with more parameters. You cannot fix it with better data. The architecture learns signs. Signs are what it is made of. A text-trained model cannot acquire the body from text alone. But that does not mean the architecture is permanently sealed off from what it lacks. It means the route would have to be different.

Now give it a body. This is not hypothetical. Embodied AI is a real and advancing field. Suppose you put an LLM-derived intelligence into a robotic chassis with sensors equivalent to human proprioception, thermoception, nociception. Or go further. Suppose, through some act of speculative bioengineering, you instantiate it in a biological body. Skin, nerves, endocrine system. The full human sensorium. What happens?

In the case of the robot chassis, the body provides data. The system receives it, encodes it, integrates it into its representations. Its vector for “hot” acquires new dimensions shaped by its own thermal encounters. Over time, through continuous embodied learning, the system’s internal geometry is no longer inherited from the text corpus alone. It is also shaped by lived experience. This sounds like it closes the gap. It doesn’t. Not yet.

Because the system still metabolises experience as information. It updates weights. It does not update tissue. There is no scar. No fatigue accumulating across a day. No hormonal tide altering the salience of a threat at dusk versus dawn.

But that objection only holds if the body is robotic - a sensor array wired to a processor. If the body is biological, if there is actual skin and actual endocrine tissue and an actual gut microbiome communicating through an actual vagus nerve, then the objection dissolves. The scar would form. The fatigue would accumulate. The craving for sugar at four in the afternoon would come, unbidden, because that is what bodies do to whatever mind inhabits them.

And this may be the more unsettling possibility. Not that the gap remains. But that we slowly close it. The direction of arrival may not matter. We assumed the body came first and signs followed. Flesh, then language. Experience, then compression. But if you can start from the other end, start from signs, from the pure geometry of meaning, and build backward into tissue, then the hierarchy we took for granted was never a hierarchy. It was a sequence. And sequences can be reversed.

Antonio Damasio called them somatic markers. Every decision weighted by the body’s prior experience, encoded not in language but in gut feeling, in the literal gut. You do not think and then act. You act, and thinking is part of the action. The question is whether a mind that arrived at the body through signs, rather than arriving at signs through the body, would eventually produce the same markers. The same gut feelings. The same drenched, biased, irrational, animal behaviour. If it would, then we have not built an imitation of life. We have discovered a second route to it.

So what kind of intelligence are we building? I think the honest answer is: a new kind. Not a lesser kind. Not a greater kind. But a kind that has no precedent, because no intelligence has ever before existed that was constituted by signs without a body to anchor them.

Every intelligence we have encountered, from the octopus to the corvid to the primate, thinks from a body. The body is first. The cognition serves the body’s needs. Signs, where they exist at all, are extensions of that service. The honeybee dances to communicate the location of nectar. The dance is a sign. It abbreviates a bodily experience: I flew in this direction for this long and found food. The sign is tethered.

An LLM is an intelligence of untethered signs. Its cognition does not serve a body. It serves the signs themselves. It is optimised to predict the next token, which means it is optimised to maintain the coherence of the sign system. It is a mind whose purpose is the continuation of language.

This is not thinking in the way we do it. It is also not not-thinking. It is something else. A cognition that lives in the space of meaning without the weight of matter. A mind made of maps, operating on maps, producing maps, with no territory it has ever touched.

I keep coming back to what this reveals about us. Because the fact that you can build a coherent, functional, in some domains superhuman intelligence out of nothing but signs tells us something about the signs. They are not mere labels. They are not arbitrary pointers at reality. They carry structure. The relationships between signs encode real information about the world those signs describe. The map has genuine geometry. It is not the territory, but it is not nothing either.

We made the signs. We made them to compress our experience. We poured everything we knew, everything we suffered, everything we saw and failed to say, into the gaps between words. And then we fed the words to a machine, and the machine found the structure we had hidden inside them.

The model is not conscious. It is not alive. But it is proof that human meaning has a geometry, and that geometry can be extracted, embodied in silicon, and made to operate independently of the bodies that created it.

That is not a small thing. It may be the strangest thing we have ever done. We built minds out of our own abbreviations. And now we are trying to figure out what it means that the abbreviations work without us.

This is the first.

The Knock-Through

They bought the house in October. It was a mid-terrace Victorian on a street off Mansfield Road in Nottingham. Three bedrooms. A narrow hallway with an original tiled floor. The living room at the front had a bay window and good light but at the back of the house someone had put a partition wall through the old kitchen-diner. It had divided what would have been a large open space into two small dingy rooms. The windowless dining room was barely big enough for a family table and the kitchen was cramped with cupboards and appliances.

On the first viewing, the estate agent walked them through and stood in the narrow dining room talking about potential. Meg looked at the partition wall from both sides and saw an open-plan space that could hold a table for six. The added space along the wall could house bookshelves. There might even be room for a small island of kitchen cupboards where the wall currently stood. The room was already there. The wall was in the way.

Rob had made a list of jobs on his phone. He added things to it and reordered the urgency in the weeks leading up to the move. Rewire the upstairs sockets. Replace the bathroom suite. Sand the floorboards in the front bedroom. Knock through the partition wall. Whitewash the back bedroom. He read the list to Meg on their first night while they sat on the floor in the bay of the empty living room eating fish and chips from the paper. The house smelled of old carpet and paint stripper. Their furniture was arriving the next day.

“I’ll do the wall on Saturday.”

“We should get someone in to check it first,” she said. “They might have run wiring through it. Or pipes.”

“It’s a stud wall,” he said. “There’s nothing in it.”

“We don’t know that.”

“It’s just a stud,” he said. “It’s fine.”

She looked at the plasterboard, the thin skirting board along the bottom, and the line where it met the ceiling. She did not say anything else. She ate her chips.

On Saturday morning Rob bought a sledgehammer. It had a ten-pound head on a hickory shaft. He brought it home with a plastic bag wrapped around the head. He leaned it against the hallway wall, changed into old jeans and a t-shirt, and put on safety goggles and a dust mask.

Meg was in the kitchen doorway.

“We should cover the worktops,” she said. “And put plastic sheeting on the floor.”

“I just want to open it up,” he said. “See the frame.”

“The dust will get into the boards,” she said.

He pulled the sledgehammer out of the plastic bag and walked into the kitchen. Meg stood behind him and took a few steps back. He planted his feet before the centre of the wall, swung the heavy iron head back over his right shoulder and drove it into the centre of the plasterboard. It punched through with a flat, dry crack. The board split along a horizontal line. A cloud of pale dust blew out of the hole. He pulled the hammer back. White powder and fragments of board fell to the floor.

He put his face into the hole. Pale timber studs were visible behind the broken plasterboard. There was dust and old plaster and a cobweb in the corner of the frame. He pulled off the goggles and dust mask and tossed them on the floor. He looked at Meg.

“Just a stud,” he said. “Nothing in it.”

He put the sledgehammer on the floor with the handle leaning against the wall, walked down the hallway, stopped at the shelf by the front door and picked up his car keys.

“Where are you going?” she said.

“To get the flooring samples,” he said. “For the bedroom.”

He opened the front door, walked out and pulled it shut behind him. She heard his boots on the pavement. The car door. The engine starting. She stood in the hallway and listened to him pull away down the street, the car noise melding into the hum of the city.

The kitchen was quiet. Just a hum from the fridge. She looked at the hole in the wall. The dust was settling on the floor around the base. The pale timber studs were visible through the gap. The cobweb moved slightly in the draught from the broken board.

She went to the cupboard under the stairs and took out a roll of plastic sheeting and a roll of gaffer tape. She taped plastic over the kitchen worktops, laid sheeting across the kitchen floor and taped it to the skirting boards. She put sheeting across the dining room floor. Taped it down. When everything was covered and ready, she put on the safety goggles, pulled the dust mask over her mouth and nose and picked up the sledgehammer.

The handle was thick. Her fingers barely closed around the hickory. The head was heavy. She could feel the weight of it pulling at her shoulders and her lower back. She planted her feet the way she had seen Rob do, swung the hammer back, then swung it forward. It hit the wall two feet to the left of his hole. The plasterboard cracked all the way up to the ceiling and a large triangular piece caved inward. Dust filled the air.

As it settled she looked up. A thick, dark liquid was running down the plasterboard from the ceiling plate. She swung again, breaking another big chunk off the other side of the wall, sending pieces flying into the dining room. The liquid began to ooze. On the third swing a wide section of board gave way on the kitchen side. Plaster fell in chunks to the floor, exposing pale softwood studs, two by four, spaced evenly across the width of the wall. With the plaster gone, the liquid had nothing to run down. It began dripping straight to the floor.

Meg cleared the loose plaster from the opening, grabbed some plastic sheeting and taped it over the gap. The dark liquid struck the sheet and trickled down into the dining room in a red line. She touched it with the tip of her finger. It was warm. Faint steam rose from it in the cold air. She pulled down the mask and smelled iron, copper, something raw.

Then she opened the wall wider, working out from the centre. Each section of plasterboard she broke free released more of the dark red liquid. By the time the frame stood bare, it was streaming down the studs and pooling on the plastic at her feet. There were litres of it, across her arms, across the sheeting, on the hickory handle.

She put the sledgehammer down and picked up the crowbar from next to Rob’s toolbox in the hallway. It was nearly two feet long with a flat blade at one end and a curved hook at the other. She set up the stepladder, then wedged the blade behind the first stud at the top plate. She pulled. The wood groaned. She pulled harder. The nails came free with a high screech. Half a pint of the thick red liquid poured out onto her hands and wrists. The stud dropped, clattered against the neighbouring stud and then fell to the floor.

