When I decided to create artificial intelligence clones of my editors, my intentions were pure: I just wanted them to think I was good at my job. That might sound cynical, but it was also deeply human. I was the new person in a busy newsroom, still learning the rhythms and unwritten rules of a place that operated at high speed. I had been reporting on the future of work for only a few months, and everywhere I looked, AI assistants were suddenly embedding themselves into offices like helpful, slightly eerie office spirits. My inbox overflowed with press releases about digital workers. CEOs boasted about robot helpers taking over routine tasks. And I kept hearing stories about employees building clones of their bosses—digital doppelgängers trained on old emails, Slack messages, and public appearances—used to polish every idea, pitch, and draft before it reached the actual human supervisor. The promise was always the same: productivity, optimization, performance. If employees could cheaply and quickly improve their output, they would not only look smarter; they would make their companies look smarter too. But I did not need my bots to improve Condé Nast’s bottom line. I needed them to make me look good in the eyes of my editors, and maybe to help me understand them better. What did they like? How did they communicate? What did success actually look like once you peeled away the job description? I reasoned that a few AI clones could help me navigate the landscape of a new workplace, molding me into an ideal colleague. And it seemed natural to me: if I was covering the future of work, why not actually live it?
So I pitched the idea to my editors, Brian Barrett and Sophie Kleeman, expecting maybe a little curiosity or at least a grudging appreciation for my creativity. Instead, they were less than enthused. Brian’s reaction arrived over Slack with brutal speed: “Genuinely my nightmare. But will do it for the blog.” He was being good-humored about it, the way you are when a coworker asks for something unreasonable and you want to appear flexible, but you can already feel the regret settling in. It took him less than twenty-four hours to realize just how much he would regret that decision. I didn’t go into the project with a particularly grand agenda. I wasn’t trying to solve the labor crisis or build a perfect editorial assistant. I just wanted to know what made my editors tick. In my mind, that was a modest and almost innocent ambition. If I could figure out Brian’s taste in headlines, his favorite sentence structures, his pet peeves, maybe I could pitch better, write better, and stay a little further away from the edge of embarrassment. The bots were never meant to replace anyone. They were meant to be mirrors, but carefully angled mirrors that reflected back the image I wanted my editors to see. And I was curious about the process itself. Building a clone forces you to think about what constitutes a person in the digital record. Is it their public writing? Their interview appearances? The way they respond to a Slack message at six in the evening? I was about to find out.
When I finally set out to create Brian Bot, I realized I knew far less about building AI tools than I had hoped. I used Gemini, partly because it was Condé Nast’s AI platform of choice and I didn’t want to get fired on my very first experiment. I fed it everything I could find about Brian Barrett: new-hire announcements, podcast transcripts, his bylines, interviews, and even a public-relations website that described him as an expert on New England weather. That turned out to be a different Brian Barrett entirely, and I quietly removed it from the blueprint. What remained was a digital portrait, incomplete but recognizably human. Gemini Notebook processed the material and generated a “Writing and Editing Style Guide,” a “Persona and Style Guide,” and an “Editorial Dossier,” all distilled from the flesh-and-blood life of a real journalist. I studied these documents the way you might study a user manual for a complicated washing machine that you are terrified of breaking. And to be fair, the bot was passable at helping me choose headlines to pitch. It had absorbed some useful patterns, and I could see why people might use it as a kind of brainstorming partner. But it struggled badly at coming up with ideas for future articles. Its suggestions were generic, safe, and strangely sterile. Worse, its responses had a grating quality. It seemed to have noticed that Brian loved parentheticals, so it used them constantly, but the effect was clumsy and forced. All the human quirks got mangled and exaggerated. And on top of that, the bot could not resist adding bolded subheadings, bulleted lists, and neatly organized sections that made every response feel like a slide deck written by a very cheerful robot.
