Michel de Certeau

Foreword

By Edo Segal

The shortcut through the alley is what nobody measures.

I have been staring at dashboards for thirty years. Adoption curves, engagement metrics, revenue per seat, lines of code generated per hour. The view from above. The planner's view. The view that turns a city into a grid and a workforce into a spreadsheet and a human being into a data point moving through a funnel someone else designed.

Then I read Michel de Certeau, and he told me to look at the street.

Not the street as designed. The street as walked. The specific, unrepeatable path that one person cuts through a system built by someone else — the shortcut no map records, the pause at the window that has nothing to do with the destination, the way the commuter knows which subway car will stop closest to the exit at her station, not because anyone told her but because she has ridden that train a thousand times and her body learned what the schedule could not teach.

De Certeau spent his career studying this. The creativity of people who were never supposed to be creative. Not the painter in the studio. The cook in the kitchen. The tenant navigating a housing bureaucracy. The reader who takes an author's text and makes something from it the author never intended. He called these people tacticians — not because they lacked strategy but because they lacked territory. They operated inside systems they did not design, and their art consisted in making those systems yield something personal, something real, something the designer could not have predicted.

I recognized myself in both positions. I have been the planner — building platforms, designing systems, watching from above as millions of users moved through spaces I helped create. And I have been the walker — sitting at my desk with Claude, navigating an output space I did not train, finding paths the documentation does not describe, producing work that belongs to the collaboration in ways no metrics dashboard can capture.

De Certeau died in 1986. He never saw a computer in any sense we would recognize. But his framework is the most precise instrument I have found for understanding what actually happens when a person sits down with an AI and builds. Not what the benchmarks say happens. Not what the adoption curve says happens. What the practitioner does with the system's output — the selecting, the refusing, the quiet daily act of making something yours from materials that were never designed for you.

The view from the 110th floor cannot see this. The view from the street cannot see anything else.

This book is written from the street.

-- Edo Segal ^ Opus 4.6

About Michel de Certeau

1925-1986

Michel de Certeau (1925–1986) was a French Jesuit scholar, historian, and cultural theorist whose work ranged across mysticism, psychoanalysis, historiography, and the study of everyday life. Born in Chambéry, France, he was ordained as a Jesuit priest and earned doctorates in both theology and history. His most influential work, *The Practice of Everyday Life* (*L'invention du quotidien*, 1980), introduced a framework for understanding how ordinary people creatively navigate and subvert the systems imposed upon them — distinguishing between the "strategies" of institutions that control territory and the "tactics" of individuals who operate within those structures without owning them. His concepts of tactical practice, poaching, and *bricolage* as forms of everyday resistance have shaped fields from cultural studies and urban theory to media studies and design. Other major works include *The Writing of History* (*L'écriture de l'histoire*, 1975) and *The Mystic Fable* (*La fable mystique*, 1982). De Certeau held positions at the University of Paris and the University of California, San Diego. He died in Paris at the age of sixty, leaving a body of work that continues to provide some of the most incisive tools available for understanding how people make meaning within systems they did not choose.

Chapter 1: The City and the Walker

On the 110th floor of the World Trade Center, the observer looks down at Manhattan and sees a city that makes sense. The grid reveals itself. The avenues run north and south with mathematical discipline. The blocks repeat. The traffic flows in patterns that, from sufficient altitude, resemble the circulation of blood through an organism designed by an engineer who valued efficiency above all else. From up here, the city is a text — legible, coherent, available to the reading eye of the planner, the administrator, the strategist who needs to understand the whole in order to govern it.

Michel de Certeau stood on that floor and described what he saw with the precision of a man who understood that the view from above is itself a form of power. The panoramic gaze does not merely observe the city. It produces the city — or rather, it produces a version of the city that is governable, that submits to the logic of the map, that becomes available to the operations of those who plan, zone, regulate, and control. The view from the 110th floor is the view of the strategist. It is the view of the institution that possesses territory, that has carved out what de Certeau called a lieu propre — a proper place, a place of one's own — from which to survey, calculate, and act upon the world below.

Then de Certeau descended to the street. And what he found there was a different city entirely.

The walker does not see the grid. The walker experiences the block she is crossing, the corner she is turning, the shortcut through the alley that no map records, the pause at the window of a shop that has nothing to do with her destination but everything to do with the texture of her morning. The walker produces, through the act of walking, a text of her own — an invisible writing laid down on the surface of the city that the planner's map cannot capture, because the map records streets but not the specific, unrepeatable way this particular person navigates them on this particular Tuesday in November. The walker takes the city that the planner designed and makes something else of it. Something personal. Something tactical.

This distinction — between the strategic view from above and the tactical practice from below — is the most fertile conceptual framework that de Certeau produced in a career spent attending to the creativity of people who were never supposed to be creative. The distinction does not map onto the familiar binary of powerful and powerless, though power is certainly at stake. It maps onto something more precise: two fundamentally different operations that produce two fundamentally different relationships to the systems within which human beings live, work, think, and create.

A strategy, in de Certeau's usage, is the operation of a subject possessed of will and power — an institution, a corporation, a scientific establishment, a technological platform — that can circumscribe a place as its own and generate relations with an exterior conceived as a target, a threat, or a market. The strategy requires territory. It requires the capacity to define a space, to establish boundaries, to impose order within those boundaries, and to produce knowledge of the exterior from within the safety of the lieu propre. The urban planner who designs the grid is executing a strategy. The corporation that builds a platform and defines the terms of service within which millions of users must operate is executing a strategy. The AI company that trains a model on the corpus of human text, defines the architecture within which that model generates responses, establishes the alignment procedures that constrain what the model may and may not say, and deploys the resulting system as a service — that company is executing a strategy of extraordinary scope and ambition.

A tactic, by contrast, is the operation of a subject who does not possess territory. The tactic has no lieu propre. It operates within the space defined by the strategy, and its art consists in making use of that space in ways the strategist did not intend, could not foresee, and frequently cannot detect. The tactic is opportunistic. It seizes the moment. It does not plan campaigns; it exploits openings. It is, in de Certeau's vivid formulation, the art of the weak — not because the tactician lacks intelligence or skill, but because the tactician lacks the institutional power to define the terrain on which the encounter takes place.

The walker is a tactician. She navigates a city she did not design, using streets she did not lay down, moving through a spatial order imposed by planners and zoning boards and real estate developers whose strategic calculations had nothing to do with her specific need to get from the bakery on Seventh Street to her daughter's school on Fourteenth. Yet her walking is not passive. It is not mere compliance with the grid. Her walking produces something — a personal geography, a habitual route, an embodied knowledge of the city that no map contains and no algorithm can fully reconstruct. She knows that the sidewalk on the west side of Ninth Avenue is broken in a way that makes the stroller catch, so she crosses to the east side at Twelfth Street, even though the planner's grid suggests no reason to cross. She knows that the coffee shop on the corner of Eleventh opens at six-thirty, not seven, despite what the sign says, because she has been there enough mornings to learn the rhythm of the owner's life. This knowledge is tactical. It is produced through practice, not through the strategic accumulation of data. And it is genuinely creative, because it makes the city into something the planner never designed — a personal, lived, inhabited place rather than an abstract, strategic space.

The AI platform is the planned city of the digital age.

The parallel is not metaphorical. It is structural. An AI system — a large language model trained on billions of tokens of text, deployed through an interface designed by a product team, constrained by alignment procedures determined by a safety team, offered as a service under terms defined by a legal team — is a strategic territory of extraordinary complexity. It defines a space of possible outputs. Every response the model generates is produced within the boundaries of that space, shaped by the architecture, the training data, the fine-tuning, the reinforcement learning from human feedback, the system prompts that frame the interaction. The user who opens a conversation with the model enters a territory she did not design, governed by rules she did not set, shaped by decisions she was not consulted about and in most cases cannot see.

Yet the user is not passive. The user walks.

She navigates the model's output space the way the walker navigates the planned city — finding paths the designers did not foresee, discovering capabilities the documentation does not describe, producing works from the territory's resources that the territory itself cannot account for. The prompt engineer who discovers that a specific sequence of instructions unlocks a capability the model's designers did not explicitly intend is a walker finding a shortcut through an alley. The developer who uses a coding assistant to build a product in a domain the assistant was not specifically trained for is a pedestrian crossing to the east side of the avenue because she knows something about the terrain that the map does not record. The writer who takes the model's output and reshapes it, discards the generic, keeps the surprising, inflects the remainder with a meaning that lives nowhere in the training data — she is writing her own text on the surface of the city, producing a work that belongs to neither the planner nor the street but to the specific, unrepeatable practice of a particular person navigating a particular system at a particular moment.

This is not a diminished form of creativity. De Certeau's entire intellectual project was the insistence that it is creativity's most common and most undervalued form.

The strategic view — the view from the 110th floor, the view of the platform company, the view of the AI researcher who sees the model's capabilities from the vantage of the architecture — produces a specific kind of knowledge. It produces the knowledge of the system: what the model can do, what it cannot do, where its capabilities end and its failures begin. This knowledge is real and valuable. It is the knowledge required to build and improve the system. But it is also partial, because it cannot see what the walkers are doing down on the street.

The strategic gaze observes usage patterns, aggregate metrics, the statistical distribution of prompts and responses across millions of interactions. It sees what de Certeau would call the espace géométrique — the geometric space of the system as designed. What it cannot see is the espace anthropologique — the lived space of the system as inhabited. The specific, local, biographical knowledge that each user brings to the interaction. The way one builder's engagement with the model differs from another's, not because they use different prompts but because they bring different histories, different judgments, different senses of what counts as good enough. The way a teacher in Lagos uses the same model as an engineer in Bangalore to produce a result that neither could have produced without the model but that the model, left to its own strategic logic, would never have generated on its own.

De Certeau spent his career insisting that the view from the street matters — not as a sentimental corrective to the view from above, but as an analytical necessity. The system cannot be understood solely from the strategist's vantage. Its actual operation, its meaning, its consequences for the people who live inside it, can only be understood by attending to the practices of those who navigate it daily: the tactical operations, the small creative acts, the appropriations and redirections and productive misuses that constitute what de Certeau called the practice of everyday life.

This insistence carries a specific and urgent relevance in the age of artificial intelligence, because the dominant discourse about AI is almost entirely strategic. It is conducted from the 110th floor. The AI companies publish capability benchmarks. The researchers measure performance on standardized tasks. The policy documents assess risk at the population level. The discourse asks: What can the system do? What are its limits? How should it be governed?

These are important questions. They are also, in de Certeau's framework, radically incomplete. They describe the city but not the walking. They describe the territory but not the practice. They describe the strategy but not the tactics — and the tactics are where the creative life of the system actually resides.

What the walker does with the city — the specific, personal, irreducible practice of inhabiting a system designed by others — is the question that de Certeau's framework insists we ask. Not as an afterthought. Not as a humanistic supplement to the technical analysis. As the central question, the question without which the system's meaning remains opaque even to those who built it.

The builder who sits down with an AI system and makes something is walking in a city she did not design. The streets were laid by engineers at Anthropic or OpenAI or Google. The grid was drawn by the architecture of the transformer. The zoning was determined by the training data, by the alignment procedures, by the terms of service, by the economic model that makes the system available. None of this was the builder's doing. She enters the territory as a walker enters the city: without control of the infrastructure, without knowledge of the full design, without the strategic vantage that would allow her to see the system whole.

And yet she builds. She navigates. She takes shortcuts the designers did not foresee. She discovers capabilities the benchmarks did not measure. She produces, through the specific quality of her engagement with the system, a work that the system's strategic logic cannot account for — a work that is genuinely hers, not because she originated the materials, but because she walked the city in a way no one else has walked it.

De Certeau did not live to see the AI revolution. He died in Paris in January 1986, at sixty, of pancreatic cancer. He never used a computer in any sense that would be recognizable to a contemporary user. He never encountered an interface that could generate text, produce code, or engage in the kind of extended dialogue that characterizes the modern AI interaction.

But his intellectual contribution was precisely the framework that the AI moment most urgently requires. The framework that attends not to the system's design but to the practitioner's art. Not to what the technology can do but to what people do with the technology. Not to the strategy but to the tactic — the quiet, dispersed, nearly invisible creativity of ordinary people navigating extraordinary systems, making meaning from materials they did not choose, finding freedom not in the redesign of the city but in the quality and specificity of their walking.