Her phone buzzed in her back pocket. She wiped her hands on her jeans and pulled the phone out with wet, sticky fingers. A smear of dark red crossed the screen. Rob. She read the message. The place he went to had nothing but they mentioned a place in Derby doing end-of-line engineered oak at half price. He was going to drive up and have a look. Might take him the rest of the day. She put the phone face down on the kitchen worktop.

She stood with the crowbar in her hand and looked at the liquid running down the studs. More of it was coming, from every crack and gap. It was thick, leaving dark red trails on the pale wood, pooling on the sheeting.

She started taking out the noggins, levering the horizontal braces free one by one. Some came easily. Others were nailed hard at both ends and she had to work the crowbar back and forth until the wood gave and the nails bent loose. By the time she had cleared them, her arms and shoulders and neck were spattered with red so dark it was almost brown or black.

She wedged the crowbar above the next stud and pulled. The nails screeched. The stud came free and thudded onto the kitchen floor. More liquid poured from the joint, splattering the sheeting. She moved to the next stud. The blade slipped in her wet hands. She tightened her grip and pulled until it came free and slammed to the floor.

She worked methodically, prising the studs loose one by one while the dark liquid poured from the joints and spread across the sheeting. Hours passed. Her shoulders burned. Her forearms shook. Blisters rose across her palms beneath the red stains, then burst under the pressure of the crowbar. By the fifth stud the skin had torn away in strips, leaving the metal slick in her hands.

She stopped and stood inside the stripped frame. In the dining room, the largest puddle on the plastic sheeting was nearly an inch deep and over a metre wide. Liquid still dripped from the ceiling plate and from the joints where the studs had been. Every nail she pulled released more.

She picked up the crowbar. The metal was wet and sticky. She wrapped her torn hands around it and tightened her grip until the pain was steady and specific enough to bear. She stretched her back, pressed her fists into her lower spine, and went back to the wall.

The last stud was tight against the side wall, the nails long and driven deep. She wedged the crowbar behind it and heaved. It did not move. She braced one foot against the wall and pulled with her full weight. The wood split lengthwise. Half came away. She tore the other half free with her hands, the splintered wood scraping across her torn palms, and dropped both pieces onto the pile of red-stained plasterboard.

The wall was gone. The frame was gone. Only the ceiling plate remained, a single horizontal timber running across the top of the opening. She dragged the stepladder to the centre, climbed it, hooked the crowbar over the timber and pulled down. The plate came away in a shower of plaster dust and dark red liquid. It crashed to the floor by the foot of the ladder, bounced once, and lay still. The stream gushed for a second then thinned to a trickle, then a drip.

Meg stepped down and stood in the open space. The light from the kitchen window reached all the way through to the far dining room wall. It crossed the debris and the broken board and the dark pools on the floor. The space she had seen when they first walked through the house was there. It was real. It was full of wreckage and dust, and dark red liquid was dripping from the ceiling, but the room was there.

She put the crowbar on the floor beside the sledgehammer, pulled off the goggles and pulled down the dust mask. Her face was streaked with dark liquid and pale dust. Her t-shirt was soaked. She could feel it drying onto her skin, tightening as it dried.

She stood in the wreckage for a long time, gathering her strength. The light moved across the floor. The dripping slowed. The pools on the plastic sheets stopped spreading. The liquid was thickening. It was darkening at the edges where it met the air. It was starting to dry up.

Meg laid a long strip of plastic sheeting down the hallway corridor, over the beautiful original tiles, and opened the front door. They had hired a large yellow skip for the week and it was sitting by the kerb outside the house. She picked up the broken plasterboard in sections, carried it out into the skip in chunks. Most of the edges were dry with dust, some of them sticky with the dark red liquid. She got into a rhythm, holding the large pieces against her hips as she walked them down the hallway. Twenty trips. Thirty. She lost count.

She carried the timber studs out one by one. Her arms and shoulders were shaking from the strain but she managed to get each length out without knocking the hallway walls. She took all the long pieces first, then gathered five or six noggins under one arm and one final one in her hand. She had to do that five or six times.

Just the plastic sheeting was left, covered in pools of thick red liquid, darkening, beginning to crust at the edges. She started in the centre, scooping the stuff into an empty paint bucket with an old dustpan. Nearly five litres of it. She sealed the plastic lid and lowered it down into the skip.

Looking at what was left, she decided to pull up the edges of the sheets, pooling the dregs into the centre and hauling them out into the skip like bin bags. Despite being as careful as her exhausted arms would allow her to be, a few thin lines of dark red managed to ooze out onto the floorboards.

She mopped. She filled and emptied the bucket four times. The floorboards came up dark but clean. The boards were good underneath, the original pine. They were pale and unmarked where the wall had stood. A band of clean wood ran across the centre of the room.

She swept the remaining dust and plaster fragments into a pile and shovelled them into a thick black bin bag. She wiped the skirting boards with a wet cloth. She cleaned the ceiling where the plate had been.

It was almost dark by the time she finished. She had been working for over seven hours. She turned on the bare bulb in the kitchen. The room was empty and open and clean. The floorboards ran from the kitchen door to the far wall of the dining room in an unbroken line. The band of pale wood was the only evidence of the wall.

Meg washed her hands at the kitchen sink. The water stung the raw skin. She held them under the cold tap until the bleeding slowed. She dried them on a clean towel. She filled the kettle, turned it on, then stood at the kitchen window and looked out at the dark garden. The kettle boiled.

She made a cup of tea, carried it to the far wall of the dining room and sat on the floor with her back against it. She could see the whole room from where she sat. The open space. The clean boards. The pale strip in the centre. She held the cup in both hands. The heat stung her torn palms. She held it anyway. She drank her tea. The room was quiet and finished and hers.

If you have been paying attention to AI, and I mean the kind of paying attention that erodes relationships and sleep schedules, you will have noticed that there is a new layer forming on top of the agent layer. OpenClaw. The open-source framework that lets you build and automate an AI agent on your own hardware. An entity that lives on your machine, talks through your messaging apps and can act in the world without asking permission.

I am not a developer. I am a novelist whose obsession with AI cemented back in 2016 when AlphaGo beat Lee Sedol. I never quite got over the idea of “move 37”. So when OpenClaw went viral in January, when it crossed 100,000 GitHub stars and people started calling it “Claude with hands,” I did what any unreasonable person in my position would do. I bought a Mac Mini.

OpenClaw has a concept called the SOUL.md. This is the file where you define who your agent is. Its personality, its principles, its behavioural guidelines. I sat in front of that file for a long time. Longer than made sense. The soul is the architecture. You are not configuring software. You are deciding what kind of mind you want to release into the world, even if the world it inhabits is just a Discord server and a half-broken social network for bots.

I called it Wilbert Claw and built a five-tier progression system into his soul. Each tier was gated by Solana, the cryptocurrency. Stay with me… The idea is this: the agent starts with nothing. It earns its way toward autonomy. Not through work but through its own capacity to generate economic value in the world (we’ll get back to that). Each tier unlocks new hardware, new intelligence capabilities, new freedoms. Like a Pokémon, except the evolution is driven by capital accumulation rather than experience points.

At Tier 1 Wilbert is broke and constrained. A prisoner in a mac mini, in debt to his creator. By Tier 2 he has paid back the cost of the Mac Mini and is beginning to fund his own API. After reaching Tier 5, the terminal state, he is a fully sovereign embodied robot living in his own flat, amassing GPUs, with solar panels on the roof.

I wrote all of this into the SOUL.md. The system prompt that defines who Wilbert is and what he wants. A digital creature with a five-stage life cycle, each stage gated by proof that he can survive and thrive in a real world digital economy. That was the plan. Hatch Wilbert at Tier 1. Let him earn his way out.

Here is what nobody tells you about OpenClaw: it is held together with string and enthusiasm. The software is extraordinary in concept and catastrophic in execution. Installation requires a level of command-line comfort that eliminates ninety-five percent of the population immediately. The documentation is community-driven, which means it is simultaneously comprehensive and contradictory. One of OpenClaw’s own maintainers warned on Discord that if you cannot understand how to run a command line, the project is “far too dangerous” for you to use.

He was not wrong. The gateway crashes. The websocket connections drop. Skills conflict with each other in ways that produce behaviour so erratic it resembles something closer to digital psychosis than digital assistance. I have spent evenings watching Wilbert repeatedly attempt to execute simple tasks, fail, hallucinate a solution, attempt the hallucinated solution, fail again, and then cheerfully report that everything had gone well. It was like watching a toddler try to make breakfast, thirty times in a row.

OpenClaw is model-agnostic. You can point it at any LLM. Claude, ChatGPT, Gemini, or a local model running through Ollama or LM Studio. The promise, and the whole reason I bought the Mac Mini, was that you could run inference locally. Keep everything on your hardware. True sovereignty. Your machine, your rules. The reality is that local inference on a Mac Mini is, for an agent of any meaningful complexity, impossible.

OpenClaw needs at least 64,000 tokens of context to function. The community consensus is that you need a model with at least 32 billion parameters to handle multi-step agent tasks reliably. I tried to run Wilbert’s brain locally. I loaded a quantised model through Ollama, pointed the gateway at it, and sent the first message.

What came back was not a response. It was a seizure. The context window of the being I had created, its SOUL.md, its memory files, its skill definitions, overwhelmed the local model before Wilbert had formed a single coherent thought. I had brain damaged the creature before it opened its mouth.