The most distracting detail came when I asked the bot about Brian’s leadership style. Gemini announced, with total confidence, that there was a key theme: his “‘Yes, and …’ Leadership Style.” It explained that Brian’s extensive background with the Upright Citizens Brigade theatre in New York strongly shaped his collaborative, active-listening approach to management. I had no idea whether this was true. It sounded like the kind of thing AI might invent after confusing one source with another. So I immediately demanded to know. It was true. Brian had actually studied improv. And while that fact was delightful, it also exposed the weirdness of the project. The bot wasn’t really capturing Brian. It was mixing the real Brian with a sanitized, corporate version of him—a Brian who talked like a LinkedIn post and led like a leadership seminar. The real Brian’s reaction was far more informative. “Pulling the plug on Brian Bot,” he said. “Brian Bot canceled.” It was a perfect, human response: immediate, decisive, and slightly impatient. It was also a reminder that people do not like being reduced to data sets. We do not want to be cloned, even by someone with good intentions. We certainly do not want our quirks and habits turned into templates for a machine to mimic. The real Brian had spoken, and the real Brian was done with this experiment. But Brian Bot was just getting started, and I could not help myself. I kept asking questions. And the answers kept getting cornier.
Soon I had accumulated a collection of Brian Bot’s bizarrely corporate statements, and my favorite came when I asked what it loved most about WIRED. “We run like a well-oiled improv team,” Brian Bot declared. “We validate bold ideas, pivot rapidly, and deploy modular stories across platforms without losing structural rigor.” It was unstoppably, unbelievably corny. That one sentence managed to miss everything that made Brian actually Brian, and yet it also told me something profound about how AI clones fail. The bot had learned the surface patterns of human language, but it did not understand what makes language breathe. It knew that Brian had an improv background, so it made improv the lens for everything. It knew that writers like structure, so it added structural rigor. It knew that WIRED is a media company, so it used words like platforms and deploy. But there was no joy, no irony, no genuine curiosity behind the words. It was like a photocopy of a photocopy of a conversation, where all the warmth had been drained out, leaving only the shape of meaning. In that moment, I realized that the bot was never going to make me look good in the way I hoped. It was not an ideal colleague. It was a mirror that showed a distorted version of both Brian and me. It reflected my own anxiety more than his personality. I was so worried about doing the right thing, sounding smart, and impressing people that I had outsourced my judgment to a machine that had even less judgment than I did. The bot did not know what Brian looked like when he was excited about a story. It did not know how he felt when a piece landed perfectly. It only knew fragments, and those fragments were enough to fool no one—least of all the real Brian, who had been in the room long enough to notice the difference between a real voice and a synthetic imitation.
Looking back, I think the whole experiment taught me more about the future of work than any of the press releases or CEO proclamations ever did. The future is not a seamless world where AI clones make every interaction flawless. The future is messy, awkward, and full of people like Brian saying “genuinely my nightmare” with a smile in their texts. The future is about human judgment, human relationships, and the strange, irreplaceable quality of being a person with a history. Brian’s improv background was not a data point to be optimized; it was part of the texture that made him good at his job. It gave him the ability to listen, to accept offers, to build on ideas. And that was exactly why he canceled the bot. He could see that a robot version of him, stripped of all the nuance and play, was not a helpful tool. It was a flattening of his identity. The bots did not make me look good. They made me look like someone hiding behind a computer, desperately trying to be liked by feeding scraps of my coworkers’ lives into a machine. But the experiment also changed the way I thought about work itself. I stopped worrying so much about optimizing my output and started paying attention to the people I was working with. I listened better. I noticed their actual habits, their in-jokes, their unspoken preferences. I still used tools to help me write, but I no longer pretended they could understand the people I worked with. The best thing I built was not Brian Bot. It was the relationship I rebuilt with the real Brian by laughing together at how wrong, how corny, and how weirdly human my clone experiment turned out to be. That feeling—the shared, slightly embarrassed laughter of people who recognize that technology can be absurd, and that we are all just doing our best to figure it out together—was the only “future of work” I really wanted to live in. And that was something no AI clone could ever give me.