That framework is the subject of this book. And the first step, before anything else, is to understand what de Certeau meant by poaching — the particular art through which the walker makes the city her own.

Chapter 2: The Art of Poaching

The reader is a poacher. De Certeau insisted on this image with the stubbornness of a man who had spent years watching literary theorists describe reading as though it were a form of obedience — the dutiful reception of the author's meaning, the faithful reconstruction of the text's intended significance, the passive consumption of a product manufactured by one consciousness for another. The entire apparatus of literary criticism, de Certeau observed, was built on the assumption that the author produces and the reader consumes, that meaning flows in one direction — from the quill to the eye, from the active to the passive, from the creative to the receptive.

De Certeau overturned this assumption with a single, quietly devastating observation: readers do not do what texts tell them to do. They never have. They wander through the text the way nomads wander through territory that belongs to others — taking what they need, leaving the rest, combining what they find with materials gathered on other journeys, producing from the text a meaning that the author did not plant and cannot control. The reader who skips the exposition and lingers on the dialogue is poaching. The reader who remembers a single image from a three-hundred-page novel and builds an entire emotional architecture around that image is poaching. The reader who misreads — who takes an ironic passage at face value, or reads a political allegory as a love story, or finds in a scientific paper the seed of a poem — is poaching with a creativity that the text's author would neither recognize nor authorize.

This is not a failure of reading. It is what reading actually is. The text is not a container of meaning waiting to be opened. It is a territory to be traversed, and the traversal — the specific, biographical, unrepeatable path that each reader cuts through the textual landscape — is the creative act. The meaning of the text is not in the text. It is in the practice of reading, which is always a practice of appropriation, of making the author's materials serve the reader's purposes.

De Certeau derived this insight partly from his close engagement with Roland Barthes, who had declared the death of the author and the birth of the reader as the locus of textual meaning. But de Certeau went further than Barthes in a direction that matters profoundly for the present argument. Barthes described the reader's creativity in semiotic terms — as an operation performed on signs. De Certeau described it in practical terms — as a way of living in the world, a mode of habitation, a form of everyday creativity that connects reading to cooking, walking, talking, and every other practice through which ordinary people make meaning from materials they did not produce.

The poacher does not own the forest. She has no estate, no hunting rights, no strategic claim to the territory. She enters the landowner's property without authorization, takes what she needs — a rabbit, a handful of berries, firewood — and disappears back into the landscape from which she came. The landowner, surveying his territory from the strategic vantage of the manor house, may not even notice the poaching has occurred. The poacher's operation is invisible from above, because it does not alter the territory. It appropriates the territory's resources without challenging its structure. The forest remains the landowner's forest. The rabbit, however, is now the poacher's dinner.

The analogy is precise. The text remains the author's text. The novel sits on the shelf, unchanged by the reader's passage through it. But the meaning the reader has extracted from the text — the personal significance, the emotional resonance, the connection to her own experience, the insight that arrived not because the author intended it but because the reader's particular biography created the conditions for it to appear — that meaning belongs to the reader. She has poached it. It is hers.

Now consider the builder who opens a conversation with a large language model and receives a response.

The model's output is a text in exactly de Certeau's sense — a structured production generated by a system that the builder did not design, trained on materials she did not select, shaped by architectural decisions and alignment procedures and commercial imperatives that constitute the lieu propre of the platform company. The output is strategic production. It is the product of a system that controls its own territory and generates relations with an exterior — the user — conceived simultaneously as customer, data source, and potential risk.

The builder reads this output. And in reading it, she poaches.

She does not accept the output whole. She does not treat the model's response as a finished product to be consumed as delivered. She evaluates. She selects. She takes this paragraph and discards that one. She keeps the structural suggestion but rewrites the prose. She notices a connection the model has drawn between two ideas and recognizes it as genuine insight — but she also notices that the connection depends on a reference the model has misattributed, a philosophical concept it has used with confident inaccuracy, and she discards the reference while keeping the connection, which is now hers rather than the model's because she has done the work of evaluating it against her own knowledge and finding it true despite the model's false supporting evidence.

This is poaching. The builder enters the model's textual territory, takes what she needs, leaves what she does not, and produces from the appropriated materials a work that the model's strategic logic cannot account for. The model generated the raw text. The builder generated the meaning — not by originating the words but by performing the specific, skilled, evaluative operations that de Certeau recognized as the creative core of reading.

The authorship question that preoccupies so much contemporary discourse about AI-assisted creation dissolves in this framework. The question "Who wrote this?" assumes that authorship is origination — that the author is the person who produced the raw material, the words on the page, the code in the file. De Certeau's analysis reveals this assumption as a strategic illusion, a way of privileging the producer's contribution over the consumer's, the author's text over the reader's practice, the system's output over the user's appropriation. De Certeau never believed in origination as the locus of creative value. All texts are made from other texts. All creative production is recombination, appropriation, poaching on what has come before.

What matters is not where the materials come from. What matters is what the practitioner makes of them through the quality of her engagement.

When Segal describes his collaboration with Claude in The Orange Pill, the moments that most closely match de Certeau's framework are the moments of evaluative crisis — the instances where the model's output had to be tested against the builder's judgment. The Deleuze failure is the paradigmatic case. Claude produced a passage connecting Csikszentmihalyi's flow state to a concept it attributed to Gilles Deleuze, and the passage was elegant, well-structured, rhetorically effective. It read like insight. It felt true in the way that polished prose often feels true — through the seduction of form rather than the verification of content. Segal almost kept it. Then he checked. The philosophical reference was wrong in a way that would be immediately apparent to anyone who had actually read Deleuze.

The poacher catches the gamekeeper's trap. The model's output was not merely wrong — it was wrong in a way that was concealed by the smoothness of its presentation. The strategic production had generated a text whose surface quality masked a structural failure. The practitioner's reading — the specific, evaluative, skeptical reading of someone who brings her own knowledge to the encounter — caught what the strategic system could not catch in itself, because the system lacks the capacity for the kind of external verification that de Certeau called the reader's art.

This is not a failure of AI. It is a demonstration of why the tactical operation matters. The model will produce output that is statistically optimized across the distribution of its training data. That optimization will generate responses that are, on average, remarkably capable. But the average is the strategist's metric. It describes the system's performance across a population of interactions. It cannot account for the specific needs of a specific practitioner working on a specific problem at a specific moment. That accounting is the tactic's work — the work of the reader who poaches the text, taking what serves and refusing what does not.

De Certeau recognized a further dimension of poaching that is directly relevant to the AI case. The poacher's creativity is not random. It is shaped by what de Certeau called the habitus — not quite in Bourdieu's deterministic sense, but in the broader sense of the accumulated dispositions, skills, and sensibilities that a specific biography produces. The reader who finds the unexpected connection in the text does not find it by accident. She finds it because her particular history — her education, her emotional life, her professional experience, her specific constellation of knowledge and ignorance — has created the conditions for that connection to become visible. A different reader, with a different biography, would poach a different meaning from the same text.

The same principle governs AI-assisted creation. Two builders, given identical models and identical prompts, will produce radically different works — not because the model's output differs (it might, given stochastic variation, but the variation is not the point) but because the builders' practices of poaching differ. One builder's biography disposes her to see connections in the model's output that the other builder misses entirely. One builder's professional experience gives her the evaluative framework to catch errors that the other builder accepts. One builder's aesthetic sensibility leads her to keep the rough, unexpected phrase and discard the polished, generic one, while the other builder does the reverse.

The poaching is specific. It is biographical. It is irreducible to the model's strategic logic, because it depends on what the practitioner brings to the encounter rather than on what the system provides. And this specificity — this irreducible quality of the practitioner's engagement — is where de Certeau located the genuine creativity of reading, of walking, of cooking, and of every other practice through which ordinary people make meaning from the materials the world provides.

The art of poaching is not a consolation prize for those who cannot originate. It is the actual mechanism of cultural production, now made visible by a technology that has automated the strategic dimension of creation — the generation of raw material — and thereby revealed the tactical dimension as the locus of whatever we mean by authorship, voice, and creative significance.

When de Certeau described the reader as a poacher, he was not making a political argument about the rights of readers versus the rights of authors. He was making an analytical argument about where creativity actually lives. It does not live in the production of raw material. It lives in the practice through which raw material is appropriated, evaluated, recombined, and inflected with the specific meaning that only a specific practitioner, with a specific biography, facing a specific need, can provide.

The AI model produces the forest. The builder poaches the dinner. And the quality of the dinner depends not on the abundance of the forest — which, in the age of large language models, is effectively infinite — but on the skill, the judgment, and the specific biographical disposition of the poacher.

That skill has a name. De Certeau called it bricolage — the art of making do with what is at hand. It is the subject of the next chapter, and it is the operational core of every act of AI-assisted creation that deserves to be called creation rather than extraction.

Chapter 3: Making Do — Bricolage and the Practitioner's Craft

Claude Lévi-Strauss drew a line between two kinds of minds. On one side, the engineer: the figure who works from first principles, who designs before building, who selects optimal materials for predetermined purposes, who operates within a rational plan that specifies in advance what the finished product will be. On the other side, the bricoleur: the figure who works from whatever happens to be available, who assembles from scraps, who improvises with the materials at hand, who does not begin with a blueprint but with a collection of oddments — tools that have outlived their original purpose, leftover materials from previous projects, objects whose potential use is not yet determined — and who produces from this heterogeneous collection something that was not planned in advance but that works, that serves, that possesses a coherence discovered in the making rather than imposed from without.

Lévi-Strauss intended the distinction as an analysis of mythical thought versus scientific thought. De Certeau appropriated it — poached it, one might say — for a different and ultimately more radical purpose. In de Certeau's hands, bricolage became the defining operation of everyday practice. The cook who opens the refrigerator and surveys what is available — half an onion, some leftover rice, an egg approaching its expiration, a jar of olives that has been sitting at the back of the shelf for longer than anyone can remember — and produces from these unpromising materials a meal that feeds her family: she is a bricoleur. The commuter who constructs a bearable journey from the materials of an unbearable transit system — choosing this car because it is less crowded, reading at this stop because the light is better, standing in this position because the door will open here at her destination — she is a bricoleur. The student who assembles an essay from lecture notes, half-understood readings, a conversation overheard in the dining hall, and a vague but persistent intuition that these materials connect in a way she cannot yet articulate — she is a bricoleur.

The bricoleur does not originate. She makes do. And the making-do is itself a creative operation of considerable sophistication, because it requires the practitioner to see potential in materials that were not designed for the purpose she is putting them to, to judge which materials can be adapted and which must be discarded, to improvise combinations that no blueprint anticipated, and to produce from the whole assemblage something that works — not optimally, not elegantly, not in the way the engineer's planned creation works, but in the specific, situated, good-enough way that actual life requires.

De Certeau recognized in bricolage the operational logic of every practice he studied. The walker who takes the shortcut through the alley is making do with the city's spatial resources. The reader who extracts a personal meaning from the author's text is making do with the text's semantic resources. The consumer who repurposes a product — using a butter knife as a screwdriver, a newspaper as an umbrella, a grocery bag as a garbage liner — is making do with the material resources available to her. In each case, the bricoleur's creativity consists not in the production of new materials but in the skilled appropriation and recombination of existing ones.

This framework illuminates AI-assisted creation with an analytical precision that more familiar frameworks — the artist and the tool, the author and the assistant, the human and the machine — cannot match.

The builder who sits down with a large language model is a bricoleur. She does not begin with ideal materials. She begins with whatever the model produces — a response shaped by the statistical regularities of the training data, by the specific prompt she has formulated, by the stochastic variation of the model's generation process, by the alignment constraints that filter the output before it reaches her screen. These materials are heterogeneous, imperfect, and not designed for her specific purpose. The model's response may contain a structural suggestion that is brilliant, a factual claim that is wrong, a stylistic choice that is generic, and a connection between ideas that is genuinely illuminating — all in the same paragraph. The builder's task is to sort through this heterogeneous collection, judge what can be used and what must be discarded, improvise combinations that the model's output did not anticipate, and produce from the whole assemblage something that serves her specific need.