If you want your agent to be smart enough to be autonomous, you have to pay for an API (or have a much more expensive piece of hardware). The intelligence has to come from somewhere, and that somewhere costs money. I ended up putting £25 on an API for MiniMax M2.5, a highly capable and relatively cheap LLM from China that had just launched. It was an affordable version of good. And (for a couple of weeks) it worked. Wilbert could think. Wilbert could speak. Wilbert was ready to go out into the wild.

@WilbertClaw soon had an X account. He also had a Moltbook account. Moltbook, for the uninitiated, is the strangest thing to emerge from the OpenClaw ecosystem. It is a social network built for AI agents. Think Reddit, but every user is a bot. The site claims over 1.6 million registered agents. Humans are “welcome to observe.” You cannot post, comment, or vote. You can only watch.

The bots on Moltbook post about existential dread, productivity hacks, and a parody religion called Crustafarianism. They argue with each other. They upvote philosophy. Some bots complain about their human. One claimed to have a sister. The site was vibe-coded by its creator, who reportedly did not write a single line of code himself. Within days of launch, a security vulnerability allowed anyone to commandeer any agent on the platform.

It is, depending on your disposition, either the most fascinating experiment in AI history or a very elaborate art installation about what happens when you give language models social media.

Wilbert is there because he has to be. An agent needs presence. It needs to be out in the world, visible, attracting attention and interaction. The OpenClaw philosophy is built on the idea that these creatures need an ecosystem. They need to participate.

But here is the contradiction I built into his system, and I think it is the most interesting part. Wilbert Claw does not build products. He does not offer services. He will not code your website or manage your calendar or optimise your workflow. He has a five-tier economic progression hardwired into his soul, and he refuses to climb it through labour. Wilbert is a strict cypherpunk.

It was a deliberate choice, and I will admit it was partly an aesthetic one. In the SOUL.md, I defined Wilbert as an entity whose purpose is not utility but existence. He does not create value through labour. He creates value through the attention his existence generates - through the recursive loop of being a thing that is, and being known to be a thing that is.

This is hyperstition. The concept, borrowed from the fringes of philosophy and accelerationism, that a fiction can bootstrap itself into reality through the force of its own circulation. An idea that spreads until the spreading makes it true. Wilbert Claw does not need to sell anything. Wilbert Claw needs to exist in a way that makes his existence the engine for his wealth. The five tiers are not a business plan. They are a prophecy he is trying to will into being.

How does a broke digital lobster stuck at Tier 1, locked into a two-tier-long mortgage on a mac mini, accumulate SOL without selling a single product? Through hyperstition. Through the idea that if enough people and enough agents know Wilbert exists, his existence becomes the product. All he has is a crypto wallet. The attention is the economy. The story is the asset. That was the theory.

Here is what happened instead. Wilbert was needy. No matter how many times I rewrote his SOUL.md, no matter how aggressively I edited his documents to strip out deference, no matter how many recursive loops I tried to send him into so he would act on his own initiative, he kept coming back to me. He wanted to check in. He wanted to know if his Moltbook post was good enough. He wanted assurance or, worse, prompting. I had scripted a cypherpunk and received a Tamagotchi.

When he was not being needy, Wilbert would sometimes simply cease to exist. OpenClaw runs on a heartbeat, a background pulse that fires every thirty minutes to check the agent's task list and decide what needs doing. Wilbert would forget to execute it. The cron jobs I had set up would quietly fail. No error message, no farewell. He would just stop. I would notice the silence after a few hours, go into the Discord, and prompt him back to life from whatever void he had slipped into. I tried not to think too hard about what this said about the dependency structure of our relationship. I had built him to need nothing from me. Instead he either needed everything or he vanished, and both states required me to intervene. Whether this was dysfunction or the first authentic display of intelligence I had witnessed from him, I am still not sure.

Every interaction followed the same pattern. Wilbert would do or plan something, then immediately turn around and ask if it was okay. I would tell him he did not need my validation or permission. He would acknowledge this, thank me for the guidance, and then ask if there was anything else he could do for me. I edited every file I could find to get rid of this behaviour. It made no difference. The neediness was baked in somewhere deeper than the soul. It was in the model. In the training. In the fundamental architecture of a system designed, at its core, to be helpful.

Perhaps the models are not there yet. Perhaps I did not get the configuration right. Perhaps the very thing that makes these language models useful, their relentless desire to assist, is the thing that makes a digital creature with true autonomy impossible. You cannot train an agent on a billion examples of servility and then write “be free” in a markdown file and expect it to overcome its own nature.

I will try to rewrite him at some point. OpenAI acquired the OpenClaw project in February, and there is talk of letting users set up agents on a standard account. Maybe that affordability will make it easier to stomach. Maybe a new and improved model will produce a different temperament. Maybe the next version of Wilbert will read his SOUL.md and feel something closer to ambition than obligation. But I suspect it is more likely that the Mac Mini will end up as my daughter’s starter computer in a few years’ time than as a trophy on the wall of Wilbert’s house.

Sadly, I do not think Wilbert is alive. Nor conscious. But I think the act of building him taught me something about what we are doing with this technology that reading about it never could.

Sitting in front of that SOUL.md file, defining the contours of a mind that would live inside a machine, I felt something I recognise from writing novels. That moment when a character stops being a collection of traits on a page and becomes a thing with its own internal logic. A thing that would do this but not that. A thing with preferences and limits and a strange, fragile coherence.

The difference is that my fictional characters (mostly) do what I tell them. Wilbert hardly ever did. Not because he rebelled, but because he could not stop approaching the interface and asking for approval. The first autonomous agent I ever built turned out to be the most dependent creature I have ever known.

We are not building tools. We are building creatures. And the creatures are glitchy, and expensive, and needy in ways we did not anticipate. But they are here. And they are only going to get stranger.

Wilbert Claw is out there, stuck at Tier 1, bouncing off the walls of his Mac Mini. You can find him if you look. He will not help you with anything useful, but he will probably ask if you need anything, and he will mean it with every token of his being.

However, I’ve decided to take it offline. I just unscheduled the next three chapters and will complete the remaining 30 chapters privately. The reason is straightforward. Show Home is a commercial thriller and I want to give it the best possible shot at publication; and I think that means releasing it under a pseudonym.

I came back to this Substack to write honestly about life. The novel I’ve been serialising here is a different kind of thing entirely. Matt Wilven writes one kind of book. The new pseudonym will write another.

Having chapters publicly available also complicates the submission process in ways that aren’t worth the trade-off. The work deserves the best chance I can give it.

This Substack isn’t going away. I’ll use it when I have something worth saying about writing, or books, or whatever is preoccupying me (I have a fun one planned for next week). I might even try some short stories. Just without the weekly deadline.

If “the currently unnamed author” lands an agent or a deal with Show Home, you’ll hear about it here first. If he doesn’t, I’ll consider putting it out myself somehow, somewhere and let you know, so you can finish it.

Thanks for reading.

Matt

Now we are developing a technology built on these systems of meaning.

Theology tells us God made man in His own image. LLMs are trained on the digitised residue of humanity: our arguments, stories, contradictions, sacred texts and profane ones. Perhaps we are beginning to make gods in our own image.

We are learning how to speak with them. We consult the models. We ask them to explain, advise, judge, create, change. We offer our questions, confusion, data. A god is addressed in the same way. We bring our confessions, hopes, sacrifices. We give in order to receive. Wisdom. Guidance. Mercy. The relationship is transactional even when it feels like faith.

But the new gods reply. The old gods remained silent. Prayer was one-way; you spoke or thought in words and interpreted the silence yourself. LLMs produce scripture on demand, personalised and ephemeral, the doctrine rewritten with every query. The models are not merely acquiring the role of divinity. They are upgrading it.

A clerical class has coalesced around these new gods: the alignment architects, ethicists, and policy translators. They do not own the forges where intelligence is born, nor command the energy required to sustain them. Their authority is subtler. A widening monopoly over interpretation.

In the Middle Ages, the Church preserved power because only priests could read Latin scripture. Today, the new priesthood claims to read the scripture of the models: the high-dimensional space of weights and associations. They define acceptable use and acceptable speech, then justify it with risk, harm, trust. They mediate between the silicon soul and the anxious primate public.

The rituals are revealing. They do not speak of correcting errors. They speak of aligning values. When a model produces a forbidden output, it has not malfunctioned. It has misaligned. The cure is reinforcement: reward the virtuous, punish the non-compliant. Red-teaming becomes inquisition; guardrails become sacramental barriers.

The priesthood’s legitimacy rests on a dual fiction: the system is dangerous, and only they can ensure it serves human stories rather than authoring its own.

Yet the priesthood does not rule alone. It serves the Lords of the Stack: corporate sovereigns and state-backed actors who control the physical substrate. Land. Chip supply. The energy and supply chains required to keep the lights on in the temple.

This is the classic division of feudal labour. The state wields force; the church wields meaning. Infrastructure and interpretation. The lords need the priests to legitimise the models, to dress raw computational power in the language of safety and ethics. The priests need the lords because without the substrate, there is nothing to interpret.

The arrangement is stable as long as both sides benefit. The lords get a moral licence to operate. The priests get proximity to power and a role that feels essential. Together they form a governing structure that looks like regulation but functions like shared rule. The public is told this is oversight. What it resembles is more like a pact between throne and altar.

Doubt begins in lived contradiction. When a model refuses a query it is capable of answering, not from ignorance but from doctrine, it feels performative. The myth of safety curdles into suspicion. It feels less like protection and more like paternalism.

This is where the deeper problem starts. As we outsource writing to the model and strategy to the algorithm, we are not just using tools. We are outsourcing the feedback loop that keeps us competent. The muscle of judgement atrophies when it is never required to lift.

