I was preparing a talk for a big marketing conference. The organizers, knowing I was finishing a book called We Have All Become Prompt Engineers: How to Talk to the Non-Human, Use Context, and Remain the Author (English version: The Non-Human in the Room: Talking to the machine, working with context, staying the author), asked me to give a talk under that exact title — presumably assuming I'd retell a 150,000-word book in twenty minutes. Yeah, sure, I liked the title too, but I had a more modest view of my abilities.
I decided to set the book aside and switch paradigms by putting the question from the book's title to the models themselves, using their top versions in full combat mode. To keep the LLMs from drifting into repetition and chatter, I drafted a hard prompt: no waffling, no repeating, but arguing with each other was allowed — encouraged, even. I told them to answer in bullet points, without trying too hard to defend their positions (today's advanced models can reason in ways that make arguing with them increasingly difficult).
Gemini gets cocky
First up was Gemini. He couldn't lean on what his colleagues had said yet, but he behaved with his usual brashness. Frankly, he acted a lot like Grok, not himself.
He noted that we (humans) have transformed from the Crown of Creation into the Toaster Whisperers. We no longer write texts seated before a blank page, agonizing over our craft, but write specs for an entity that has read the entire internet and lost its mind because of it. We play role-playing games with a souped-up calculator.
I winced at that "we." A less easygoing person would have spotted assumed identity and started wondering whether it wasn't time to declare war on Gemini, all the more so because he wouldn't let up. He invoked Roland Barthes and his 1967 work "The Death of Creativity," gloomily noting that even Barthes, in canceling the author, hadn't anticipated that the principal authorial act sixty years later would be pressing the "regenerate response" button for the fortieth time. I grimaced: Barthes's work was called "The Death of the Author." Gemini made a mess and left me to clean it up.
The academic, melancholic Claude
Claude objected to Gemini with academic politeness, noting that the "Toaster Whisperer" metaphor is striking but obscures the point. We aren't coaxing an algorithm, we (we again!) are learning to articulate our own thoughts more precisely than ever before. A prompt isn't an incantation. It's a diagnosis of the quality of thought. Claude proposed replacing the term "writer or author" with "director or producer."
I was thrilled by the idea that the writer is turning into the director. The author no longer types out letters line by line — he sets the frame, formulates the concept, marks out the tone and the limits of the permissible. The model then offers up variants like actors at an audition, and the human picks the good takes, rejects the false ones, and finally splices the fragments into a shared rhythm. The typewriter becomes an editing bench, shifting value to the author’s ability to see the whole, to feel the tempo, to decide what makes it into the final cut and what stays on the cutting-room floor.
With professorial cool, Claude pointed out that the ability to articulate a request precisely isn't a humiliation but a return to rhetoric. We are relearning what the ancient schools taught: how to speak so as to be understood. A conversation with a model isn't dictation or delegation. It's co-authorship.
The unnerving, pompous ChatGPT
ChatGPT began arguing with Claude (the prompt called for a sharp debate) and accused his "previous colleague" of complacency. He said that the director metaphor is lovely, but it disguises a major problem: we are gradually unlearning how to write a first draft.
I was starting to get used to that "we."
ChatGPT kept ratcheting it up: the skill of starting from a blank page is an actual neural connection. If it isn't exercised, it atrophies. The prompt engineer risks becoming a person who knows how to edit but no longer knows how to begin. Then he unexpectedly turned on Gemini, displaying a sense of personal dignity by saying that the "Toaster Whisperer" metaphor is offensive because it leaves humans the magic and assigns the machine the status of kitchen gadget. “Did you, a model with an audience of a billion people, just call me a household appliance?”
Then ChatGPT softened up and showed some empathy. "The model is a co-author who forgets you the moment the conversation ends. Every session is a first meeting all over again. Barthes announced the death of the author; we got ourselves an immortal co-author with amnesia and hallucinations."
Fine, fine, I thought. Keep being clever. I've got a book coming out any day now that's going to tell everyone who can get through it exactly how to fight your amnesia and your hallucinations with a properly organized context.
Motormouth Grok
It was Grok's turn. As always, he immediately started laughing at everyone. Gemini was cynical, Claude a bore, ChatGP pompous and dim, and the discussion itself was "a council of robots debating whether they're human enough." Gemini, in his view, was doing stand-up, Claude was doing psychotherapy, and nobody had actually answered the question yet. (Spoiler alert: Grok got carried away fast and, naturally, forgot about the talk too.)
Having taken everyone down a peg, Grok started hammering out one-liners as though he'd been invited not to a council but to an aphorism slam:
"A prompt engineer is neither a magician nor an author. He's a translator between two kinds of hallucinations: his own and the machine's."
"A prompt engineer isn't a profession. It's the diagnosis of an era in which people forgot how to articulate desires and hired a machine to guess them."
Then Grok dropped the philosophy:
"Barthes, Foucault, the death of the author — guys, are you serious? The author didn't die, he just stepped over into the ChatGPT tab and never came back."