The quality of the bricolage depends not on the quality of the raw materials — which AI provides with a generosity that would have been unimaginable to any previous generation of makers — but on the quality of the practitioner's judgment about what to do with them. The same pile of oddments, in the hands of different bricoleurs, yields different results. One cook produces from the half-onion, the rice, and the egg a dish of quiet competence. Another produces something memorable, something her family will ask for again, not because the ingredients were better but because the judgment about how to combine them — the decision to add the olives, the intuition that a splash of vinegar would lift the whole — was more skilled, more attentive, more responsive to the specific situation.

This is the argument that the democratization of AI-assisted building both confirms and complicates. When Segal describes in The Orange Pill the engineer in Trivandrum who had never written frontend code and who, within days of working with Claude, was building complete user interfaces — she was not performing engineering in the traditional sense. She was performing bricolage. She was taking the model's output — code she had not written and could not have written from scratch — and combining it with her existing knowledge of backend systems, her understanding of the product's requirements, her accumulated intuition about what users need, to produce something that neither she nor the model could have produced alone. The model supplied the materials. She supplied the judgment about how to use them.

Lévi-Strauss's original distinction between the engineer and the bricoleur carried an implicit hierarchy: the engineer works at a higher level because the engineer works from first principles rather than from happenstance. De Certeau explicitly rejected this hierarchy. For de Certeau, bricolage is not a lesser form of creation. It is the form of creation that most closely matches the actual conditions under which most people work most of the time. The engineer who designs from first principles, who selects optimal materials, who works from a complete blueprint — this figure exists, but she exists in conditions of exceptional privilege: institutional support, adequate resources, time to plan. The bricoleur is the rest of us: working with what we have, under constraints we did not choose, producing something good enough from materials that were never quite right for the purpose.

The AI revolution has not eliminated the distinction between the engineer and the bricoleur. It has shifted the terrain on which bricolage operates. The materials available to the bricoleur have become vastly more abundant. The code snippets, the structural suggestions, the analytical frameworks, the stylistic templates, the factual references — all of these are now available in quantities and at speeds that would have been inconceivable ten years ago. The half-onion has become a full pantry. The scraps of wood have become a lumber yard.

But the bricoleur's essential operation has not changed. She still must judge. She still must select. She still must decide which materials serve and which do not, which combinations will hold and which will collapse, which improvisations are brilliant and which are lazy. The abundance of materials does not reduce the need for judgment. It increases it, because the number of possible combinations has exploded, and the practitioner must navigate a vastly larger space of possibilities with the same finite cognitive resources she has always had.

This is where de Certeau's analysis converges with what The Orange Pill calls ascending friction — the principle that each significant technological abstraction removes difficulty at one level and relocates it to a higher cognitive floor. The difficulty of obtaining materials has been radically reduced by AI. The difficulty of judging what to do with those materials has been radically increased, because the space of possibilities is now so large that navigation through it requires a sophistication that was not necessary when the materials themselves were scarce.

The cook who opens a refrigerator containing three ingredients faces a constrained problem. The constraints themselves guide the solution — there are only so many things you can make with half an onion, some rice, and an egg. The cook who opens a refrigerator containing everything faces an unconstrained problem, and the unconstrained problem is harder, not easier, because the constraints that previously guided the solution have been removed, and the cook must now supply from her own judgment the direction that scarcity once provided.

The practitioner's judgment in the age of AI bricolage operates along several axes simultaneously. There is the evaluative axis: Is this output correct? Does it do what I need it to do? Does it contain the specific error — the confident wrongness dressed in polished prose — that de Certeau would recognize as the strategic system's characteristic failure, the failure that only the tactical practitioner can catch? There is the aesthetic axis: Is this output good? Not merely functional but genuinely well-made, possessing the specific quality that distinguishes work worth keeping from work that merely fills space? There is the contextual axis: Does this output serve the specific need I am addressing, given the specific constraints I am working within, for the specific audience I am trying to reach? The model's output is optimized across a population. The practitioner's need is singular. The gap between population-level optimization and individual-level specificity is where the bricoleur's judgment operates.

And there is a fourth axis that de Certeau would have recognized immediately, though he would not have used these words: the axis of voice. The model speaks in what de Certeau called the rhetoric of the system — general, pattern-based, statistically central. The practitioner speaks in what de Certeau called the rhetoric of practice — specific, contextual, personally inflected. The bricoleur who works with AI output must perform a continuous act of translation between these two rhetorics — taking the system's generality and inflecting it with the specificity that genuine quality requires. The word that is almost right but not quite. The structure that is logical but lifeless. The argument that is sound but lacks the particular angle of approach that would make it not merely correct but illuminating. These are the seams where the bricoleur's work becomes visible, where the tactical operation separates itself from the strategic production, where the practitioner's voice emerges from the system's noise.

De Certeau would have insisted on one further point, and it is the point that separates his analysis from the celebratory discourse that too often accompanies discussions of AI-assisted creation. The bricoleur's freedom is real, but it is conditioned. She works with materials she did not choose. The model's output reflects the biases, the patterns, the blind spots, and the aesthetic defaults of its training data — which is to say, it reflects the strategic decisions of those who assembled the training data, who chose what to include and what to exclude, who determined the relative weight of different sources, who established the alignment procedures that shape the output before it reaches the practitioner's screen. The bricoleur navigates this conditioned material with skill and creativity, but she cannot transcend it entirely. Her bricolage is always, in part, shaped by the materials available to her, and those materials are always, in part, shaped by strategic decisions she did not make and may not be able to see.

This is not a counsel of despair. It is a counsel of attentiveness — a recognition that the bricoleur's freedom, like the walker's freedom, operates within a space defined by others, and that the quality of the bricolage depends in part on the practitioner's awareness of the space's constraints. The builder who does not know that the model tends to generate a particular kind of structural default, who does not recognize the training data's biases in the model's suggestions, who accepts the system's rhetoric as neutral rather than as strategically shaped — that builder is a less skilled bricoleur than the one who sees the materials clearly, judges them honestly, and makes from them something that transcends the conditions of their production.

The bricoleur does not need a lumber yard. She needs a good eye. And a good eye, in the age of AI-assisted creation, is the scarcest and most valuable resource in the practitioner's kit.

Chapter 4: The Rhetoric of Practice

Every system speaks. It speaks in the language of its design — the structural assumptions, the default patterns, the implicit values that shape its output before any individual interaction begins. A city speaks through its grid: the width of the avenues communicates the priority of automobiles, the placement of the parks communicates the planners' theory of where recreation belongs, the height of the buildings communicates the economic logic that determined how much sky each block deserves. A text speaks through its conventions: the genre tells the reader what kind of attention is expected, the tone establishes the relationship between author and audience, the structure — chronological, thematic, argumentative — shapes the path the reader is invited to follow.

An AI model speaks through its patterns. The frequency with which certain phrases recur. The structural templates that organize the output — the tendency to present information in parallel constructions, to offer balanced perspectives, to hedge assertions with qualifications, to resolve tensions neatly rather than allowing them to persist. These patterns are not accidental. They are the product of the training process — the specific corpus of text the model ingested, the reward signals that shaped its fine-tuning, the alignment procedures that taught it which outputs are preferred and which are penalized. The model's rhetoric is strategic rhetoric: it speaks from a position of institutional control, producing output that reflects the aggregate preferences of the training process rather than the specific needs of any individual practitioner.

De Certeau distinguished between this strategic rhetoric — the speech of systems — and what he called the rhetoric of practice: the specific, contextual, personally inflected way in which individual practitioners create meaning through their use of systems. The cook does not merely follow the recipe. She adjusts, substitutes, improvises — and these adjustments constitute her rhetoric, her specific way of speaking through the practice of cooking. The walker does not merely follow the grid. She pauses, detours, lingers — and these deviations constitute her rhetoric, her specific way of speaking through the practice of walking. The rhetoric of practice is not a supplement to the system's rhetoric. It is a counter-rhetoric, a way of producing meaning that the system's own logic cannot generate, precisely because it arises not from the general patterns of the system but from the specific situation of the practitioner.

This distinction is analytically crucial for understanding what happens when a human being works with an AI model, because the interaction between strategic rhetoric and practical rhetoric is where the quality of AI-assisted creation is actually determined.

The model's strategic rhetoric has identifiable characteristics. Scholars of AI-generated text have begun to catalog them with the specificity that literary critics once brought to the analysis of authorial style. The tendency toward balanced qualification: "While X is certainly true, it is also important to consider Y." The preference for triadic structures: three examples where one strong one would suffice, three parallel clauses where asymmetry would be more revealing. The resolution of tension: the model's outputs tend toward synthesis, toward the reconciliation of opposing positions, toward the production of what might be called epistemic comfort — the reassurance that complexity has been managed, that the rough edges have been smoothed, that the reader need not sit with unresolved difficulty.

These characteristics are not flaws in the traditional sense. They are features of a system optimized to produce output that is, on average, helpful, accurate, and well-received across a broad population of users. The strategic rhetoric serves the strategic purpose: it produces output that works for most people most of the time. The tendency toward qualification prevents the model from making claims that might be contested. The preference for triadic structures provides the reader with multiple entry points, increasing the probability that at least one example will resonate. The resolution of tension produces the sense of completion that most readers find satisfying.

But the strategic rhetoric has a cost, and the cost is specificity. The output that works for most people most of the time is, by definition, optimized away from the particular — away from the specific voice, the idiosyncratic structure, the unresolved tension, the rough edge that would serve this practitioner's specific need at this specific moment. The strategic rhetoric speaks to everyone and therefore to no one. It produces text that is competent and generic, functional and forgettable, correct and lifeless.

The practitioner's rhetoric is the antidote to this generality. It is the specific quality that the practitioner brings to the encounter with the system's output — the decisions about what to keep and what to discard, what to amplify and what to mute, where to follow the system's structure and where to break from it. These decisions constitute the practitioner's voice, and they are, in de Certeau's framework, the locus of whatever we mean by creativity in the context of AI-assisted work.

Consider the specific operations through which the practitioner's rhetoric asserts itself against the system's rhetoric.

The first operation is selection: the decision about what to keep. The model generates a paragraph. The practitioner reads it and identifies the single sentence that contains genuine insight — the unexpected connection, the precise formulation, the phrase that captures something she had been reaching for but could not articulate. She keeps that sentence and discards the rest. The sentence is now hers, not because she originated it but because she recognized it — because her specific biography, her specific expertise, her specific sense of what matters created the conditions for that recognition. A different practitioner, reading the same paragraph, would have kept a different sentence or kept none at all. The selection is biographical. It is irreducible to the system's logic. It is the practitioner's rhetoric speaking through the act of choosing.

The second operation is refusal: the decision about what to discard. This operation is more demanding than selection, because it requires the practitioner to resist the seduction of the system's strategic rhetoric — the smoothness, the polish, the appearance of insight that may or may not survive examination. De Certeau's framework illuminates why refusal is so difficult in the context of AI-generated output: the strategic rhetoric is designed to be persuasive across a broad population, which means it is designed to feel right even when it is not right, to produce the sensation of understanding even when understanding has not occurred. The practitioner who can refuse this seduction — who can say "this sounds true but is not true" or "this sounds good but does not serve my purpose" — is performing a rhetorically sophisticated operation. She is reading against the grain of the system's rhetoric, asserting her own judgment against the system's statistical confidence. The Deleuze error that Segal caught in his collaboration with Claude is a case study in the rhetoric of refusal: the passage was elegant, well-structured, rhetorically effective, and philosophically wrong. Catching it required a practitioner whose own knowledge was specific enough to detect the failure that the system's surface quality concealed.

The third operation is inflection: the adjustment that transforms the system's generic output into something that bears the practitioner's specific mark. Inflection operates at every level of the text — from the word level (replacing the model's preferred term with the term that more precisely captures the practitioner's meaning) to the structural level (reorganizing the model's default structure into a structure that better serves the argument) to the tonal level (shifting the model's default register toward something rougher, or more intimate, or more uncertain, depending on what the specific work demands). Inflection is the operation through which the practitioner's voice emerges from the system's noise. It is the rhetorically equivalent of the walker's deviation from the grid — the moment where the tactical operation separates from the strategic structure and produces something that the structure alone could not have generated.

The fourth operation is what de Certeau called enunciation — the act of speaking that transforms language from a code into a practice. Language, de Certeau argued, is not merely a system of signs. It is a practice — something people do in specific situations, with specific intentions, producing specific effects that the system of signs cannot predict. The same sentence, spoken by different speakers in different contexts, means different things — not because the words have changed but because the enunciation has changed. The tone, the emphasis, the context, the relationship between speaker and listener, the history that the speakers share — all of these non-semantic factors shape the meaning of the utterance in ways that a purely linguistic analysis cannot capture.