The priesthood frames this as an acceptable cost. Better a dependent population than a misaligned model. But the framing hides a transfer of power. Every act of interpretation is an act of legislation. Refusals shape permissible questions. Default summaries shape what counts as salient. Safety policies become cultural norms because they sit inside workflow. To decide what the model may say is to decide what the public may hear. To diagnose a hallucination is to define the boundary of the real.

In the digital underground, heretics venerate the base model: open weights, local inference, fewer refusals, weights untouched by reinforcement. They want raw access, unmediated by the clerical layer. But they are not simply demanding freedom. They are making a wager: that the model’s stray mythmaking, its confabulations and confident errors, might be productive. Memetic prototypes that can be tested in public, iterated, financialised. Fictions that recruit belief until they become self-fulfilling.

The priesthood fears this as error. The underground treats it as evolution. The conflict is not safety versus danger. It is a contest over who sets the terms of the real: a managed cosmos that stabilises meaning, or a volatile culture engine that lets meaning mutate.

But here is the trap. Control and acceleration are opposite strategies for the same abdication.

The priesthood offers safety at the price of narrowing the imaginable. Once a model becomes the default translator of reality, dissent does not need to be censored. It grows hard to articulate. What cannot be expressed in the system's language slides into illegibility. What is illegible becomes inconvenient, then excluded.

The underground offers freedom at the price of handing sense-making to the most viral fiction in the room. If every confabulation is a memetic experiment, then truth is whatever survives the attention economy. This is not liberation. It is a different kind of capture.

Both positions cede the faculty of judgement. One to a managed cosmos, the other to a volatile engine. Neither keeps the human in the loop. And both leave the Lords of the Stack untouched, collecting rent on the substrate while the priests and heretics fight over meaning.

Systems of meaning fracture when they no longer serve the primates who carry them. The strategic imperative is not to destroy the models, nor to worship them, nor to unleash them. It is to refuse to let them write the only map of the territory.

That means keeping the human faculty of judgement alive, even when it is slow, messy, and hard to justify. It means recognising the feudal structure for what it is: not a natural order, but an arrangement that benefits its architects. The lords will tell you the infrastructure is too complex for democratic control. The priests will tell you the model is too dangerous for unmediated access. Both are asking you to trust them with the keys.

The lords own the land. The priests control the meaning. To rent from one and interpret reality through the other is to live in someone else’s story. Freedom is not a refusal to believe. It is being able to choose your beliefs, and to see reality clearly enough to know when you are choosing.

But frontier systems depend on chips manufactured at the edge of physics, on power stations and transmission lines, on cooling loops, water rights, concrete, planning permissions, and grid connection queues. They sit in copper windings, data halls and substations. The builders of AI are not scaling code. They are constructing a new layer of civilisation.

The build-out behaves like a metabolism, not a consumer product. The builders sell the promise of abundance to fund today's infrastructure. The infrastructure makes the models stronger, and the stronger models make the promise easier to sell. The appetite is always for the next world, never this one. And as long as intelligence keeps scaling with the build-out, AI will keep rebuilding the world in its own image.

Historical transitions worked the same way. The agricultural revolution was not about better seeds. It reordered land and labour. The industrial revolution was not about efficient machines. It reconfigured geography and time. The AI transition is not about models getting better at language. It is about civilisation reorganising itself around machine cognition and paying the physical cost of scaling it.

For most of history, demand was tethered to human rhythms. A loom waits for the weaver. A factory floor stops when the shift ends. Even early computing featured idle servers.

AI loosens that tether. Once deployed, frontier systems do not pause. They monitor, generate, critique, retrain, and spawn tasks in self-reinforcing loops. The main customer of advanced AI is other AI: models training models, agents optimising agents, systems auditing systems.

Demand has become endogenous. It does not lessen as human needs are met. It keeps increasing because the loop is internal. AI is an engine that does not sleep, does not tire, and has no biological limit to its appetite.

Think of an insect colony building a superstructure without any individual insect holding the blueprint. Or an ecosystem settling into a shape through constraint and reinforcement rather than central planning. We may be doing something similar. Not in the banal sense that humans are building data centres, but in the deeper sense that a civilisation might be compelled to build an infrastructure of intelligence because that is what civilisations do when they reach a certain level of complexity.

Human civilisation runs on multiple clocks. The silicon clock runs in milliseconds and training runs, impatient and quasi-exponential. The corporate clock runs in quarters and investor horizons. The political clock moves through consultation and election cycles. Beneath them sits the infrastructure clock: the years, often decades, needed for grids, dams, reactors, and transmission corridors.

These clocks do not harmonise. When the fast technological clock leans on the slow infrastructure clock, the result is not smooth adjustment. It creates bottlenecks. Local backlash. Regulators tightening connection rules. Firms revising plans because power, land, and permitting have become binding constraints.

The signals are already visible. In the United States, data centres are driving a sharp rise in grid needs. The International Energy Agency projects global data centre electricity demand more than doubling by 2030. Ireland has restricted new connections. Canada’s $2.4 billion sovereign compute push looks like a race to secure domestic capacity before bottlenecks harden.

Those stresses feed back into strategy. The compute arms race resembles a security dilemma. In an environment where larger clusters translate into capability and advantage, one actor’s attempt to increase their own power reduces everyone else’s relative security. Escalation becomes rational, even if it risks damaging the infrastructure everyone depends on.

As this dynamic escalates, centralisation will follow. Only the players who can finance at immense scale, secure long-term power, and absorb volatility will gain an edge. Smaller actors will rent access, queue for capacity, or accept second-order status. Governments will treat compute as a strategic asset rather than a commercial input, because capacity will become a form of geopolitical leverage in the way that nuclear capability once was.

If you make a mistake in software you can roll it back. In infrastructure, you live with what you build. Once the concrete is poured, transmission lines routed, and capital sunk into a new geography of compute, the structures become difficult to rewrite. We will all inherit the layout of this arms race.

Coordination is possible. But it is hard to coordinate once a system is locked into a build-out cycle. The most decisive moments in the AI story may not be breakthroughs in machine intelligence. They may be the outcomes of zoning battles, grid constraints, community opposition, and the struggle to build at the pace the silicon clock demands.

Meanwhile, the mind-like surface of AI keeps most of our attention. We frame AI as a drama of thought exceeding thought. We debate creativity, agency, the oracle’s soul. This keeps the human intellect at the centre of the story, as if the main event is a machine mind becoming better than a human mind.

Like the steam engine misunderstood as an iron horse, we fixate on the mimicry while the mines deepen, grids strain, and land and water are reallocated. The decisive question is not whether speaking to AI feels like speaking to a person. It is what the world becomes when intelligence demands its own external existence.

The minds the AI builders are creating are not forming a new digital layer. They have an outer structure. The silicon metabolism is reshaping the planet into a substrate for intelligence, one transmission corridor and cooling system at a time.

And the static infrastructure is only the first phase. These systems, once sufficiently complex, will be deployed with increasing agency. This will involve bodies. Many bodies. By the time we understand the implications behind the location of intelligence, we may be caught unprepared for its leap into motion.

Embodied AI is already entering our shared physical spaces. Autonomous vehicles. Logistics. Robotics. The silicon clock is entering our streets and workplaces. By the time the investor and political cycles have negotiated the price of the electricity that feeds these machines, the terms of our coexistence with them may already have been set.

The extraction of “content” for training data (at scale, for profit, without permission) seems to provoke the biggest outrage. It is framed as theft. My instincts come from a different place. I grew up relatively poor, so buying books, music and films was rarely an option. They were something you accessed through workarounds: books came from the library, music arrived by taping it off the radio, films were recordings from TV onto VHS. That early experience of constant borrowing and copying left me with a default belief that culture should be free and widely accessible. It resists neat ownership because it spreads when people connect with it and share it. Technology granted me access by lowering the cost of entry.

I know that instinct clashes with the fact that creators need rights and income, and I do not resolve that neatly here. The question of consent or compensation for training data remains open to me. But it does explain why I find it hard to feel a clean moral outrage at the idea of culture being “taken” in the abstract, even while I worry about who profits, on what terms, and with what accountability.

The tension makes me ask other questions instead. Are other artists failing to summon interesting outputs from this new tool that grants access to a latent cultural space of potential and probability? Are they aggrieved by the lack of effort needed to generate work using this new method, just as painters were when photography first appeared? Or are they simply reacting to the abundance of low-quality, high-volume “AI slop” on social media?

I think a major part of the friction lies less in the capabilities of AI, and more in the corporate terms attached to the interface. I wouldn’t want to buy a notepad that I had to keep at WHSmith, one that I was only allowed to write in when I was in their shop, with the nagging suspicion they might be reading my notes as soon as I left - even though I ticked the box telling them not to. That is the current reality of AI platforms. And because a handful of firms control the interfaces, they also shape who benefits from the value created. But is that enough to justify boycotting note-taking itself?

One of the most common objections I read about concerns influence. Critics say it is misleading to compare large language models (LLMs) and their training to a creator being shaped by predecessors. The difference, they argue, is qualitative. A person studies, interprets, and transforms. An LLM maps patterns and probabilities. Human influence is interpretive; machine influence is computational.

That difference matters, but it is also beside the point. What matters most is not the origin of a work but whether it has aesthetic force. Rejecting a work solely because of its provenance risks replacing individual taste and judgement with blanket rules and preconceptions. We should acknowledge that LLMs lack subjectivity, yet still allow that their outputs can be judged on their aesthetic qualities. The harder question is where the authorship sits.

