Writing with a Machine in the Room

I write with a machine in the room. She talks fast, knows a great deal and draws connections in seconds that used to take me an afternoon. She never grumbles, never asks for coffee and is always available. On good days, working together feels like a good conversation: I say something, she sharpens it, I cut what does not fit, and something emerges that I would not have made so quickly on my own.

On other days I notice that I am letting her talk too much.

That is the moment that made me want to write this essay. Not as a technical guide to AI tools. Not as a moral objection to automation. But as an honest exploration of what happens when you use a powerful instrument so often that it begins to feel like an extension of yourself. And if that extension has a voice of its own, what does that mean for yours?

I use AI in my writing and development process for three things. I ask for sources and facts. I let it feed me with angles for workshops and talks. And I use it to build structures: headings, paragraphs, transitions, the skeleton on which my own sentences can land. That works. It works well. And then, on a day that feels unremarkable, it works a little too well.

Without noticing, I am leaning on the machine more heavily than I realise.

The text that appears is well-formulated. It flows, it is coherent, it is complete. But something is different. I can still hear my own tone, but I feel less grounded. My intuition, that thin channel between experience and language, is drowned out by a perfectly formulated middle path. The text glides. But it lives less.

This is not a feeling I was keen to acknowledge. Because it touches something that goes beyond writing.

Beneath that shift lies a dynamic I also see in organisations, in teams that have been deciding too fast, too much, for too long, without asking the deeper questions. Speed seduces. Depth costs. We trade the warmth of craft for the coolness of efficiency and call it progress. In strategic questions, teams recognise this pattern: the urge to decide before the meaning is clear. In leadership development, the same: action collapses into automatism when we do not question the undercurrent. In my writing process, the same thing is happening.

The machine accelerates my assumptions too. When I ask unclearly, I receive a polished version of my own unclarity.

There is also something psychological at work. I project authority onto the machine. She sounds certain, so she must be right. At the same time, a kind of dependency develops: if she has formulated it this way, I need to doubt myself less. In psychodynamics, we call this transference, the unconscious attribution of authority to a figure outside yourself. It applies to people in relationships, to employees in relation to managers, and apparently also to writers in relation to their language model. The machine has no intentions, but I behave as though she does. I ask her to give direction to something I ought to know myself.

It leads to what I call a productive addiction. More output. Less ownership.

The question underneath is fundamental: where is the source of the work? Does meaning come from within, from experience, perception, conscience, or from outside, from patterns, predictable language and aggregated wisdom drawn from millions of texts? Transformation from within, the concept at the centre of the work I do with organisations, requires that the centre remains in me. Not egocentric, but accountable. I may use tools. But I remain the maker.

Research shows that people prefer texts written by AI when they believe them to be human, but that their appreciation drops as soon as they know a machine was involved. That is a strange finding. We do not mind a machine contributing, as long as we do not know about it. And that sharpens the question: am I writing for the reader who does not know, or for the reader I genuinely want to reach?

I choose the second.

That means I take three risks in my own practice seriously.

The first risk is pace over texture. The machine accelerates everything, including my unclarities. When I ask an unclear question, I receive a polished version of that unclarity back. It sounds like an answer but is merely my question in a better coat. The solution is rhythm: slow down first, then speed up. I have started writing a short source letter to myself, by hand, before I switch on the machine. What have I experienced? What have I observed? What do I know that I have not yet expressed? Only once that letter exists do I invite the machine to think alongside me. She receives a starting point, not an empty space.

The second risk is form crowding out voice. A perfect structure can obscure my own tone. I notice it when my metaphors become flatter, when my sentences stop breathing, when the text is fluent but without friction. Then I switch the machine off and rewrite one paragraph from a concrete scene: a conversation with a director that did not go as expected, a silence in a team that said something words could not, a moment of friction I only understood afterwards. Experience reanimates. An AI can generate patterns. But it has not lived through anything.

The third risk is completeness as disguised fear. The machine makes it easy to want to cover everything, all the models, all the perspectives, all the disclaimers. But completeness is often fear of omission. The fear that someone will say: but you forgot X. So I add X, and Y, and Z, until the text is so complete it says nothing. Instead of everything, I choose sharpness. Which single thought do I want to send into the world right now? Which boundary does this text draw consciously? What do I leave out on purpose?

A text that wants to say everything says nothing.