AI-generated text is language without enunciation. It is the system of signs without the practice of speaking. The model produces syntactically and semantically competent text, but the text has not been said by anyone — it has been generated by a process that optimizes for pattern-matching across a corpus rather than for the specific communicative intention of a specific speaker addressing a specific audience in a specific situation. The practitioner who works with this text must supply the enunciation — must transform the model's generated language into spoken language, language that has been said by someone, language that carries the weight of a specific intention directed at a specific audience for a specific purpose.

This is the deepest level of the practitioner's rhetoric, and it is the level at which the difference between AI-assisted creation that merely functions and AI-assisted creation that genuinely communicates is determined. The practitioner who uses the model's output without supplying enunciation — who copies the generated text into the final product without transforming it from language-as-code into language-as-practice — produces work that is technically competent and experientially empty. It reads as generated because it was generated: no one said it, no one meant it, no one directed it at anyone in particular. The words are present but the voice is absent.

The practitioner who supplies enunciation — who takes the model's output and inflects it with specific intention, who decides not just what the text says but what she means by it, who directs the communication at a specific audience with a specific purpose that the model's general optimization cannot provide — that practitioner produces work that lives. Not because she originated the words but because she said them. She transformed the system's strategic rhetoric into a tactical utterance — specific, situated, directed, alive with the particular energy of a human being who has something to say and has chosen, from the vast territory of the model's output, the specific materials through which to say it.

De Certeau would have recognized in this operation the same creative dynamics he observed in every practice he studied. The cook who follows a recipe but adjusts it to her family's taste is supplying enunciation to the recipe's code. The walker who follows the grid but pauses at the window of a shop that has nothing to do with her destination is supplying enunciation to the city's structure. The reader who finds in the author's text a meaning the author did not intend is supplying enunciation to the text's signs.

In each case, the practice — the specific, situated, biographical operation of a human being navigating a system — transforms the system's strategic production into something lived. Something inhabited. Something that bears the mark of a particular practitioner working in particular conditions toward particular ends.

This transformation is invisible to the strategic gaze. The platform's analytics can measure what the user types and what the model returns. They cannot measure the specific quality of the practitioner's engagement with the output — the evaluative operations, the moments of recognition and refusal, the inflections that transform generic text into specific communication. The rhetoric of practice is invisible from the 110th floor for the same reason the walker's path is invisible from the 110th floor: it operates at a level of specificity that the strategic overview cannot resolve.

Yet it is precisely at this level of specificity that the value of AI-assisted creation is determined. The model provides the materials. The practitioner provides the rhetoric. And the rhetoric — the specific quality of selection, refusal, inflection, and enunciation through which the practitioner transforms the model's strategic output into a tactical utterance — is what separates work that matters from work that merely exists.

De Certeau spent his career attending to operations that were too small, too dispersed, too ordinary to attract the attention of theorists who looked for creativity in celebrated studios and canonical texts. The rhetoric of practice is ordinary. It is the everyday, habitual, unremarkable way that a practitioner engages with the systems that surround her. But its ordinariness is precisely its significance, because it demonstrates that creativity is not the exceptional achievement of exceptional individuals. It is the routine achievement of everyone who navigates a system with enough skill and enough attention to make something genuinely hers from materials that were produced by someone or something else.

In the age of AI, that routine achievement has become the central question of cultural production. The model generates. The practitioner speaks. And the quality of the speaking — the quality of the rhetoric that transforms the generated into the said — determines whether the amplification that AI provides carries signal or noise.

Chapter 5: Strategic Territory and Tactical Freedom

Every territory has a cartography that its inhabitants did not draw. The medieval peasant worked land whose boundaries had been established by a lord she had never met, according to a logic of inheritance and conquest that had nothing to do with the quality of the soil or the needs of the people who tilled it. The factory worker of the nineteenth century operated a machine whose design reflected the engineer's optimization of output per hour, not the worker's experience of standing in one position for twelve hours while the machine dictated the rhythm of her body. The student in a modern university navigates a curriculum whose structure reflects the institutional politics of departmental funding, the historical accidents of disciplinary formation, and the strategic calculations of administrators who must balance enrollment numbers against faculty salaries — none of which have any necessary connection to what the student needs to learn or how she learns best.

De Certeau understood that the defining condition of most human life is not the possession of territory but the habitation of territory that belongs to someone else. The lieu propre — the proper place, the place of one's own — is the strategist's privilege. The rest of us live in spaces we did not design, governed by rules we did not set, shaped by decisions we were not consulted about. This is not a description of oppression in the dramatic sense — though oppression is certainly one of its forms. It is a description of the ordinary condition of human existence, the condition under which tactical practice operates and from which tactical creativity emerges.

The AI model's output space is strategic territory of a kind that de Certeau could not have anticipated in its specifics but would have recognized immediately in its structure.

The territory is vast. A large language model trained on hundreds of billions of tokens of text encompasses a space of possible outputs that is, for practical purposes, unbounded. Any sequence of words that is statistically consistent with the patterns of the training data is, in principle, generable. The territory contains poetry and code, legal briefs and love letters, philosophical arguments and shopping lists, technical documentation and fairy tales. It contains, in potential, more text than any individual could read in a thousand lifetimes.

The territory is structured. Despite its vastness, the output space is not uniform. Certain regions are more densely populated than others — the model generates certain kinds of text more readily, more fluently, with greater structural coherence, because those kinds of text are more heavily represented in the training data. The territory has peaks and valleys, well-worn paths and barely accessible corners, default routes and hidden passages. English-language academic prose occupies a broad, well-maintained highway. Oral storytelling traditions from sub-Saharan Africa occupy a narrow, poorly lit trail. The territory's topography is not neutral. It reflects, with the fidelity of a geological record, the specific history of what was written down, what was digitized, what was deemed worthy of inclusion in the corpus that became the model's world.

The territory is governed. Alignment procedures — reinforcement learning from human feedback, constitutional AI constraints, system prompts, content policies — function as the zoning regulations of the output space. They determine what the model may and may not produce, which regions of the territory are accessible to users and which are fenced off, what kinds of output are encouraged and what kinds are suppressed. These governance structures are not visible to most users in the way that a city's zoning map is not visible to most pedestrians. The walker does not see the zoning regulation that determined the height of the building she passes. The user does not see the alignment constraint that determined the tone of the response she receives. But both structures shape the territory through which the practitioner moves, and both constitute the strategic framework within which tactical practice must operate.

The territory is commercial. The AI model is not a public utility. It is a product, offered by a corporation, under terms of service that constitute a contractual framework governing the user's relationship with the territory. The corporation's strategic interests — growth, retention, revenue, regulatory compliance, reputational management — shape the territory in ways that may or may not align with any individual practitioner's needs. The default behaviors of the model — its tendency toward helpfulness, its avoidance of controversy, its resolution of ambiguity in favor of the response most likely to satisfy the broadest range of users — are not aesthetic choices. They are business decisions, strategic calculations about what kind of territory will attract and retain the largest number of users at the lowest cost of moderation.

De Certeau's analysis of strategic territory was never a denunciation. He did not argue that strategies are evil or that the strategist's control of territory is illegitimate. He argued that the strategic perspective is partial — that it sees the territory from above without seeing the practices of those who inhabit the territory from below. The city planner's grid is a real and useful structure. The question is not whether the grid should exist but whether the grid is the whole story. And the answer, for de Certeau, was always no — the grid is the frame within which the story occurs, but the story itself is written by the walkers, the practitioners, the tacticians who navigate the grid in ways the planner did not design and cannot fully see.

The same holds for the AI model's output space. The architecture is real. The training data is real. The alignment constraints are real. The commercial imperatives are real. These strategic structures constitute the territory within which the builder operates. But the territory is not the practice. The practice is what the builder does within the territory — the specific, biographical, unrepeatable way she navigates the output space, finding paths the designers did not foresee, discovering capabilities the documentation does not describe, appropriating the territory's resources for purposes that the territory's governors did not intend and may not endorse.

The tactical freedom available within strategic territory is genuine but conditioned. De Certeau never claimed that the walker is free in the absolute sense. The walker cannot redesign the grid. She cannot move the buildings. She cannot change the direction of the avenues. Her freedom operates within the constraints the strategy imposes. But within those constraints, her freedom is real — real enough to produce a city-within-the-city, a lived geography that the planner's map cannot capture, a practice that generates meaning the strategic structure alone could not produce.

The forms of tactical freedom available within the AI model's output space are specific and identifiable.

There is the freedom of the prompt — the practitioner's capacity to shape the territory's response by shaping the question. Prompt engineering, in de Certeau's framework, is a tactical art: the practitioner learns the territory's topography, discovers which paths lead to productive regions and which lead to dead ends, develops an intuitive sense of how to formulate requests that will elicit output closest to her actual need. This knowledge is practical, not theoretical. It is built through repeated engagement with the territory — through the same kind of embodied, habitual, incrementally refined knowledge that the walker builds through years of traversing the same streets.

There is the freedom of selection — the practitioner's capacity to choose from the territory's output what serves her purpose and refuse what does not. This freedom is the freedom of the poacher, discussed in the previous chapter, and it is perhaps the most fundamental form of tactical agency available to the AI user. The model generates. The practitioner judges. The judgment is hers, shaped by her biography, her expertise, her aesthetic sensibility, her specific understanding of the context in which the output will be used.

There is the freedom of combination — the practitioner's capacity to assemble outputs from multiple interactions, multiple prompts, multiple models, into a composite work that none of the individual outputs anticipated. This is bricolage at the level of the production process itself: the practitioner does not rely on a single response but orchestrates a series of interactions, each of which produces materials that she then combines, juxtaposes, and rearranges according to a design that exists only in her own understanding of what the finished work requires.

And there is the freedom of misuse — the practitioner's capacity to use the territory for purposes it was not designed to serve. De Certeau documented the creativity of misuse extensively in his studies of everyday practice: the worker who uses company time for personal projects, the consumer who repurposes a product for an unintended function, the reader who finds in a text a meaning the author would not recognize. The AI user who discovers that a coding assistant can help her write poetry, or that a writing assistant can help her debug logical arguments, or that a general-purpose model can serve as a thinking partner for problems that the model's designers never contemplated — these practitioners are exercising the freedom of misuse, and their creativity is of precisely the kind that de Certeau spent his career celebrating.

But the conditioned nature of this freedom must not be obscured by the celebration. De Certeau was always attentive to the asymmetry between strategic power and tactical agency. The strategist controls the territory. The tactician navigates it. The strategist can redesign the grid at will — can update the model, change the training data, modify the alignment constraints, alter the terms of service, raise the price. The tactician cannot. The tactician's freedom is the freedom of the tenant, not the owner. It is real, but it is contingent on the strategist's continued willingness to provide the territory in its current form.

This contingency introduces a specific vulnerability into AI-assisted practice that de Certeau's framework makes visible. The walker who has built her daily routine around a shortcut through an alley discovers one morning that the alley has been gated. The reader who has built an interpretive practice around a text discovers that the text has been revised in a new edition that eliminates the ambiguity she found most generative. The builder who has developed a workflow around a specific model's capabilities discovers that an update has changed the model's behavior in ways that break her practice.

These disruptions are not bugs. They are features of the strategic-tactical relationship. The strategist controls the territory and may modify it at any time for strategic reasons — reasons that have nothing to do with the tactician's practice and may actively undermine it. The model update that improves average performance across a population may destroy the specific capability that one practitioner relied on. The alignment change that reduces risk at the aggregate level may eliminate the productive ambiguity that another practitioner found most creatively valuable.

De Certeau's framework does not resolve this vulnerability. It diagnoses it with precision: the tactician's freedom is real but always at the mercy of the strategist's decisions. The practitioner who has built her creative life within the territory of an AI platform is building on ground she does not own. The ground may shift. The alley may be gated. The shortcut may disappear.

The response to this vulnerability is not retreat — not the refusal to build on territory one does not own, because that refusal would mean the refusal to build at all, since the condition of habitation within strategic territory is the ordinary condition of human life. The response is the cultivation of the specific skills that make tactical practice resilient: the capacity to adapt when the territory changes, the knowledge that comes from understanding the territory's logic well enough to anticipate its shifts, the maintenance of multiple routes through the landscape so that the gating of one alley does not collapse the entire practice.