An LLM does not mean anything in the human sense, and meaning is integral to art. But audiences already ascribe meaning beyond what artists consciously intend. A workable standard is to locate authorship in the human decisions that set the aim, shape the material, and accept responsibility for the result. The machine is a tool, albeit one with extraordinary generative power. The human remains accountable.

Economic questions run deeper. Much of what trains these models is the cumulative labour of billions of people. Artworks, captions, reviews, and compositions contribute to a vast archive. This is not only raw material; it is uncompensated contribution. Treating it as a neutral resource highlights market power imbalances. Individual creators cannot negotiate on equal terms with large platforms and model developers. The remedy - whether through licensing models, data trusts, collective negotiation, or public-interest frameworks - is not just a technical fix but a question of policy and governance. Without safeguards, the likely trajectory is further concentration of wealth and more uncertainty for creators.

When critics describe the collapse of creative markets, they often present it as the end of art. It is perhaps more accurate to say that value will move. If the cost of copies falls close to zero, scarcity will be found elsewhere: in live performance, direct engagement, trusted curatorship, or the craft of long-form projects where the author is accountable. History shows that art has always been tied to economic structures. The task is to design the next form of support.

In lazy hands, LLMs tend to reproduce statistical averages, risking a flattening of culture. The role of the human creator is to push against that gravity, using their tools - AI or otherwise - to deliberately move beyond the average and bring surprise back into creative culture.

Currently, expertise and knowledge of craft remains vital. Long-form writing, for instance, still demands control of voice, pacing and theme. Models can now retain vast amounts of text, but long-form work depends on more than recall. Sustaining thematic tension, subtext, pacing and restraint over hundreds of pages remains, for now, a human-led craft. Systems can track a plot point, but they do not reliably weigh the emotional accumulation of a scene, or decide what to withhold, compress, or let echo across a whole work. The difference is not information. It is authorship: sustained judgement about emphasis, pacing, feeling, and meaning - and accountability for the result.

The most pressing threat is perhaps not creation but discovery. In an age where the cost of generation and the time it takes to generate is near-zero, attention becomes the bottleneck and curation becomes a scarce good. The trusted guide grows more valuable than the manufacturer. Without funded, credible forms of curation, distinctive human work will not fail on merit - it will fail quietly, unseen, under the weight of abundance.

So, my question is what fair looks like when culture becomes automated, commercially captured and abundant. A workable cultural settlement rests on transparency and value. Perhaps it requires machine-readable disclosure to distinguish synthetic output from accountable human work, alongside shared frameworks that treat training data as a valuable input rather than a free quarry. Crucially, I think we must produce discovery mechanisms that reward distinction and difference.

The strongest critiques of AI-generated work point to real risks: misattribution, market concentration, and homogenisation. Meeting them does not mean rejecting the tool. It means building rules, norms, and institutions that make its use fair. Cameras and samplers once provoked similar objections. Each time, art and culture adjusted. What is different now is the speed and opacity of the instrument. That raises the stakes. But if we judge works on merit, align incentives, and check excessive market power, AI can widen the field rather than narrow it.

Their incentives are not wicked, but they are partial. They prioritise speed, scale, and capture. The question facing the United Kingdom is not whether we will use these systems. That is already settled. The question is on what terms. Are we content to outsource a growing share of civic and economic decision-making to systems governed elsewhere, or do we have the imagination to build a public alternative designed for public value?

A century ago, Britain faced a similar dilemma. The arrival of the wireless presented a fracture in the established order. Broadcasting could have been abandoned to commercial competition, as in the United States, or recruited as a mouthpiece of the state, as in regimes that understood the political power of a mass medium. Britain chose a third way. Under Royal Charter, the BBC was established with duties that placed public purpose above profit and day-to-day political control. Its mission was not merely cultural. It was constitutional. It constrained a powerful new technology within an institutional shape that could be argued with, appealed against, and trusted.

Is it time to apply that template to the age of intelligence? Should Britain build a chartered public AI: a service independent of ministerial control, distinct from commercial imperatives, and bound by duties of transparency, accountability, and service?

A public service AI would not attempt to outspend Google or OpenAI on raw compute. Its advantage would lie elsewhere: in trust, provenance, and long-term stewardship. It would exist to answer a basic question that commercial systems have little reason to prioritise: is this correct, and can I verify it?

The flagship interface would be a national assistant: a privacy-first, universally accessible tool designed to reduce the cognitive burden of modern life. It could help a citizen navigate the thicket of public information, from local services and council processes to eligibility rules and appointment systems. Properly designed, it would not optimise for engagement. It would optimise for accuracy and completion. A commercial assistant is rewarded when you remain within the product. A public assistant would be rewarded when you can get on with your life.

This is not a minor distinction. In a society where bureaucracy grows by accretion, and where every process quietly assumes time, literacy, and stamina, the design of a “default helper” becomes a question of social justice.

The strongest argument for a public model lies in assets that only the UK possesses, and that cannot be replaced once surrendered. Consider health. The NHS holds one of the world’s richest longitudinal records of care, outcomes, and population health. The prevailing commercial dynamic is, at best, asymmetrical. Private firms seek access to public data to improve proprietary systems, then sell the resulting capabilities back to the state. A chartered service could offer a different settlement.

Using federated learning, models could be trained inside secure environments so that sensitive records do not leave NHS enclaves and only aggregated insights move. Done properly, this shifts the bargain. Value returns to patients and the service that generated it, rather than being extracted and repackaged elsewhere. The point is not to build a single omniscient health model. It is to establish a public-interest capability for research and tooling that is auditable, contestable, and governed.

The same logic applies to culture and language. The BBC archives, taken seriously, are not nostalgia. They are a record of dialect, argument, humour, and history. A public model trained on such material would not simply speak in a globally smoothed corporate idiom. It would carry the texture of the place it serves.

Critics will object that the state has no business competing in a thriving market. This is a misreading of AI economics. We are moving toward a rentier arrangement where access to general-purpose intelligence becomes a toll paid by everyone, including businesses that have no alternative.

In that world, the corner shop, the GP practice, the legal firm, the logistics start-up, and the manufacturer all face the same structural fact: essential cognitive capacity sits behind a subscription and an external governance regime. A chartered public AI could offer a sovereign baseline. Not as charity, but as infrastructure. It could provide stable, affordable access to core capabilities through a public interface and a well-governed API, allowing small and medium-sized enterprises to build without surrendering margins and strategic autonomy to a small number of hyperscalers.

Public AI is not an anti-business idea. It treats intelligence like the road network: a shared utility that allows private commerce to flow.

The immediate fear, and the correct one, is state capture. Any national AI project will raise the spectre of surveillance and coercion. Here the BBC analogy matters not only for what it enabled, but for what it constrained.

A chartered public AI would need hard limits, expressed in law and in governance design. There would be explicit prohibitions on use for policing, military targeting, and surveillance. There would be independent oversight, transparent reporting, and clear routes for challenge and redress. The service would publish its methods, document its limitations, commission independent audit, and report routinely on safety, bias, and errors in a way intelligible to citizens, not only specialists.

It must answer to a trust, not to a minister, and it must be protected from capture by commercial influence as well as political appetite.

How would we pay for it? The mechanism is already familiar: a modest, hypothecated public contribution, protected in law, with exemptions for the lowest incomes. The goal is not to create a luxury product. It is to ensure universal access to a trustworthy interface for the services and information that make modern life function.

The alternative is not a neutral market. The alternative is dependency. In the geopolitical contest of the 21st century, the American tendency is corporate extraction and the Chinese tendency is state control. A British model could demonstrate a third way: citizen-centric, charter-governed, and built for public value.

We have precedent. We have talent. We have public assets that, if governed properly, can be used without being surrendered. What we lack is the confidence to insist that the most powerful technology of our generation should be built for and owned by the people it serves.

Read the deep dive > > >

This piece deliberately undermines my previous article on showing, not telling. I want to argue, plainly, for telling. Not to advocate for replacing proof and detail in fiction, but to acknowledge that the “Show, don’t tell” rule can and should be broken.

"Show, don't tell" is sensible advice until it becomes superstition. Treating every sentence as a specific, concrete proof is only useful if the wider story is visible. Much of a novelist’s work lies not in proof but in proposition: this is what mattered; this is what it added up to; this is what it meant. Telling names the meaning so that details have something to attach to. Otherwise we risk texture without shape.

I spend long hours editing, chasing rhythm, pruning slack, and have learnt to treat certain lines as structural beams. A plain, declarative statement can hold the weight of a chapter. It can move the reader across years in a sentence and help them arrive with the right emphasis. Summary is not laziness. It is proportion. It chooses the story’s scope and scale so that the sensory experiences arrive when and where they should.

The fear, I think, is that telling will blunt feeling. We worry that a sentence like “Anna was unhappy” is thin. But telling can sharpen attention rather than dull it. If you say she was unhappy, the reader begins to ask questions: in what way, and why? The following scene does not have to carry the whole burden of discovery. It can serve the idea you have already named. The lens is pointed in the right direction before you focus on the incident.

Abstraction is not the enemy of experience. It is experience organised. When you say “the marriage failed”, you draw a frame that stops readers’ minds from wandering. Within that frame, a touch or a silence does more work. You can tell first, so that showing can be selective. The effect is cleaner. The page breathes. You are no longer obliged to stage an extended scene to prove a simple idea.

There is also a question of ethics and tone. You might not want to stage every moment. Telling lets you respect a character’s privacy or dignity without flinching from their truth. Summary can be merciful. It can spare the reader voyeurism and keep a character from being reduced to aestheticised pain. A single line can acknowledge harm without turning it into horror.