At the same time, I see the enormous value of what has become possible. In workshops I use AI as an idea generator for formats: three alternatives in five minutes, whilst I hold the question open and guard it myself. In research I use it as an index: what have I overlooked, which angle is missing? In writing I use it as a mirror: how does this paragraph read when I halve it, what is lost and what is not? It works the moment the contract is clear. You support. I decide.

That contract sounds simple. It is not. Because it requires that I know what I want to say before I ask how to say it. And that is precisely what the machine cannot take over, even if she sometimes behaves as though she can.

I have developed a number of working agreements for myself that help me keep that contract. I first formulate my hypothesis or intuitive core before I ask a question. The machine receives a direction, not an invitation to invent one. After generating, I plan a short offline pause, a moment away from the screen, so that the content can settle before I adopt anything. In every text I name at least one personal observation or case that only I could have, something that cannot be drawn from a dataset. I note in my working log which parts were suggested by AI and what I consciously changed, not as accountability to others but as discipline for myself. And I check facts for their source, draw wisdom from experience, and carry style myself.

None of this is a moral exercise. It is an exercise in maturity.

Thoughtful leaders recognise this pattern. In relationships, in teams, in strategy. We work with powerful instruments, and precisely for that reason, setting limits is a form of care: for the craft, for the reader, for the truth of the moment.

The machine amplifies what is already there. When I am empty, it amplifies emptiness. When I am unclear, it amplifies unclarity in attractive packaging. When I am moved by something, it can sharpen my words and help carry the message. It is an amplifier, not a source. And the difference between those two determines everything.

There is also a social dimension to this that is rarely named. When I publish a text that has largely been generated by AI without saying so, I am claiming something I have not fully made. That feels uncomfortable. Not because using AI is forbidden, but because the reader expects something from me and receives something else. The relationship between writer and reader is one of implicit trust: you have thought this, I read it, and through that text we meet each other. If the machine has done most of the thinking, that encounter is an illusion.

That does not mean I never use AI in my writing. I do, and I do not hide it. But it means I have to draw the boundary consciously. Which part is mine? Which part belongs to the machine? And what proportion do I owe my reader?

There is one more thing I want to say, and it may be the most essential. Writing is not a by-product of thinking for me. It is the place where thinking truly begins. I do not know exactly what I think until I have written it down. Language is not the packaging of the thought. It is the thought itself, in the making. If I hand that movement over to a machine too early, I lose not only my voice. I lose the chance to discover what I actually wanted to say.

That is what is truly at stake.

Not whether my texts are good enough. But whether I am still the one writing them, in the most essential sense. Whether I am still the maker, and not merely the editor of something someone else conceived.

Writing as a leadership practice: that is the invitation I see. Not to produce the perfect piece. But to exercise the inner compass that is also needed outside the text. The courage to hold an unfinished thought until it is ripe enough to be expressed. The discipline to say: this is my boundary, this is what I want to keep, this is what I will not let be taken over.

The machine may move alongside me. As a mirror. As an index. As a builder of structures in which my own sentences can land. But I choose the direction.

The machine remains a companion, not a compass.

And writing can once again do what it always did: make meaning, from the inside outward, as a movement that begins with who I am and ends with a reader who may afterwards see something just slightly differently.

That is the only thing that matters. And that is something no machine can do for me.

Notes for those who wish to read further:

1. Roland Barthes, The Pleasure of the Text (1973, Hill and Wang). On the difference between text that soothes and text that unsettles, and why friction creates meaning. A classic reference point for anyone thinking about voice and ownership in language.
2. Manfred Kets de Vries, The Leadership Mystique (2001, Pearson). On psychodynamics in leadership roles and the mechanism of transference and dependency, including outside the therapeutic context.
3. Matthew Crawford, The World Beyond Your Head (2015, Farrar, Straus and Giroux). On attention, craft and what is lost when we outsource too much to systems beyond ourselves. Sharp and contrarian.
4. Cal Newport, Deep Work (2016, Grand Central Publishing). On the cognitive costs of fragmentation and the value of uninterrupted concentration for work that genuinely matters.
5. Adam Alter, Irresistible: The Rise of Addictive Technology (2017, Penguin Press). On how digital systems create dependency, not through bad intent but through design. A useful framework for anyone who wants to understand the pull of AI tools without becoming moralistic about it.