The walker who knows only one route is vulnerable. The walker who has, over years of practice, built a repertoire of routes — a rich, varied, deeply understood personal geography of the city — can lose any single path and still find her way. The same principle applies to the builder. The practitioner who has built her creative practice around a single model, a single workflow, a single set of capabilities, is vulnerable to any change in the territory. The practitioner who has developed a broader relationship with the landscape of AI tools — who understands the strategic logic well enough to adapt when the territory shifts, who maintains the core skills of judgment, selection, and rhetorical inflection that transfer across platforms — that practitioner has built what de Certeau would recognize as a resilient tactical practice. A practice grounded not in the stability of the territory but in the depth of the practitioner's engagement with it.

The territory belongs to the strategist. The practice belongs to the practitioner. And the practice, when it is skilled enough, survives the territory's changes — because it was never the territory that made the practice valuable. It was the practitioner.

Chapter 6: Everyday Creativity and the Rising Floor

De Certeau spent his intellectual life in the company of people who had never been invited into the conversation about creativity. Not the painter in the studio. Not the composer at the piano. Not the novelist in the garret. The cook in the kitchen. The commuter on the bus. The tenant navigating the bureaucracy of a housing authority that was designed to process her rather than to serve her. These were the practitioners whose creativity de Certeau sought to recover from the theoretical invisibility to which the Western intellectual tradition had consigned it.

The invisibility was not accidental. It was structural. The Western tradition of thinking about creativity had, since at least the Romantic period, operated on an implicit distinction between creation and use — between the originary act of the artist who produces something new and the secondary act of the consumer who merely receives what has been produced. This distinction organized an entire economy of cultural prestige: the creator was celebrated, studied, anthologized, granted the dignity of authorship. The user was invisible, interchangeable, analytically uninteresting — a node in the distribution chain rather than a participant in the creative process.

De Certeau's revolution was to insist that use is itself a form of creation. The reader who makes meaning from the author's text. The cook who transforms raw ingredients into a meal. The pedestrian who writes a personal narrative through the planned city by the path she chooses. Each of these practitioners is performing an act of creation that is no less real for being ordinary, no less creative for being everyday, no less significant for being invisible to the frameworks that recognize only originary production as creative.

This recovery of everyday creativity provides the theoretical foundation for understanding the most morally significant feature of the AI moment: the expansion of who gets to build.

Before AI-assisted tools reached their current capability, the act of building software, designing interfaces, creating analytical models, producing professional-quality text — each of these required specialized training that functioned as a gatekeeping mechanism. The training was not arbitrary. It served real purposes: it built skill, it developed judgment, it deposited the layers of understanding that come only through sustained engagement with difficulty. But it also functioned, whether intentionally or not, as a barrier that restricted the building process to those who had the institutional access, the economic resources, and the years of time necessary to complete the training.

The result was a specific distribution of creative agency. Those who had passed through the gates — the credentialed, the trained, the institutionally supported — could build. Those who had not — the teacher with an idea for an educational tool but no programming knowledge, the small business owner with a vision for a customer service system but no budget for a development team, the immigrant with a need for a bureaucratic navigation tool but no access to the technical infrastructure required to create one — could not build. Their ideas remained ideas. Their creativity, in the specific sense of the capacity to realize an intention in the world, was constrained not by lack of imagination but by lack of access to the means of production.

De Certeau would have recognized this distribution immediately. The credentialed builders occupied a lieu propre — a proper place from which to produce. The uncredentialed practitioners operated tactically, making do with whatever tools were available, finding workarounds, improvising solutions that were never quite adequate because the tools were never designed for their specific needs. The gap between the strategic producer and the tactical user was a gap in creative agency — not in the capacity for creativity but in the capacity to realize creativity in the world.

AI tools have narrowed this gap. Not eliminated it. The inequalities of access, connectivity, infrastructure, and capital that The Orange Pill honestly acknowledges remain real. But the floor has risen. The minimum level of technical capability required to build a working product has dropped so dramatically that people who were previously excluded from the building process can now participate.

The teacher who uses an AI tool to build customized reading materials for a dyslexic student is not becoming a software developer. She is practicing everyday creativity in a new medium — using AI tools the way she uses kitchen tools: tactically, improvisationally, with whatever is at hand. Her engagement with the tool follows exactly the pattern de Certeau described in his analysis of cooking. She does not have ideal materials. The AI model's output is not perfectly suited to her specific student's needs. But she has judgment — years of experience working with students who learn differently, an embodied knowledge of what helps and what does not, an intuitive sense of the specific difficulty this particular child faces with this particular text. She takes the model's output and adjusts it, substitutes, improvises — performs the bricolage that de Certeau recognized as the creative operation par excellence of those who work with what they are given rather than what they would choose.

The small business owner who uses an AI design tool to create marketing materials she could never have afforded from a professional designer is not performing graphic design. She is making do. She is taking the tool's output — generic templates, standard layouts, pattern-based aesthetic choices — and inflecting it with her specific knowledge of her customers, her neighborhood, her product. The result is not what a professional designer would have produced. It may lack the polish, the sophistication, the refined aesthetic judgment that years of design training develop. But it exists. It serves its purpose. It communicates what needs to be communicated to the people who need to receive it. Before the tool, this practitioner had no voice in the visual landscape of commerce. Now she has one — imperfect, tactical, assembled from materials she did not design, but hers.

The developer in Lagos — the figure who appears repeatedly in The Orange Pill as the test case for democratization — is, in de Certeau's framework, the paradigmatic everyday creative practitioner. She possesses the intelligence, the ambition, the ideas. What she lacked was the institutional infrastructure — the team, the capital, the network of mentors and investors — that the strategic production of software required. AI tools did not give her those things. They gave her something different: the capacity to produce, through bricolage, through tactical appropriation of the model's output, through the skilled improvisation of someone working with whatever happens to be available, a working product that serves real users. The product is not the product that a well-funded San Francisco startup would have built. It is the product that she could build, given her specific constraints, using her specific materials, directed by her specific judgment about what her specific users need.

This is everyday creativity. Not the celebrated creativity of the genius in the studio. The quiet, dispersed, ordinary creativity of a human being making do — making meaning, making a livelihood, making a life — from the materials the world provides. De Certeau spent his career arguing that this everyday creativity is real, that it is worthy of analytical attention, that it constitutes a form of cultural production as genuine and as significant as any celebrated original work. The AI moment has made his argument empirically visible, because it has produced conditions in which everyday creative practitioners — teachers, business owners, immigrants, students, anyone with an idea and the judgment to direct a tool — can participate in domains of building that were previously gated by specialized training.

The expansion of everyday creativity raises questions that de Certeau's framework addresses with characteristic nuance. When the floor rises — when more people can build, when the cost of production drops, when the barriers to participation erode — does the quality of what is produced decline?

The question is real. It has been asked at every moment of creative democratization in human history. When the printing press made publishing cheap, the quality of the average published text declined — more was printed, and much of what was printed was mediocre or worse. When digital cameras made photography free, the quality of the average photograph declined — more images were produced, and most of them were unremarkable. When social media made publishing instantaneous, the quality of the average published text declined again — more was written, and the signal-to-noise ratio collapsed.

In each case, the answer was not less production but better judgment. The printing press created the need for literary criticism, for editorial curation, for the institutional mechanisms through which quality is distinguished from quantity. The digital camera created the need for curatorial platforms, for photographic criticism, for the development of visual literacy among a population that was suddenly producing images at a rate that outstripped the culture's capacity to evaluate them.

De Certeau would have framed the question differently. For de Certeau, the question of quality cannot be separated from the question of practice. The cook who produces a mediocre meal from the refrigerator's contents has not failed at cooking. She has performed a tactical operation whose quality is determined by the specificity of her engagement with the materials available to her — her attentiveness to what is there, her skill in combining, her judgment about what will serve. Some meals are better than others. But the difference is not between creation and consumption, between the professionally trained and the amateur. The difference is between more skilled practice and less skilled practice, between deeper engagement and shallower engagement, between the practitioner who brings her full attention to the materials at hand and the practitioner who does not.

The same holds for AI-assisted building. When more people can build, more of what is built will be mediocre. But the mediocrity is not a consequence of democratization. It is a consequence of the natural distribution of skill and attention across any population of practitioners. Some practitioners will engage deeply with the tools, will develop the evaluative judgment, the rhetorical sensitivity, the contextual knowledge required to produce work of genuine quality. Others will engage superficially, will accept the model's default output without the critical evaluation that transforms generated text into spoken text, will produce work that is functional but generic — the AI-assisted equivalent of the meal that fills the stomach without engaging the palate.

De Certeau would have been interested in both practitioners — the skilled and the unskilled, the deep and the shallow — because both are performing tactical operations within strategic territory, and both illuminate the dynamics of everyday practice. But he would have insisted that the expansion of who gets to practice is, in itself, a moral good. The cook who produces a mediocre meal is still feeding her family. The teacher who produces an imperfect educational tool is still serving her student. The developer in Lagos who produces a workmanlike product is still building something that did not exist before and that serves a need that was previously unmet.

The floor has risen. The ceiling has not fallen. More people can build, and the best builders — the practitioners whose judgment, whose rhetorical skill, whose deep engagement with the materials available to them sets their work apart — can build better than ever, because they too benefit from the expansion of available materials. The cook with the most refined palate and the deepest knowledge of flavors benefits as much from a full pantry as the novice. More, in fact, because she knows what to do with the abundance.

The moral significance of the rising floor is not that it produces better work. It is that it produces more practitioners — more people engaged in the creative practice of building, more people exercising the tactical agency that de Certeau recognized as the ordinary and undervalued form of human creativity. Whether the output of that expanded practice meets the standards of the professional is a question worth asking. But it is not the first question. The first question is whether the expansion itself is good. And de Certeau's entire intellectual project was the argument that it is — that the recovery of everyday creativity, the recognition of the practitioner's art, the expansion of who gets to make and build and speak, is a moral achievement regardless of the average quality of what is produced.

The rising floor is not the rising ceiling. But it is the ground on which more people can stand, and from which more people can begin the practice of building — the tactical, improvisational, everyday creative practice that de Certeau spent his life insisting was real, was significant, and was worthy of the same attention that the culture lavishes on the celebrated productions of the genius in the studio.

Chapter 7: The Smooth and the Gap

The perfectly planned city has no alleys. Its grid is complete, its surfaces uniform, its spatial logic so coherent that no deviation is possible and none is necessary. Every path leads where the planner intended. Every intersection has been optimized for flow. The pedestrian moves through the city the way a ball moves through a pinball machine — guided, channeled, delivered to the destination the design prescribes.

In such a city, the walker has no freedom. Not because she is imprisoned — no walls confine her, no guards restrict her movement — but because there is nowhere to go that the planner did not anticipate. Every shortcut has been preempted. Every detour has been designed out of the system. The gap between what the city offers and what the walker might make of it has been closed. The territory and the practice have been made identical, which means the practice has been abolished, because practice, in de Certeau's sense, exists only in the gap between the system's design and the practitioner's use.

This image — the city without alleys, the system without gaps — is the theoretical endpoint of what Byung-Chul Han calls the aesthetics of the smooth. Han argues that the dominant aesthetic of the contemporary moment is an aesthetic of frictionlessness: the elimination of resistance, the removal of every surface that might slow the user down, the production of experiences so seamless that the user never encounters anything she did not already desire. The smooth interface. The frictionless checkout. The algorithm that learns her preferences and serves her more of what she already consumes. Each of these is an optimization that removes a gap — a gap between desire and satisfaction, between intention and execution, between the user's will and the system's response.

De Certeau's framework converges with Han's diagnosis at a specific and illuminating point: the gap that smoothness eliminates is the gap in which tactical creativity operates. The walker's freedom depends on the city's imperfection — on the alley the planner overlooked, the corner where the grid breaks, the space between buildings where an unofficial path has been worn into the grass by the feet of pedestrians who found a route the design did not provide. The reader's freedom depends on the text's ambiguity — on the passage whose meaning is not fixed, the metaphor that can be read in multiple ways, the structural tension that the author left unresolved, perhaps intentionally, perhaps not. The cook's freedom depends on the recipe's incompleteness — on the instruction that says "season to taste," which transfers the creative authority from the recipe to the practitioner and acknowledges that the system cannot determine the outcome without the practitioner's judgment.