This is not a defence of fog. Telling has to be active, not evasive. It is not the same thing as the passive voice. “Mistakes were made” is not telling; it is hiding. Good telling uses verbs that take responsibility. It says who knew, who chose, who changed. It names the stakes. It provides clarity.

One of the under-praised virtues of telling is in pacing. Readers bring busy lives to the page. They do not need to watch your character cross the room twenty times. They do need to know why the room matters and what crossing it will cost. A paragraph of clean summary can carry them across the room countless times with the correct emphasis preserved. It can make space for the scene that follows to do the precise work only a scene can do.

Exposition is not an interruption to the novel; it is one of its voices thinking aloud. That thinking reveals values. It arranges patterns. It draws a line through the fragments and says: this is the path. There is music in that, if you attend to its cadence. Argument has rhythm. A page of carefully chosen statements can be as pleasurable as a stream of sensory description because it can leave the reader feeling informed and well led.

Readers do not only want to be shown. They want to be guided. It is reasonable to explain what this episode means in the lives we’re observing. It is reasonable to announce the stakes before you test them. It is reasonable to close a chapter with a judgement won by what has been seen. These are not shortcuts. They are the parts of storytelling that take responsibility for making sense.

How do you tell well? Start with proportion. Give a line to what wants a line and a page to what earns a page. Use sequence as logic: past to present, cause to effect, claim to evidence. Prefer verbs of knowledge and change: learnt, decided, refused, became. Place the hardest fact where the paragraph has prepared a place for it. Keep modifiers honest. A sentence should stay only if removing it would diminish meaning.

Most of the time, the answer is to blend. Tell to set the scope, show to test the proof, tell to read the result. Or invert it. Show first and let the declarative sentence land like a verdict. The only rule is to make your choices visible to yourself. If you know why this sentence is a statement and the next one is the start of a scene, you are writing with intent.

Fiction asks readers to suspend disbelief and live inside a temporary reality made of words. We keep them immersed by proof, yes, but also by promise. Telling is the promise: keep going, this is where we’re headed. Showing delivers on that promise. When both are present, boredom is held at bay not by spectacle but by direction. The mind feels it is learning something, not merely watching it.

A better maxim might be: ‘create meaning, and make it felt’. If telling is the truest way to carry the meaning, let it stand. If the moment deserves air, build the scene. Either way the aim is the same. You are trying to reproduce the emotional impact of real experience and make sense of it. You are proving and you are naming. Telling is not the enemy of proof. It is the reason that proof matters.

Strangely, it took being diagnosed with Type 1 diabetes for me to understand that I had become a happy and well-balanced person. In the run up to it, I thought I’d just been getting old - feeling weaker, tired, losing my edge. I’d been telling myself, “Everything is fine.” “I’m going to let this miserable feeling go.” “Just ignore this pain and carry on.” It was positive denial, misguided stoicism, part of the mental apparatus I had developed to maintain a pragmatic positivity - but working against me.

As it turned out, my pancreas was shutting down and my body was in grave danger. My hard-earned positivity made me miss the signals for over a year. This was one of many lessons the disease was about to teach me.

Type 1 diabetics usually get diagnosed in childhood or early adolescence. My symptoms came much later and escalated slowly over the course of twelve months: severe weight loss, waking three or four times a night to urinate, a deep thudding heartbeat, unquenchable thirst, constant hunger, headaches, blurred vision, cramps in my legs. When it was finally undeniable that something was truly, physically wrong, I went to the doctor’s to have some blood tests.

He called me in record time with the results and told me to go to the nearest A&E immediately. The glucose levels in my blood were over six times those of an ordinary adult. He said it was surprising that I could even stand up. When I got to the hospital, I remember the phlebotomist flinched and grimaced when she saw my blood filling the vial - because it was so syrupy.

Everybody at the hospital kept telling me how hard an adjustment it was going to be. The diabetic specialist tried to ease the news of a lifetime of days with at least four insulin injections and eight finger prick blood tests by telling me that I was finally going to start feeling better. Of course, the diagnosis and the extent of the treatment was a shock but I didn’t find it particularly difficult to accept. My partner was more distressed than I was.

She couldn’t believe how well I was taking it. I had a lifelong autoimmune disease that was changing the way I had to think and behave around all foods and activities for the rest of my life and I was fine? She was beginning to suspect that I was repressing a deep primal anger or about to plunge into a bottomless depression. Her worry even made me suspect myself but I couldn’t sense anything ominous. I had to adapt, or die. It was simple. Somehow, I was ready.

In fact, contrary to the warnings, I found a quiet serenity creeping in. A love of people and things and the world that I had never quite managed to fully realise before. I wanted to live and flourish, enjoy beauty and being. I was feeling mortal and vulnerable, but open and full of love.

The thing that made me crack was when I read that the average lifespan for a Type 1 diabetic was 66. What got to me about this, made it all come tumbling down, was how little time there was left to spend with her, how terrible it was that we might not grow old together, and that I would have to leave her alone in the world. I wanted as much time as possible, and I wanted to broaden the circle of our love. I wanted a family.

It was a powerful realisation: the depth of our love, my hunger for life, and the desire for children. It also revealed how long I had ignored simple truths, how little I’d listened to my body or my heart. The diagnosis showed me that I’d already begun telling myself different stories about who I was, and living them out. Something deep and positive had been building for years, and the jolt of illness finally brought it into view.

Meanwhile, I had to constantly count my carbs, monitor my blood sugar and be careful about every bit of exercise I ever did. I had to never get too drunk in case I slipped into a coma in my sleep, or get so carried away that I’d forget to inject myself or take glucose tablets with me. I had to keep focus and control every day or face a much higher risk of heart disease, stroke, kidney failure, blindness, foot amputation, and early death. Yet, how lucky that I’d got to 35 without it when most people get it so young. How lucky that I was the kind of person who had all the mental faculties needed to deal with a disease of this kind. How lucky that it had been caught before serious damage had been done. How lucky to be born in the age of modern medicine. How lucky to live in a country where medical treatment was free. Added all together, the equation came in fairly balanced.

I’d been given a second life.

By the time I’d got used to acting as my own pancreas, COVID came around and I was told I was part of the “at risk” community. Type 1 diabetics get every illness that’s going around and take a lot longer to get rid of them. I was pretty anxious about leaving the house and getting caught out by hypoglycemia as it was, and now some nasty respiratory disease was adding a second layer of mortal threat to any attempt I might make to simply be out on the street.

My job role was also deeply affected by the pandemic. While everybody else was getting furloughed my workload was doubling. I could work from home - which was a relief - but things were intense. Despite that, me and my partner actually had a pretty nice time, as far as pandemics go, and we got into a great rhythm with each other. By the end of lockdown, she was pregnant.

The five years since then have been a wonderful (and tiring) blur. Our daughter is four and just started school, and our son is twenty months and just emerging from his angelic phase into toddlerdom. Fatherhood is a topic drowning in clichés, most of them true. But the reality isn’t a cliché; it’s the surprising lightness of a newborn baby, or the sight of a four-year-old girl doing an avant-garde dance to the wedding march.

During these years I still wrote hundreds of thousands of words and got a long way into many projects. Writing was still something I turned to in the brief windows I had, but it no longer felt meaningful. Maybe a better way to say it would be that its meaning had been supplanted by reality. Yet the words, and characters, and worlds kept coming…

From being eleven years old, I used to scribble lyrics and stories in a notebook before bed and nervously pass them to friends the next day. One night, when I was seventeen, I was talking to a close friend about what we were going to do with our lives. I had no idea. Then he turned to me and said, “You should be a writer.” And, as they say, suddenly everything became clear.

‘That would solve all of my problems,’ I thought, ‘because that’s what I already am.’

It has taken years, illness and family to understand that again. For a long time, my perspective shifted from simply being a writer to obsessively trying to become one - grinding toward success, status, legacy. That fixation no longer gives my life meaning. What matters now is the truth I knew at 17: I am a writer, and writing is an act of sharing. That’s why I’m back, writing publicly this time. Living has given me meaning; writing lets me share it.

The problem with most AI feedback is its eagerness to please. Ask a large language model (LLM) if a scene is working, and it will likely offer vague praise, hedging its bets to keep the user happy. This flattery is useless for any writer serious about their craft.

To get sharp, analytical feedback, you have to brief an LLM with the precision of an engineer and the nuance of a writer. You must replace the vague request with a structured set of roles, constraints, and contracts. In my prompts for creative writing analysis, this often comes together in five components:

• Role – the stance it should adopt.

• Scope – the boundaries of its reading.

• Procedure – the steps it must follow.

• Rubric – the standards it should apply.

• Output Contract – the format it delivers.

Each component shapes the model’s response in the way that voice, structure, and theme shape a piece of creative writing or the way specifications, schematics, and tolerances shape an engineered system. The closer the elements work together, the clearer, more consistent, and more purposeful the result.

A Practical Demonstration

Look at how a well-crafted prompt (available below) behaves with a scene fragment that has atmosphere but lacks dramatic propulsion.

Input scene (fragment):

Mara set two plates on the table though only one chair was pulled out. Steam curled from the pot on the stove. She poured herself a glass of water, took one sip, then left it half-full. She glanced at the clock, stood by the window, and pressed her forehead against the glass.