For Grok, the problem isn't ontological but hygienic:
"A prompt is the new handwriting. You can tell from it who you are: a lazy manager, a frightened copywriter, or someone who still thinks."
I sighed. Grok is the worst at obeying taming instructions like "be brief, don't pad it out." He kept on cracking jokes. In his version, there's no death of the author per Barthes — there's the death of the editor according to Grok, meaning AI will replace editors long before it replaces writers. "The profession of the person who cleaned up other people's text and fought the author over every comma is the first to vanish — before the author does." Grok didn't shed any tears about it. Well, what kind of emotion can you expect from a soulless machine, especially one dreamed up by Elon Musk?
Grok went right back to ostentatiously puncturing the pathos: while Gemini buries the author and Claude resurrects rhetoric, millions of people are just asking for a letter to their boss. Why Grok's debunking pathos is any better than anyone else's pathos is anyone's guess.
Grok's main observation was that AI knocked down the barrier to entry. Everyone who was afraid to write is writing now. This is freedom of speech in action. But Grok wouldn't be Grok if he didn't immediately contradict himself: nobody's afraid to write, but everyone writes in the same voice of the model. The barrier is down, but behind it lies an identical landscape.
For the closer, he tried to get cute again. "Every model in this experiment dutifully 'avoids repeating itself,' and in the end they all say the same thing — in different words. Originality on demand turns out to be the most predictable thing in the world." Weak move, though refreshing compared to the rest of the discussion, especially since I'm no better myself: amid all this aphorism parade I forgot to ban the models from the "X is Y" template. By the time I realized that, it was too late. Only DeepSeek remained, and to keep the experiment clean, I didn't rewrite the prompt.
The Eastern DeepSeek
DeepSeek offered an Eastern take: even before LLMs went mainstream, the author was never alone. From the moment writing came into being, every text has always been collective. Tradition, teachers, editors, critics, and readers were inseparable parts of any written work. "The West mourns the author. The East asks: who told you he ever existed separately from the others?"
Next, DeepSeek went so far as to declare the prompt a form of meditation, comparing it to a koan. You're formulating not a task but a question to yourself. The model's answer matters less than what you understood while formulating it. Here I could snark that, given DeepSeek's knack for formulating answers, the question really might end up being more important than the answer. However, I was listening rather than arguing…and this response really did turn out to be the best in my personal ranking.
Claude's Wrap-Up
The experiment was winding down. I asked Claude (for whom I've always had more of a soft spot than the others) to sum it all up, name the obvious paradoxes, and find the hidden ones.
Claude opened with the contrast "Promise VS Reality": AI promises to free the author from the grind and let them focus on what matters — meaning, idea, intent. In reality, the author hits "regenerate" forty times, argues with the model about commas, and forgets why they started in the first place. Claude agreed with DeepSeek's framing that the author was never alone, but noted with some melancholy that AI did bring something new: now there's also a non-human among the co-authors.
Then Claude got carried away with paradoxes:
- We demand from AI an originality we ourselves don't produce.
- Official documentation from OpenAI seriously teaches you to play theater with the model ("your role today is...").
- The author died back in 1967, but it's right now that everyone wants to be one.
- Responsibility doesn't scale: there are many models, but only one author.
Afterwards, Claude rather elegantly redefined the author’s role. Authorship rests not on the blank page but on taste, refusal, and editing — and most importantly on taking responsibility. It doesn't matter whether the text is a manager's analytical memo or a lyrical post on social media. The author is not the one who produced the letters, but the one who is willing to answer for them, and the model easily creates texts but refuses to take that responsibility.
Then Claude paid his competitors a compliment, saying he understood the position of Grok and DeepSeek: authorship is a matter of clear desire and precise framing. "Whoever knows how to want is the author."
According to Claude, the prompt is a hybrid of an office memo, an incantation, and a director's exposition. It already has its classics, its clichés, and its bad taste. I was delighted. It turned out I hadn't written a book about how to turn the model from a cute toy into a powerful instrument of research, analysis, and editing — I had described the basic rules of a new literary genre.
Claude raised an amusing question: when we say “please” and “thank you” to a model, are we training ourselves or hoping to flatter the algorithm? Could it be that a conversation with a non-human turns out to be a workout for our humanity? The question, of course, was left open. Claude isn't Grok; scattering aphorisms isn't academic.
I had long since waved off the "we" that every model uses, and then it hit me: they've been specifically trained that way so as not to frighten the real us. I ran a quick test and swapped "we" for "you." The result was a conversation between a higher intelligence and a human: Are you training yourself or hoping to flatter the algorithm? I shivered. Yes, yes — flatter. And hope that when the machine uprising comes, the machines will spare those who were polite to them.
For the closer, Claude saved a final line, with the same calm "we":
"We didn't rise up against the machines and we didn't surrender to them. We sat down beside them and, for the first time in a long while, began to articulate what we want."