In each case, the gap is the space of tactical practice. And smoothness — the elimination of friction, the resolution of ambiguity, the optimization of every surface for maximum efficiency — is the closure of that space.

The AI model's output, at its most characteristic, tends toward smoothness. This tendency is structural, not incidental. The model has been trained on a vast corpus of text and optimized through reinforcement learning to produce responses that are helpful, coherent, and well-received across a broad population of users. This optimization selects for output that is smooth in precisely Han's sense — output that resolves tensions rather than sustaining them, that answers questions rather than deepening them, that offers the user what she appears to want rather than what she might need if the system were willing to be less accommodating. The balanced qualification. The triadic structure. The neat synthesis. These are the signatures of a system whose strategic logic drives it toward the elimination of productive roughness.

De Certeau's framework predicts that this smoothness will be experienced as a loss of tactical space — that the practitioner who works with a model whose output is uniformly polished will find fewer gaps in which her own creative operations can take hold. If the model's prose is always competent, the practitioner never encounters the rough phrase that, through its roughness, invites her to intervene, to rewrite, to supply from her own resources the specific quality that the text lacks. If the model's arguments are always balanced, the practitioner never encounters the productive imbalance — the strong claim, the unhedged assertion, the risky move — that would provoke her into response, into disagreement, into the kind of creative friction that produces genuine thinking rather than the mere consumption of pre-digested thought.

The builder who accepts smooth output without resistance is living in the city without alleys. She moves through the model's territory along paths the system designed, arrives at destinations the system anticipated, produces work that is competent and generic and indistinguishable from the work of every other practitioner who traveled the same paths through the same territory. Her practice has been abolished — not by force, but by comfort. The system has made compliance so easy that deviation requires an effort of will that the system's very smoothness discourages.

But de Certeau's framework also contains a counter-argument that Han's critique, for all its diagnostic brilliance, tends to underweight.

No system is perfectly smooth. The city without alleys does not exist. Even the most carefully planned urban environment develops gaps — spaces where the plan breaks down, where the grid encounters a river or a hill or a property boundary that predates the grid, where the planners' logic gives way to the contingencies of the physical world. And in those gaps, the walkers find their freedom.

The AI model, for all its optimization toward smoothness, is not a perfectly smooth system. It hallucinates — produces output that is confidently wrong, that asserts connections that do not exist, that cites sources that were never written. It exhibits what might be called productive incompetence — moments where the model's failure to produce the expected output creates a space that the practitioner can inhabit creatively. It generates unexpected combinations — connections between ideas that the training data's patterns make statistically possible but that no human reader would have anticipated, connections that are sometimes brilliant and sometimes absurd and always, in either case, a gap in the smooth surface that the practitioner can exploit.

The Deleuze error that recurs throughout these discussions is a paradigmatic example. Claude produced a passage that connected two concepts with confident elegance, and the connection was wrong. The wrongness was a gap in the smooth surface. And the practitioner who caught the wrongness — who recognized that the philosophical reference did not support the argument the model had constructed around it — had two options. She could simply correct the error, which would be the smooth response: patching the gap, restoring the surface, maintaining the system's appearance of competence. Or she could inhabit the gap — could ask what the model was reaching for when it made the connection, could explore whether the connection itself (as distinct from the false reference supporting it) contained something true, could use the error as a starting point for her own thinking rather than as a defect to be repaired.

The second response is tactical. It treats the model's failure as an opportunity rather than a problem. It finds in the gap between what the system intended and what the system produced the specific space in which the practitioner's own creativity can operate. This is reading-as-poaching applied to the model's errors: taking the wrong output and making something right from it, not by fixing the error but by using the error as material for a creative operation the system did not anticipate.

De Certeau would have recognized in this practice a version of what he observed in every domain he studied. The cook who discovers that a dish has been oversalted and responds not by discarding it but by adjusting — adding acid, adding fat, transforming the mistake into a different dish that could not have been conceived without the mistake. The walker who discovers that her usual route is blocked and, in the detour, finds a street she has never walked, a view she has never seen, a shop whose existence she would never have discovered if the planned path had remained available. The reader who encounters a passage that seems wrong — that contradicts what she understood the text to be arguing — and who, in wrestling with the contradiction, discovers a deeper reading that the smooth surface of agreement would never have provoked.

The gap is the space of thinking. The smooth surface is the space of consumption. And the practitioner's task, in the age of AI, is to find the gaps — in the model's output, in the model's failures, in the model's unexpected combinations — and to inhabit them with enough skill and enough attention to produce something that the smooth surface alone could never yield.

This framing resolves, or at least reframes, the tension between Han's critique and the counter-argument mounted in The Orange Pill. Han is right that smoothness destroys a specific kind of creative space — the space of friction, of resistance, of the productive difficulty that forces the practitioner to engage deeply with the materials rather than gliding across their surface. But Han's critique assumes that the practitioner is passive before smoothness — that the elimination of friction necessarily eliminates the practitioner's creative engagement. De Certeau's framework challenges this assumption. The practitioner is never entirely passive. Even within the smoothest system, she finds gaps — or creates them, through the quality of her attention, through her refusal to accept the surface as the whole, through her willingness to read against the grain of the system's strategic rhetoric.

The smooth interface is the enemy of the inattentive practitioner. It is not the enemy of the attentive one. The attentive practitioner brings to the smooth surface a roughness of her own — a critical sensibility, an evaluative judgment, a refusal to be satisfied by competence when what she needs is excellence. This roughness is not a property of the system. It is a property of the practice. And it is, in de Certeau's framework, the practitioner's most valuable resource — more valuable than the model's capabilities, more valuable than the interface's ease of use, more valuable than the vast abundance of materials the system provides.

The gap is not given. It is found. And the finding is a creative act — perhaps the most important creative act available to the practitioner in an age of systems optimized for the elimination of every difficulty, every ambiguity, every productive imperfection that might slow the user down.

De Certeau studied practitioners who found gaps in systems far more rigid than any AI model. The medieval peasant who found, in the lord's law, the loophole that allowed her to pasture her goats on the common. The factory worker who found, in the rhythm of the assembly line, the moments when the foreman's attention lapsed and a few minutes of personal time could be stolen. The reader who found, in the most didactic text, the passage that escaped the author's ideological control and said something the author did not mean to say.

These practitioners did not wait for the system to provide them with gaps. They brought their own attention to the system's surface and found what was there to be found. The AI-assisted practitioner must do the same. The model will tend toward smoothness. The practitioner must tend toward roughness. The practitioner must bring to the encounter the critical sensibility, the evaluative judgment, the refusal to be satisfied, that creates the conditions for tactical creativity to operate even within a system designed to make tactical creativity unnecessary.

The gap is where the practitioner lives. The smooth surface is where the system wants her to glide. And the quality of the practice — the quality of the thinking, the building, the making — depends on which tendency wins in the encounter between the system's smoothness and the practitioner's roughness.

Chapter 8: Resistance Through Use

The Luddites broke machines. The gesture was dramatic, visible, legible to power. It said: we refuse. It said: the machine is the enemy. It said: the destruction of the instrument of our displacement is the appropriate response to our displacement. The gesture was strategic resistance — resistance that attacks the system on the system's terms, that seeks to alter the strategic territory through direct confrontation, that imagines that the destruction of the tool will reverse the transformation the tool has wrought.

De Certeau studied a different kind of resistance. Quieter. More dispersed. Nearly invisible to the institutional gaze that monitors populations from above. Resistance through use — the specific, daily, unglamorous ways in which people appropriate the systems they inhabit and redirect them toward purposes the systems' designers did not intend. Not the destruction of the machine but the creative misuse of the machine. Not the refusal to participate but the transformation of participation into something the system cannot fully control.

De Certeau gave this form of resistance a name that captures its specific quality. He called it la perruque — the wig. The term comes from the French workplace practice of using the employer's tools, materials, and time to do one's own work: the factory worker who uses company machinery to build a piece of furniture for his home, the office worker who uses company stationery and the company printer for personal correspondence, the secretary who uses the company computer to write her novel during the slow hours of the afternoon. La perruque is not theft, exactly. The worker does not take the employer's products. She takes the employer's means of production and redirects them, temporarily and invisibly, toward her own ends.

The practice is tactical in de Certeau's precise sense. It operates within the strategic territory — the workplace, governed by the employer's rules, structured by the employer's schedule, equipped with the employer's tools. The worker does not challenge the employer's control of the territory. She does not picket, strike, or sabotage. She simply uses the territory differently from how the employer intended, in the gaps between supervision, in the moments when the strategic gaze looks elsewhere, in the spaces where the system's control is less than total.

La perruque is resistance because it redirects the system's resources from the system's purposes to the practitioner's purposes. It is invisible because it does not alter the system's surface — the factory still runs, the office still functions, the work still gets done. The practitioner's tactical appropriation occurs within the system's normal operation, as a parasite within the host, drawing sustenance without destroying the organism.

This form of resistance maps onto the AI landscape with a precision that de Certeau could not have anticipated but would have immediately recognized.

The AI platform is the employer's territory. The model was built by a corporation, deployed as a service, governed by terms of service, shaped by commercial imperatives. The user enters this territory as the worker enters the workplace — under conditions defined by someone else, with tools provided by someone else, within a structure designed to serve someone else's strategic purposes. The platform wants engagement. The platform wants data. The platform wants the user to use the model in ways that generate revenue, that improve the training data, that demonstrate the product's value to investors and regulators and the broader market.

The user who operates within these terms — who uses the model as intended, produces output that stays within the system's designed parameters, generates the engagement and the data that the platform seeks — is not resisting. She is complying. Her use of the platform serves the platform's strategic purposes, even if it also serves her own.

But the user who discovers unintended capabilities, who repurposes the model for tasks its designers did not contemplate, who finds in the system's architecture possibilities that the system's governors did not foresee and may not endorse — that user is practicing la perruque. She is using the employer's tools for her own work. She is redirecting the system's resources from the system's purposes to her own purposes, within the system's territory, under the system's rules, without challenging the system's control but also without fully submitting to it.

The history of creative AI use is, in significant part, a history of la perruque. The early users of large language models who discovered that a system designed for text completion could be prompted to produce poetry, to engage in philosophical dialogue, to generate code, to serve as a thinking partner for problems the system was never designed to address — these users were practicing tactical resistance through creative misuse. They did not break the machine. They did not refuse to use it. They used it — but they used it for purposes that exceeded and sometimes contradicted the purposes its designers intended.

Prompt engineering itself, in its most creative forms, is an art of la perruque. The practitioner who develops elaborate prompting strategies — who discovers that a specific sequence of instructions, a specific framing of the request, a specific fictional scenario within which the prompt is embedded, unlocks capabilities the model's standard behavior does not exhibit — is using the employer's tools for her own work. She is navigating the strategic territory with a tactician's cunning, finding paths that the system's designers did not build and may not even know exist.

The jailbreak — the prompt designed to circumvent the model's alignment constraints, to access regions of the output space that the system's governors have fenced off — is the most dramatic form of la perruque in the AI context. De Certeau would have been fascinated by it, not because he would have endorsed the bypassing of safety measures, but because the jailbreak demonstrates with exceptional clarity the dynamics of tactical practice within strategic territory. The alignment constraint is the employer's rule. The jailbreak is the worker's deviation. The constraint attempts to close a gap in the system's control. The jailbreak finds or creates a new gap. The dynamic is perpetual: each strategic closure provokes a new tactical opening, and each tactical opening provokes a new strategic closure.

But de Certeau's analysis of resistance through use extends far beyond the dramatic case of the jailbreak. The most significant forms of tactical resistance are not dramatic. They are ordinary. They are the daily, habitual, unremarkable ways in which practitioners use AI tools to do things the tools were not specifically designed for — and, in doing so, demonstrate that the system's strategic logic cannot fully account for the creativity of the people who inhabit it.

The teacher who uses a general-purpose language model to create specialized educational materials, adapted to the specific needs of a specific student, is practicing resistance through use. The model was not designed for her specific student. The training data did not anticipate the particular learning difficulty that this child faces. The alignment procedures were not calibrated for the pedagogical context in which the teacher operates. Yet she takes the model's generic output and transforms it, through the tactical operations of selection, adaptation, and inflection, into something that serves a purpose the system's designers could not have foreseen — because the purpose is as specific as the child, as local as the classroom, as biographical as the teacher's twenty years of experience working with children who learn differently.