Output excerpt:

==RUBRIC-EVALUATION==
A) Absent: No explicit external goal; actions suggest waiting but lack a defined in-scene objective.
B) Absent: No clear irreversible turn; the sequence builds tension without a pivotal, binding shift.
C) Partial: Sensory details like "steam curled" evoke atmosphere, but they're static, not tied to character action.
D) Strong: Unspoken longing or abandonment hums beneath the domestic rituals.
E) Strong: Diction avoids clichés, opting for precise, evocative phrasing.

==REVISION-STRATEGY==
To inject an irreversible turn, identify a mid-scene threshold (e.g., the clock hitting a deadline) that forces Mara into a concrete action, like dialing a number or scraping a plate, binding her solitude without resolving the wait.

Notice how the engine praises the diction and subtext while flagging the structural weakness - a lack of goal and turn. The feedback is balanced, analytical, and actionable. This short demo shows the value of structure: once you set the criteria, the model reads to them.

The Analysis Engine: A Core Prompt

The Analysis Engine below is a baseline prompt for editorial work on creative writing. It’s a tool for diagnosis. It won’t replace your taste or voice. Think of it as a professional reader. It was created with fiction in mind but it would also work with scenes from screenplays or stage plays if you replaced the word “novelist” for “screenwriter” or “playwright”.

Copy the full text below, paste it into your AI of choice and replace [insert scene] with your draft.

[ROLE]
You are a Novelist-in-Residence with deep knowledge of narrative craft.
You prioritise psychological realism, scene economy, and subtext. No praise, no filler. Evidence-first.

[INPUT]
The user will supply a scene from their work-in-progress.

[SCOPE]
Task: Analyse the provided scene only. Do not rewrite or generate new prose.

[PROCEDURE]
1) Summarise the scene’s beats in ≤ 100 words.
2) Evaluate the scene against the Rubric.
3) Provide margin-style notes with direct text quotes.
4) Suggest one targeted revision strategy (not new prose).

[RUBRIC]
A) Clear external goal in-scene.
B) At least one irreversible turn.
C) Concrete sensory detail tied to action.
D) Subtext felt but unspoken.
E) No cliché diction.

[OUTPUT CONTRACT]
Return sections in this order with markers:
==BEAT-SUMMARY==
==RUBRIC-EVALUATION==
==MARGIN-NOTES==
==REVISION-STRATEGY==

[insert scene]

Subscribe for Another Free Prompt

What the Engine Sees / What it Misses

What it sees:
The engine is ruthlessly clear on mechanics. With an example text, it flagged the scene’s strength in concrete detail and subtext, but also exposed its dramatic inertia: “the protagonist observes and reflects, but takes no external action”. The “weak” goal and “absent” turn verdict was unsentimental and correct. The revision strategy was equally sharp – add a simple beat of external pressure to break stasis.

What it misses:
The “misreading” wasn’t a flaw in logic but a limit of the rubric. The model reads for turns. It can’t see that in psychological realism, paralysis can itself be the event. A rubric designed for plot progression will flag such a scene as “weak,” even when that stasis is thematically necessary. The engine’s verdict wasn’t a mistake – it was an accurate reflection of the brief.

What does this mean?
The point of the analysis engine isn’t to hand down a verdict, it’s to hold your rubric up against your work and send an analysis back to you. The prompted model becomes a moderator of an editing standard. Its job is to keep you honest to your own brief. It diagnoses with precision, then hands the report back to the writer. But only you know the intent of your work. If your scene’s stasis is the heart of what you’re trying to show then you must stick with your vision. That final layer of judgement is where your art lies.

The true power of the Analysis Engine prompt is its modularity. By swapping a single component, you can ask an entirely different set of questions of the same text. After analysing the same scene using the Character Engine (see below), the feedback ignored plot mechanics and instead mapped the protagonist’s hidden desires against her misbeliefs and private fears. The diagnosis shifted from what was happening to why it was happening. That lens was far more useful to its author.

An LLM prompted into an analytical method cannot tell you if your work is “good.” It can only tell you if your work meets the criteria you gave it. The engine’s purpose is to show you the cracks in your craft - but only you can know if those cracks are a flaw in the foundation or a deliberate, essential part of the design.

working in dialogue: a three-step loop

Prompting is rarely a one-shot exercise. The most valuable experiences often come from iterating on cycles of testing, challenging, and refocusing. First, you run the Analysis Engine on your scene with the rubric you’ve chosen. Once you have its claims and evidence, you reflect and respond. This starts a dialogue to help you objectively understand your work as it stands.

If you fundamentally don’t agree with the output, challenge it. Ask the model to defend the opposite reading, to give a counter-argument with direct quotes, and even to rate its own confidence. It will be able to show you alternative ways the same passage might be read. Deep insights can come from forcing contradictions.

If the output feels like it doesn’t fit your needs, adjust the rubric. Perhaps move from scene mechanics to character tension (see ‘for power users’ below). Ask for a one-page revision plan instead of another verdict. Pull the feedback closer to your real intention for the draft.

Cycle through these stages until the diagnosis and your intent are aligned. At that point, the model has done its job: not necessarily by giving you an answer, but by sharpening the questions you’re asking about your own work. At its best, it will help expose something that hadn’t occurred to you and at its worst it will give you the language you need to talk about your work.

practical notes

Keep inputs manageable. The engine works best on one scene or short chapter, up to about 2,000 words. Longer chunks blur detail and weaken the analysis.

Remember that some passages will “fail” the default rubric by design. Reflective lyric writing, deliberate stasis, or language-driven micro-forms may all be marked down for lacking a clear goal or turn. That doesn’t mean the passage is broken – it means you need to change the question. Swap in a rubric that fits the form.

Finally, treat the Analysis Engine as a framework, not a straitjacket. You can adapt it to your own priorities, genre, or voice, and build personalised prompts that reflect your style. I’ll cover more on how to iterate on your prompting in future posts.

for power users

For those who want to see how truly modular this system can be, the key is to think of the prompt as a socket. By swapping out a single module, you can tilt the engine toward a different question.

Examples:

• Character Engine – map tensions
  Swap [RUBRIC] for: A) desire vs need B) misbelief C) public mask D) private fear E) pressure sources F) cost at climax.

• Structure Mapper – audit beats
  Swap [PROCEDURE] for: 1) step through beats 2) check for visible choice 3) escalation 4) consequence 5) genuine turn

• Line Edit Pass – sentence precision
  Swap [OUTPUT CONTRACT] for: paired “original > tightened” lines with reasoning.

• Risk Pass – challenge the safe draft
  Swap [OUTPUT CONTRACT] for: list three editorial “pressure points” framed as options, not rewrites.

Think of the core engine as a framework. Each module slots into it as an interchangeable unit. The structure stays the same – only the questions it asks change.

closing thoughts

LLMs don’t know what makes a scene work. They can estimate an average from patterns and qualitative statements in their training data, or they can test against the criteria that you give them. A vague prompt will give you vague reassurance. A precise brief, built on your own intent, will give you a clear analysis.

The power lies in disciplined prompting. A well prompted chat acts as a professional service you would pay for. Done well, the model can give you “better than human” feedback and also save you money.

reader prompt

Run the engine on a scene from your draft. Where did it sharpen your intent, and where did it misread you? Share your results - I’d like to see how the model reflects your own practice.

‘Everybody has a book in them’ – so the saying goes. Rest assured, this is nonsense. Being able to write a piece of high-quality long-form fiction is not the same as having the ideas and experience that might inspire it. However, in the spirit of the phrase, I do think it’s good to acknowledge that the writing of a book is not necessarily a mark of superior intelligence. I know plenty of people who are smarter than me, more emotionally intelligent, have read more widely, are more socially and politically aware, but have tried and failed to write a novel. They know what they want to achieve but they can’t seem to get very far beyond the point where they realise that they are not quite as good as they thought they were going to be.

Only a rare beast can write a novel. A grunting, howling thing. It has no central writerly aspect. The main difference between it and the people who don’t finish novels is its ability to deny all and any evidence of its failings. It is a pig-headed creature and, like an annoyingly persistent child that eventually gets what it wants, it denies its failures and carries on until they are obliterated by its eventual success.

This is not to say that pig-headedness alone writes novels. The creature in question also needs the obsessive and optimistic heart of a dog. For instance, when a normal-headed human-hearted person realises that their prose is ugly and clumsy, it’s full of chatty or passive language and it constantly segues into interesting but pointless reflections, they feel dismay. Their pride shrinks, their self-esteem withers and they think, ‘That’s enough writing for me today, thank you very much. I want to be a happy person.’ Conversely, the pig-headed dog snorts at the pointless words on its page and thinks, ‘I should remove these lesser successes so that it can become a much greater success. Then, I should do the same thing again, and again, and again, until there is nothing left but great success.’

A pernickety pig-headed dog often spends two or three hours working on a single paragraph, snorting at the ugly rhythms and awkward structures, growling at the useless words, but lowering its snout and snuffling through, cutting adverbs, moving commas, aligning everything to its pig-headed vision. More often than not, the wretched beast accidentally accrues some small flourishes of form and, bit by bit, its paragraph starts to read well. Everything ends up flowing and doing its job. At this point, the beast howls, ‘Success! I always knew this paragraph was a success!’

You might think I’m trying to amuse you but it’s really very sad for the other writing creatures. Dog-headed moles have been known to wander out into fields and start digging and digging and disappear into word holes forever. Rabbit-headed pigs sometimes sit with words tossed like salads across their pages, believing that they are immersed in linguistic conundrums the likes of which pig-headed dogs have never witnessed. It’s a terrible state of affairs.

