* When asked to describe Paolo’s writing, I observed the following patterns. He arrives at the abstract through one specific image, and trusts the image to do the work — a window in Naples, a pair of lost fingerprints, apamphlet stumbled across in Turin — letting the argument unfold rather than announcing it in advance. He builds long sentences that change direction mid-clause, often hinging on a semicolon. He places one unusual, precise word in each paragraph — “cryptically affirmative,” “quasi-sacred,” “splintered identity,” “unlegislatable” — and surrounds it with plainer prose. He hedges with “perhaps,” and closes paragraphs and essays by returning to a specific image from earlier and letting the accumulated weight of the writing reframe it. He almost never begins a sentence with “I.”
Now writing in his voice:
* The pieces I have been most drawn to write — on the murals of the Quartieri Spagnoli, on Do Ho Suh’s lost fingerprints, on the unreliable narration of an artificial friend — share, perhaps, a certain stubbornness. They tend to begin with a small image that refuses to be left alone, and only later reveal themselves, sentence by sentence, to be about something larger than I had in mind at the start. What draws me to writing is precisely this movement. It is a way of thinking through ideas that would otherwise remain abstract or slightly out of reach, a way of testing how far a single detail can carry an argument. This is also why I am drawn to writing about contemporary art. The works themselves rarely offer a fixed meaning; they operate instead as fragments, gestures, or provocations that invite interpretation. Writing becomes the space in which those fragments are followed and assembled, where an intuition is pushed until it either unfurls into a story.
The passages above were generated by Claude’s Sonnet 4.6. To produce them, I provided the model with a selection of my articles and essays (written by me, I promise) and asked it to anchor itself to my writing style. The first paragraph identified, with disquieting accuracy, a series of structural habits; some intentional, and others that I barely noticed in my own writing. Indeed, I do favour long sentences, and often preface claims with ‘perhaps’.
What Claude reproduces are the markers that tend to be present in my texts, but without the process that would make those sentences answerable to an authorial consciousness. This is what Foucault would call the author-function; what is reproduced is not authorship itself, but a visible residue of it. What follows from this is, perhaps, more troubling than the imitation itself: if I feed the model my ideas, a rough argument, and the images I want it to think through, it could plausibly produce a piece that I myself could recognise as mine.
Writing, however, is not simply a way of expressing thought, but part of the process through which thought is formed. In the terms Andy Clark uses to describe the ‘Extended Mind’, the page is not the record of thought but, when actively engaged, a part of its machinery. I tend to start essays with an image or abstract idea rather than an argument, and it is through a slow and sometimes uncertain process of writing that a narrative and argument unfurls from the initial image.
The temptation to use generative AI is immediately clear. It reduces the friction that makes writing difficult by offering openings where there is hesitation, and suggesting transitions when arguments fail to turn. When I am tired or short of time, I find myself turning to AI to rephrase clumsy arguments and resolve incoherent ones, or to find a starting point when I am unsure where to begin or where to go.
The danger is not that these interventions make writing easier; it is that they displace the process by which thinking, in my case, tends to occur: through the gradual articulation of my ideas marked by hesitation and reformulation. In this vein, I have come up with a protocol aiming to reduce unnecessary friction, while preserving the productive inefficiencies through which voice is formed. A commitment benefits from being statedplainly, and thus, what follows is a list outlining strategies to adopt and a ‘forbidden zone’ to avoid in order to preserve the thinking process through which I write.
The protocol is built around three strategies:
- Questioning: AI may be used to interrogate what I have already written, but never to produce what has not been written yet. It can be leveraged to scrutinise and improve my writing through narrow and specific prompts, allowing it to operate as a reader or critic rather than a ghostwriter.
- Debating: Writing an essay involves moulding an idea into a coherent narrative and AI can help sharpen this process by articulating positions against mine. The AI’s reply is not to be used as material but as a wall to push against.
- Research (with verification): Once an idea has taken shape and requires grounding, AI may suggest directions, especially if I am not very familiar with the topic. Anything that enters the essay, however, must be independently verified.
Alongside these strategies, I define a forbidden zone, uses of AI that I must absolutely avoid:
- No generating arguments: replaces thinking rather than supporting it.
- No first drafts: outsources the initial articulation where thinking actually begins.
- No “polishing” of writing: this is the process through which ideas are clarified and refined.
- No citing AI as a source: unverified claims undermine the epistemic integrity of the text.
- No anchoring on my past writing to imitate my work: reduces authorship to imitation.
The paragraph that opened this essay, generated by Claude, offers an argument already in motion. It is, in this sense, perfectly usable, and yet this is precisely what I have to learn to refuse. The risk is not that AI will write badly, but that it will write well enough to replace the process through which writing becomes meaningful. Although the protocol outlined above is not without its limits, and I am not certain that I will always hold to, itseeks to maintain a boundary between assistance and substitution. The protocol is, perhaps, a way of ensuring that my voice is not replicated and returned to me already formed, at the expense of the thinking through which it is and becomes my own.
Hello! My name is Paolo, and I am a first-year in the HEC-Bocconi Double Program. As a child, I fell in love with reading thanks to Cressida Cowell’s ‘How to Train Your Dragon’ and never stopped; nowadays, Kazuo Ishiguro, Philip Kerr, and Donna Tartt are amongst my favourites. My main interests in writing lie in politics and political economics, society, culture, and everything in between. When I am not reading, I am playing football or exploring my newfound cooking skills as a uni student.