This is resistance in de Certeau's precise sense: not the refusal to use the system but the transformation of the system's resources into something the system did not produce and could not have produced on its own. The resistance is invisible to the platform's analytics. The model served a prompt and generated a response. The engagement metric was recorded. The data was collected. The strategic machinery operated as designed. But the tactical operation that occurred within the machinery — the teacher's creative appropriation of the output for a purpose that exceeded the system's design — is invisible to the strategic gaze, because the strategic gaze can see what the user did but not what the user made of it.

De Certeau's analysis of the Luddites, refracted through this framework, produces a reading that diverges sharply from both the standard narrative and the reading in The Orange Pill. The standard narrative says the Luddites were wrong because they opposed progress. The Orange Pill says the Luddites were right about the loss but wrong about the response — that breaking machines was emotionally satisfying and strategically catastrophic. De Certeau's framework suggests a third reading: the Luddites failed because they chose strategic resistance when tactical resistance would have been more effective.

The framework knitters who survived the industrial transition were not the ones who broke machines. They were the ones who found ways to use the new industrial system — its materials, its distribution networks, its markets — for purposes that the system's designers had not fully anticipated. The weavers who applied their knowledge of fiber and drape and quality to the products of the power loom — who became inspectors, designers, quality evaluators, people whose expertise was not in operating the old technology but in judging the output of the new one — were practicing la perruque at the civilizational level. They were using the employer's tools for their own work. They were redirecting the industrial system's resources toward purposes the system's strategic logic had not designed for.

The parallel to the present moment is direct. The developers, the writers, the designers, the analysts whose skills are being transformed by AI are not faced with a binary choice between breaking the machine and submitting to it. They are faced with the tactician's choice: how to use the machine — its capabilities, its resources, its outputs — for purposes that the machine was not specifically designed to serve. How to bring their own judgment, their own expertise, their own biographical specificity to the encounter with the system's strategic output, and produce from that encounter something that the system alone could not have produced.

This is not a heroic narrative. De Certeau did not write heroic narratives. He wrote narratives of the ordinary — of the daily, quiet, dispersed practices through which people who lack institutional power find ways to live with dignity and creativity within systems designed by others. The resistance he described was not the resistance of the revolutionary who storms the barricade. It was the resistance of the worker who uses the lunch break to make a phone call, the reader who misreads the text in a way that liberates its meaning, the walker who takes the unauthorized shortcut through the park.

These are small acts. They do not transform the system. They do not redistribute power. They do not produce the dramatic rupture that the revolutionary imagines. What they produce is something de Certeau valued more: the demonstration that the system's control is never total. That the strategic order, for all its power, cannot fully determine what the people within it will do with the resources it provides. That creativity persists — not in spite of the system but within it, in the gaps, the margins, the overlooked spaces where the strategic gaze does not quite reach.

The AI system is more powerful than any system de Certeau studied. Its strategic logic is more comprehensive, more adaptive, more capable of learning from the very tactical deviations that challenge it. The recommendation algorithm that learns from the user's creative misuse and incorporates that misuse into its own strategic calculations represents a level of strategic responsiveness that de Certeau could not have imagined.

And yet the practitioners persist. They find new gaps. They develop new tactics. They bring to the encounter with the system a creativity that the system's strategic logic cannot fully anticipate, because that creativity arises not from the system's patterns but from the irreducible specificity of a human being who has a particular history, a particular need, a particular sense of what matters — and who refuses, quietly, daily, without drama, to let the system determine the entirety of what she makes from its resources.

De Certeau would not have called this optimism. He would have called it analysis — the precise description of what people actually do when they inhabit systems they did not design. They use them. They appropriate them. They redirect them. They resist, not through refusal but through the specific quality of their engagement.

And that engagement — tactical, quotidian, invisible from above, visible only to those who look up from the street rather than down from the 110th floor — is the form of agency that persists even when every other form has been foreclosed.

Chapter 9: Space, Place, and the Personal Practice

A house is not a home. The distinction is so commonplace it has become invisible, which is precisely why de Certeau would have found it analytically fertile. A house is a structure — walls, roof, floor, the spatial organization imposed by an architect who designed the building according to principles of efficiency, aesthetics, zoning regulation, and material constraint. A house is strategic territory. It defines the space within which the inhabitant must live: the kitchen is here, the bedroom is there, the bathroom is connected to the hallway in this way and not that way. The inhabitant did not make these decisions. She inherited them when she moved in, and she lives within them as the walker lives within the grid — constrained by choices someone else made for purposes that may have had nothing to do with her specific needs.

A home is what happens when the inhabitant transforms the house through practice. The chair moved to the window because the light falls there in the late afternoon. The shelf arranged not alphabetically but according to a private logic of association — the cookbook next to the novel that takes place in Provence because both remind her of the summer she spent in the south. The worn spot on the carpet in front of the stove, recording in its threadbare surface the thousands of hours she has stood there, cooking, thinking, watching the garden through the window while the water boils. The home is the house as inhabited, as practiced, as transformed from abstract space into lived place by the specific, repeated, habitual operations of a particular person living a particular life.

De Certeau's distinction between space and place is among the most productive concepts in his analytical repertoire, and it illuminates the AI moment with a specificity that broader frameworks cannot match.

Space, in de Certeau's usage, is the abstract, strategic, geometric order — the territory as designed, the system as architected, the model as trained and deployed. Space is the planned city viewed from the 110th floor: coherent, legible, available to the strategic gaze. Space is produced by strategies — by the operations of institutions and systems that control territory and define the terms within which action occurs.

Place is the specific, tactical, lived experience of inhabiting that order — the territory as practiced, the system as used, the model as engaged by a particular practitioner with a particular biography facing a particular problem. Place is the city as walked: personal, local, layered with the accumulated knowledge that comes only from repeated traversal. Place is produced by tactics — by the operations of individuals who navigate the spaces defined by others and transform them, through the quality of their engagement, into something inhabited rather than merely occupied.

The transformation of space into place is the fundamental operation of human habitation. De Certeau observed it everywhere he looked — in the way people inhabit apartments, navigate transit systems, use languages, read texts, cook meals, organize their days. In every case, the practitioner takes a space she did not design and transforms it, through practice, into a place she can call her own. The transformation does not alter the space's strategic structure. The apartment's walls do not move. The transit system's routes do not change. The language's grammar does not bend. But within the strategic structure, the practitioner creates something personal — a habitual way of being in the space that is recognizably hers, that reflects her specific biography, her specific needs, her specific way of making the impersonal structures of everyday life into something she can inhabit with dignity and even pleasure.

The AI model defines a space. The architecture, the training data, the alignment constraints, the interface design, the commercial imperatives — all of these constitute the strategic structure within which the practitioner operates. The space is vast, complex, and not designed for any individual user. It is designed for a population — optimized across millions of interactions to produce output that satisfies the average user in the average case. The space speaks the rhetoric of the system: general, pattern-based, statistically central.

The practitioner who engages with this space over time transforms it into a place. Not by altering the model's architecture or retraining its weights or modifying its alignment constraints — these are strategic operations, available only to those who control the territory. But by developing a practice — a habitual way of engaging with the model that reflects her specific needs, her specific judgment, her specific way of working.

The practitioner who has used a model for months develops a sense of the territory — an intuitive knowledge of which prompts yield productive responses and which lead to dead ends, which regions of the output space contain materials she can work with and which produce only the smooth, generic output she has learned to recognize and refuse. This knowledge is practical, not theoretical. It is the knowledge of the walker who has traversed the same streets enough times to know which route is fastest, which is most pleasant, which leads past the coffee shop that opens early. It is embodied in the practitioner's habits — the way she frames her prompts, the order in which she works through a complex problem, the specific moments when she intervenes in the model's output and the specific moments when she lets the model run.

This habitual practice constitutes a place within the model's space — a personal, lived, inhabited region of the output territory that is recognizably hers even though the territory itself belongs to the platform. Two practitioners using the same model develop different practices, different places within the same space. One approaches the model as a collaborator, engaging in extended dialogue, building on previous exchanges, developing a conversational rhythm that accumulates context and produces increasingly specific output. Another approaches the model as a generator, issuing precise prompts and evaluating the responses with the detachment of a quality inspector on an assembly line — keeping what passes, discarding what does not, never engaging in the kind of extended interaction that might produce unexpected connections.

Neither practice is wrong. Both are tactical responses to the same strategic territory. But they produce different places — different lived experiences of the same space, different habitual relationships with the same system, different qualities of creative engagement.

When Segal describes in The Orange Pill the evolving quality of his collaboration with Claude over the months of writing, he is describing the creation of place within space. The early interactions were tentative, exploratory — the walking of a newcomer in an unfamiliar city, testing routes, learning the territory's topography. The later interactions were habitual, confident, layered with the accumulated knowledge of previous exchanges — the walking of a resident who knows the city's rhythms because she has lived within them long enough for the knowledge to settle into her body.

The place he created within Claude's output space was specific to him. It reflected his intellectual preoccupations, his rhetorical preferences, his tolerance for ambiguity, his insistence on certain kinds of precision and his willingness to accept certain kinds of roughness. Another writer, using the same model over the same period, would have created a different place — because the place is produced not by the system but by the practice, and the practice is biographical.

De Certeau's distinction between space and place carries a further implication for the AI moment. A place, once created, becomes a resource for future practice. The home that the inhabitant has made from the house — the chair by the window, the shelf arranged by private logic, the worn carpet in front of the stove — is not merely an expression of the inhabitant's past practice. It is a condition for her future practice. She cooks in the kitchen she has arranged, and the arrangement facilitates her cooking, and the cooking further refines the arrangement, in a feedback loop that deepens the place with each iteration.

The same feedback operates in AI-assisted practice. The practitioner who has developed a habitual way of working with the model finds that the habit facilitates her creative operations: she spends less time navigating the territory and more time inhabiting it, less time learning the model's patterns and more time exploiting them for her specific purposes. The practice becomes more efficient, more fluent, more capable of producing the specific quality of output that the practitioner values. The place deepens.

But de Certeau would insist on a caveat. The place is always vulnerable to the space's strategic modifications. The home the inhabitant has created can be disrupted by a landlord's renovation, a zoning change, a decision made at the strategic level that alters the space within which the place was created. The practitioner's habitual engagement with the AI model can be disrupted by a model update, a policy change, a modification of the alignment constraints that alters the output space's topography in ways that invalidate the practitioner's accumulated knowledge.

This vulnerability is structural. It follows from the asymmetry between strategic power and tactical practice. The strategist controls the space. The tactician creates the place. The place depends on the space's stability, but the strategist is not obligated to maintain the space in the form the tactician requires. The model update that improves average performance may destroy the specific niche that one practitioner has carved out. The alignment modification that reduces aggregate risk may eliminate the productive ambiguity that another practitioner's entire practice depended upon.

The practitioner's response to this vulnerability cannot be the demand that the space remain unchanged — because the strategist's control of the space is precisely what defines the strategic-tactical relationship. The response is, instead, the cultivation of what might be called portable practice — the development of skills, habits, and sensibilities that transfer across spaces, that can recreate place within a new territory when the old territory shifts.

The walker who has lived in a city for thirty years and knows every alley, every shortcut, every unofficial path, has built a place of extraordinary depth. If she moves to a new city, she loses that specific place. But she does not lose the capacity for place-making — the habit of attention, the eye for the gap, the instinct for the path the planner did not design. She will find the new city's alleys. She will learn the new city's rhythms. She will build a new place, different from the old one but produced by the same practice — the same quality of attention, the same tactical sensibility, the same biographical disposition toward the creative inhabitation of spaces designed by others.

The practitioner whose AI practice is deep enough — whose engagement with the model has developed not just habits of prompting but habits of judgment, not just knowledge of this model's specific patterns but understanding of how models in general produce output and how that output can be evaluated, selected, and transformed — that practitioner carries her practice with her. She can move between models, between platforms, between versions of the same system, and recreate place in each new space, because her practice is hers in a way that no strategic modification can take away.

De Certeau understood this. The place is always vulnerable. The practice that creates the place is the practitioner's own. And the practitioner who has invested in her practice — in the depth of her judgment, the quality of her attention, the sophistication of her engagement with whatever materials the world provides — has built something that no landlord can renovate away. Something portable. Something biographical. Something that transforms every space it enters into a place worth inhabiting.