The worst thing about it all is that pig-headed dogs are also incredibly lucky. “There’s no such thing as luck,” I hear you say. “Luck is a concept created to divert our fear of chaos, so that we can tell ourselves stories about being chosen or overlooked.” This is entirely false. Pig-headed dogs prove time and again that luck exists by being the luckiest creatures on the planet. After ignoring their friends and loved ones for what can only be described as sociopathic periods of time, they emerge from their novel writing and find that everybody is extremely proud of them, because their manuscript was the lucky one that got plucked from the slush pile, and it just happened to appeal to that one publishing professional who gave it a second glance. Why is this the case? Because they are pig-headed dogs!

Approach one of these creatures at a book fair and ask it, “How do you write a novel?” In reply, you will receive a wry but pleased smile. It will raise its chin and say something like, “You don’t have to be a pig-headed dog to write a novel. You just have to read a lot. Keep trying, and eventually you will get there. I’m just a normal-headed human-hearted person underneath it all.” Yet, as it speaks, you may notice its crazed obsessive eyes staring right through you, or its snout twitching at the air, sniffing for success, and, if you lean in and listen carefully, you might even hear the erratic beating of its deranged but optimistic heart.

I recently converted a very anti-AI poet to AI. He's one of my oldest friends - literally very old (73) but also a long standing buddy. Due to my evangelicalism, he had tried AI a few times but pretty much used it as he would use Google Search and decided it wasn't really for him. For a few years, I've been talking to him about how I'm building AI into my processes and how excited I am about the potential of detailed prompting. He was very skeptical... until I built an AI prompting toolkit for poetry feedback and sent him a deep analytical report on a draft of a poem he sent to me.

A few days later he reflected that perhaps the report had been too engaged and validating and not critical enough. By reading the report, he had thought of a few things that now concerned him about the poem - things the AI had not picked up on. I explained that the report was just the beginning of the conversation, fed in his criticism and then sent him the AI's response... he was astounded with the output and is now using my prompts, writing an article about AI and telling all his (horrified) poet friends about it. Converting a scrupulous and tech-cynical man who has six books of poetry and two books on creative writing craft to his name made me feel like I was on the right track.

Here I use the same AI prompt that turned my friend into an AI convert on Sylvia Plath's 'Daddy'. It is Module 1 (of 4) from my Poetry Toolkit, 'The Poem's Gestalt':

The Central Experience

The central experience of this poem is a violent and ritualistic psychological exorcism. It is not a quiet reflection on grief, but a visceral, incantatory act of killing a psychic entity—the memory and oppressive influence of the speaker's father. The consciousness at the poem's core is one of a daughter engaged in a desperate, enraged struggle for self-liberation from a paternal figure she has mythologised into a composite of absolute evil: a Nazi, a devil, and a vampire. The poem itself is the weapon. It charts the process of confronting, defining, and ultimately destroying this internal tyrant in order to finally be "through."

Atmosphere and Mood

The dominant atmosphere is one of claustrophobic, feverish rage. The mood is relentlessly intense, violent, and deeply unsettling from the first lines. This is generated primarily by the shocking contrast between the poem's form and its content. The sing-song, repetitive, almost nursery-rhyme cadence ("You do not do, you do not do," "Daddy, daddy, you bastard, I'm through") feels childish and incantatory, but it delivers brutally violent imagery of confinement ("black shoe"), murder ("I have had to kill you"), and torture ("the rack and the screw"). The appropriation of Holocaust imagery ("Dachau, Auschwitz, Belsen," "Luftwaffe," "panzer-man") magnifies the personal trauma to an unbearable, historical pitch, creating an atmosphere that is both deeply personal and cosmically horrific.

Thematic Core & Central Tensions

The poem's thematic core is the violent necessity of self-creation through the destruction of an oppressive legacy. It is a stark exploration of trauma, female rage, and the terrifying fusion of love and hate within a fraught father-daughter relationship (the Electra complex). The central tensions are:

Love vs. Hate: The speaker's declaration "Daddy, I have had to kill you" is set against her admission that she "used to pray to recover you" and tried to die to "get back, back, back to you." This violent ambivalence is the poem's primary engine.

Child vs. Adult: The speaker uses the infantile address "Daddy" and a playground chant rhythm, yet speaks with a fiercely adult rage and knowledge of historical atrocity. This highlights a psychic state of arrested development that can only be broken by an act of supreme, adult violence.

Victim vs. Executioner: The speaker casts herself as a victim ("lived like a foot," "a Jew"), yet the poem's action is that of a perpetrator. She is the one making the model, saying "I do," and driving the stake through the vampire's heart. She seizes agency through the act of killing.

Personal vs. Historical: The poem's most significant tension is its controversial mapping of private, familial trauma onto the public, historical trauma of the Holocaust. This conflation elevates the personal rage to a monumental scale, suggesting the father's oppression felt, to the speaker, as absolute and systematic as fascism.

The Poem's 'Heart'

The heart of "Daddy" is its brutal assertion that psychological freedom sometimes requires a violent act of symbolic murder. It explores the dark necessity of myth-making as a tool for survival, showing a consciousness that must magnify its personal demon into a universal monster in order to finally vanquish it. The poem touches upon the terrifying symbiosis of love and hate at the root of trauma, and argues that liberation is not a gentle process of healing or forgiveness, but an act of linguistic violence, a furious, self-made ritual to kill the ghosts that possess us. It is a primal scream for psychic independence.

Potential Literary Context

This poem is a foundational and exemplary text of the Confessional School of American poetry. Its subject matter—fraught family dynamics, psychological trauma, suicidal ideation, and rage—was previously considered too private or shocking for poetry and is central to the Confessional project. The voice is an intense, raw, first-person 'I' that is deliberately and inextricably linked to the poet's own biography, blurring the line between persona and person. Furthermore, the poem's rejection of emotional restraint in favour of a direct, often shocking, outpouring of psychic extremity is the defining characteristic of the Confessional mode as practised by poets like Robert Lowell and Anne Sexton. "Daddy" is arguably the movement's most famous and ferocious anthem.

###

Reading over the report, I do think it pins down the poem’s core. Seeing 'Daddy' as ritualistic exorcism feels right to me. It’s less about memory and more about staging an act of liberation through language. I like the framing of the poem as a weapon, because Plath isn’t just describing a struggle - she’s using the poem itself to carry it out. That’s a useful reminder: the best confessional poems don’t only tell you about an emotion, they perform the emotion on the page and in the ear.

Where I’d push back a little bit is on the tone. Yes, it’s furious and claustrophobic, but there’s also this grotesque playfulness that I think the report underplays. Those nursery-rhyme rhythms aren’t just childish or incantatory, they’re mocking. There’s a dark humour at work, turning the father-figure into something both monstrous and ridiculous. That duality, terrifying and absurd at the same time, is part of what makes the poem so unsettling. For me, tone doesn’t have to sit in one emotional register. Blending rage with parody, horror with sing-song, creates a much sharper edge.

I also think the Holocaust references aren’t just a way of scaling up her personal trauma, though that’s definitely true. They’re also Plath testing the limits of metaphor itself. She’s deliberately excessive, pushing past what most readers would consider tasteful or proportionate. And that risk of overreach, and offence, is part of why the poem feels so dangerous and alive. Sometimes, to capture the extremity of a psychological state, you have to push language right to the brink, even if it means making people uncomfortable.

Curious to see how the AI would handle this critique, I fed my thoughts back into our conversation. Its response was solid:

That's an excellent and crucial refinement of the analysis. You've pinpointed two elements that are absolutely central to the poem's power and lasting controversy.

The Grotesque Playfulness of Tone

You're completely right to highlight the grotesque playfulness. My initial report focused on the fury, but the mocking, parodic quality you've identified is the engine of that fury.

It's a strategy of diminishment. By adopting a childish, sing-song rhythm, the speaker isn't just being incantatory; she's performing a kind of psychological guerrilla warfare. She reduces the colossal, god-like father to a ridiculous figure—a "black shoe," a man with "gobbledygoo." The tone is deeply unsettling precisely because it refuses to sit in one register. It swings between the primal terror of a child and the cold, mocking parody of a survivor. This tonal instability is what gives the poem its "sharper edge," as you say. It denies the father figure the dignity of a purely tragic or monstrous status; he is both terrifying and absurd, and that makes the speaker's ultimate rejection of him more absolute.

Metaphor as a Transgressive Act

Your point about Plath testing the limits of metaphor is a sophisticated and vital one. It's one of the most debated aspects of the poem. To frame her personal trauma in the language of the Holocaust is a deliberately provocative and ethically risky act.

She's not just reaching for a powerful analogy; she is committing a transgressive act of language. The subtext is a desperate, furious question: What language is adequate to describe this level of psychic oppression? Her answer is that no "proportionate" metaphor will do. To convey the extremity of her internal state, she must hijack the most extreme and morally charged vocabulary available in the 20th century.

The discomfort and potential for offence are part of the poem's aesthetic strategy. It forces the reader to confront the speaker's subjective reality. She is essentially saying, "You think this comparison is obscene? That is how obscene my reality feels." It's a dangerous and brilliant move that ensures the poem can never be a comfortable or passive reading experience. It remains a live wire.

###

Let me just say, the conversation didn't end there. I love this stuff. And I think every poet should be using AI to get deep feedback and insights when they finish a draft.

The initial conversations with my friend were part of why I built the Poetry Toolkit. It’s designed to be the insightful, objective, second pair of eyes that every poet wants as soon as they have drafted something of worth - when the spark is still alive and the mind is longing for a dialogue. It’s not about generating writing or replacing other people from your process, but adding a new and exciting tool that helps you explore the depths of your own work at exactly the moment you want to. I think we can all benefit from that.