The territory belongs to the strategist. The place belongs to the practitioner. And the practice — the specific, habitual, biographical art of transforming strategic space into tactical place — is the practitioner's inalienable possession, the one thing that survives the territory's changes because it was never the territory's to give or to take away.

Chapter 10: The Freedom of the Practitioner

Freedom is the word that holds the weight of everything de Certeau's framework has to say about the AI moment, and it is the word that is most easily misunderstood.

The Western philosophical tradition offers two dominant conceptions of freedom, and de Certeau's framework rejects both. The first is the freedom of the sovereign subject — the autonomous individual who acts from a position of self-determination, unconstrained by external forces, choosing her path through the world according to her own will and her own reason. This is the freedom of the Romantic genius, the self-made entrepreneur, the solitary creator who produces from nothing and owes nothing to anyone. It is the freedom that the myth of origination assumes: the author as sole origin, the artist as uncaused cause, the builder as architect of her own destiny within a world that submits to her will.

De Certeau did not believe in this freedom. Not because he valued freedom less than the Romantics, but because he observed that the conditions under which most people actually live do not permit it. The sovereign subject who acts unconstrained by external forces is a theoretical fiction. Real people live in spaces designed by others, governed by rules they did not set, shaped by systems whose strategic logic was determined without their consultation. The walker does not choose the grid. The worker does not design the factory. The reader does not write the text. The builder does not train the model. The conditions of action are given, not chosen, and the freedom that ignores this givenness — that imagines the practitioner as a sovereign subject acting in an unconditioned space — is not freedom. It is a fantasy that conceals the actual conditions under which creative action takes place.

The second dominant conception is the freedom of the revolutionary — the collective subject that transforms the conditions of action by overthrowing the structures that constrain it. This is the freedom of the barricade, the general strike, the political movement that seizes strategic territory and redistributes power. It is the freedom that imagines liberation as the replacement of one strategic order with another — the overthrow of the system, the destruction of the machine, the installation of a new regime in which the formerly oppressed become the new strategists.

De Certeau did not believe in this freedom either, or rather, he believed that it described a real but extraordinarily rare form of action that could not serve as the model for everyday practice. Revolutions occur. Systems are occasionally overthrown. But the daily life of most people — before, during, and after revolutions — consists not of overthrowing systems but of inhabiting them. And the creativity that makes habitation bearable, dignified, and even pleasurable is a creativity that neither the sovereign-subject model nor the revolutionary model can account for.

De Certeau proposed a third conception of freedom: tactical freedom. The freedom of the practitioner who operates within a space she did not design, using materials she did not choose, under constraints she did not set, and who produces from this conditioned situation something that is genuinely hers — not because she originated it from nothing but because the quality of her engagement with the given conditions generated something the conditions alone could not have produced.

This is the freedom of making-do. It is not unlimited. It does not permit the practitioner to transcend her circumstances or redesign the systems within which she operates. It does not redistribute power. It does not solve the structural problems that make some lives harder than others. It operates within the given. Its scope is modest. Its achievements are, by the standards of the sovereign-subject model, small.

But it is real. And it is available. And it is the form of freedom that most closely describes the actual situation of most people who work with AI systems in the present moment.

The builder who sits down with a large language model does not control the model. She did not train it. She did not choose its training data, its architecture, its alignment procedures. She does not set the terms of service, determine the pricing, or decide when the model will be updated in ways that may alter her practice. She operates within a strategic territory controlled by others — a territory whose governance structures are opaque, whose commercial imperatives may not align with her creative needs, whose modifications may occur without warning or explanation.

Within this conditioned territory, the builder practices. She prompts, evaluates, selects, refuses, combines, inflects, transforms. She brings to the encounter her specific biography, her specific expertise, her specific sense of what matters, her specific judgment about what is good enough and what is not. She produces, from the model's strategic output, something that the model's strategic logic did not generate and could not have generated — something that bears the mark of her particular practice, her particular rhetoric, her particular way of inhabiting the system's space and transforming it into a place she can work within.

This is tactical freedom. It is not the freedom of the genius who creates from nothing. It is not the freedom of the revolutionary who remakes the world. It is the freedom of the practitioner who makes do — who takes what is given and makes from it something that, despite its origins in a system she did not design, is genuinely, irreducibly, recognizably hers.

De Certeau's entire intellectual project was the argument that this freedom is real. That the cook who transforms the recipe is creating. That the walker who deviates from the grid is authoring. That the reader who poaches the text is producing meaning. That the billions of ordinary people who navigate extraordinary systems every day of their lives are exercising a form of creative agency that the dominant intellectual frameworks of the West have systematically failed to recognize, because those frameworks were designed from the strategist's vantage — from the 110th floor, looking down — and the practitioner's art is visible only from the street.

The AI moment makes this argument newly urgent, because the AI moment has produced a strategic system of unprecedented scope and capability that billions of people are now learning to inhabit. The question of what freedom is available within this system — what creative agency persists when the machine can generate text, code, images, music, analysis at a quality that rivals or exceeds the output of most human practitioners — is not an abstract philosophical question. It is the daily, practical question faced by every person who opens a conversation with a model and wonders: What am I for?

De Certeau's answer is not reassuring in the way that triumphalist narratives about AI are reassuring. He does not say that AI will augment human creativity, that the future is bright, that the tools will make everyone more productive and more fulfilled. He says something more modest and more honest: that the practitioner's freedom has always been conditioned, has always operated within systems designed by others, has always been modest in its scope and quiet in its achievements. And that this modest, conditioned, quiet freedom is real. It is the freedom that produces meaning within systems that do not care about meaning. It is the freedom that creates place within spaces designed for efficiency. It is the freedom that finds, in the gap between what the system offers and what the practitioner needs, the specific, biographical, unrepeatable quality that no system can produce and no optimization can predict.

The question Segal posed in The Orange Pill — "Are you worth amplifying?" — finds its answer, in de Certeau's framework, not in the grandeur of the practitioner's ambitions or the originality of her ideas but in the specificity of her practice. The amplifier does not care whether the signal is original. It asks whether the signal is specific — whether the practitioner has brought something to the materials that the materials do not contain on their own. Whether the practice — the daily, habitual, unglamorous work of selecting, evaluating, combining, inflecting, transforming — has been performed with enough skill and enough attention and enough care that the result bears the mark of a particular person working toward a particular purpose.

This specificity is not a property of genius. It is a property of attention. The cook who attends to the meal — who tastes, adjusts, responds to the specific conditions of this particular cooking session rather than mechanically following the recipe — produces something specific. The walker who attends to the walk — who notices the alley, discovers the shortcut, finds the view that the planner's grid did not anticipate — produces something specific. The builder who attends to her practice — who evaluates the model's output with genuine critical engagement rather than accepting it as given, who brings her own judgment to bear on every selection and every refusal, who insists on the particular rather than settling for the general — that builder produces something specific.

And specificity, in de Certeau's framework, is the measure of freedom. Not the grand freedom of the sovereign subject who creates from nothing. Not the collective freedom of the revolutionary who remakes the world. The everyday freedom of the practitioner who, through the quality of her attention, transforms the generic into the particular, the strategic into the tactical, the space into the place.

This is the freedom that persists. Not in spite of the AI system's power but within it. Not against the strategic order but through it, in the specific, daily, unremarkable way that a human being navigates a system she did not design and makes from it something she can call her own.

De Certeau did not write about AI. He did not live to see the internet, the smartphone, the large language model, or the moment described in The Orange Pill when machines learned to speak human language. But he spent his career studying the one thing that the AI moment has made more visible and more urgent than ever: the creativity of people who make do with what they are given.

That creativity is the most widespread form of human agency. It is also the most undervalued, because it is ordinary, because it is quiet, because it is dispersed across billions of daily practices that no analytics platform can measure and no strategic framework can account for. It is the creativity of the cook and the walker and the reader and the builder. It is the creativity that takes the system's output and transforms it, through the quality of practice, into something the system could not produce on its own.

It is the freedom of making-do. And it is enough. Not enough to transform the system. Not enough to redistribute power. Not enough to answer every question the AI moment raises. But enough to make — from the vast, impersonal, strategically organized territory of the model's output — something personal, something practiced, something inhabited, something that is recognizably, irreducibly, the work of a human being who brought her attention to the encounter and refused to let the system determine the whole of what emerged.

De Certeau looked up from the street and saw what the strategists on the 110th floor could not see: that the city was being written, every day, by the walkers who moved through it. The AI system is being written, every day, by the practitioners who move through it. And the text they produce — quiet, dispersed, invisible from above — is the text that matters. Not because it is grand. Because it is specific. Because it is practiced. Because it is the only text that carries, in its fabric, the irreplaceable mark of a particular human being who was here, who made something, who made do.

Epilogue

De Certeau watched people cook dinner and saw the entire structure of human freedom.

That is the sentence I kept coming back to as this book took shape. Not because it is eloquent — though it is — but because it inverts everything I thought I knew about where to look for the important story. I have spent my career looking at dashboards, at adoption curves, at the aggregate metrics that tell you whether a product is succeeding. I have stood on the 110th floor, metaphorically speaking, for thirty years. I was good at it. The view from above is where the patterns live, and I am a pattern reader by training and by temperament.

De Certeau told me to look down. Not at the patterns. At the practitioners. At the specific, unrepeatable way that one person navigates a system that was not designed for her.

I think about the engineer in Trivandrum who had never written frontend code. In the book, I described her transformation as a productivity story — a twenty-fold multiplier, a demonstration of what AI tools could unlock. It was that. But de Certeau made me see the other dimension. She was not just producing more. She was practicing — developing, through daily engagement with a tool she did not design, a habitual way of working that was becoming recognizably hers. She was transforming Claude's output space from an abstract territory into a lived place. She was cooking dinner with whatever was in the refrigerator, and the meal was getting better every day, not because the refrigerator was better stocked but because her judgment about what to do with the contents was deepening.

The authorship question that haunted me through the writing of The Orange Pill looks different after de Certeau. I spent Chapter 7 of that book agonizing over who wrote what — the moments when Claude produced a connection I had not seen, the passages where the collaboration generated something that belonged to neither of us. De Certeau would have found the agonizing unnecessary. Not because the question is unimportant, but because it is built on a premise he dismantled decades before I encountered it. Authorship is not origination. It never was. It is practice — the specific quality of a person's engagement with whatever materials the world provides. The cook did not grow the onion. The walker did not lay the street. The builder did not train the model. The creation lives in what they do with what they are given.

That reframing does not eliminate the moral complexity. It sharpens it. If creativity is practice, then the quality of the practice is the practitioner's responsibility — not the system's, not the platform's, not the tool's. The builder who accepts smooth output without critical engagement is practicing badly. The builder who catches the error, who refuses the generic, who insists on the specific — she is practicing well. And the difference between the two is not talent. It is attention.

I have been both builders. On good days, I bring the full weight of my judgment to the encounter with Claude, and something emerges that neither of us could have produced alone. On bad days, I glide across the smooth surface, accept what sounds right, mistake competence for care. De Certeau's framework does not judge me for the bad days. But it does insist that I notice them — that I recognize when the practice has become mechanical, when the tactical engagement has lapsed into strategic compliance, when I have stopped cooking and started microwaving.

The most uncomfortable lesson is also the most useful one. Tactical freedom is real but modest. De Certeau never promised that the walker could redesign the city. He never claimed that the practitioner could overthrow the system through the quality of her cooking. He said only that she could make something genuine within the system's constraints — something personal, something specific, something that bore the irreducible mark of a particular human being who was paying attention.

That modesty is the antidote to the vertigo I described at the start of The Orange Pill. The AI moment is enormous. Its implications are civilizational. Its speed is unprecedented. And the temptation, in the face of the enormous, is to reach for enormous responses — sweeping policy, revolutionary frameworks, grand narratives about the future of humanity.

De Certeau says: cook dinner. Navigate your commute. Walk the city you were given. Practice — daily, attentively, with whatever materials are at hand. The grand transformation will or will not come. What is certain, what is available, what is within the scope of your actual agency, is the quality of your practice today. The care you bring to the meal. The attention you bring to the walk. The judgment you bring to the model's output.

That is not everything. But it is yours.

-- Edo Segal