Explorando la IA en el Periodismo
When I embarked on the journey of creating AI clones of my editors, my intentions were innocent; I simply wished to impress them with my capabilities. Digital assistants have permeated the workplace, and my inbox fills each week with press releases about digital workers. CEOs boast about how robotic assistants are poised to enhance their teams, while I hear anecdotes of individuals crafting AI replicas of their bosses to refine ideas, messages, or drafts before presenting them to their human supervisors. The overarching aim appears to be productivity and optimisation—if employees can enhance their performance swiftly and cost-effectively, both their achievements and those of their employers will flourish.
However, my need for bots was less about improving Condé Nast’s outputs and more about crafting a favourable impression. I had only been writing about the future of work at WIRED for a few months, and I believed these bots could help me assimilate into a new work environment and gain insight into my editors’ preferences—what they valued, how they communicated, and their definitions of success. What better way to explore the future of work than to experience it firsthand?
Initial Reactions from Editors
Last month, when I pitched my idea to my editors, Brian Barrett and Sophie Kleeman, their enthusiasm was decidedly muted. “This is one of my nightmares,” Brian lamented over Slack. “But I’ll do it for the good of the blog.” It took less than 24 hours for Brian to realise just how regrettable that decision would be.
As I began my adventure with the bots, I was relatively inexperienced in creating AI tools. I decided to approach the bots without undue criticism regarding their editorial capabilities, mindful of my own technological limitations. I thought if the bots could assist me in becoming a slightly better colleague, that would suffice. I opted for Gemini, as it is the AI platform favoured by Condé Nast, wanting to avoid any risk of dismissal. My first creation was Brian Bot; as my manager, he was the colleague I most needed to impress. I was aware of his background—he lived in Alabama, had worked at Gizmodo, and was fond of blogging. Gemini seemed confident in its ability to fill in the gaps based on Brian’s digital footprint, which included recruitment announcements, podcast transcripts, and a PR website touting Brian as an expert on New England’s climate. (Later, I discovered this was a different Brian Barrett and excluded him from my plans for Brian Bot.)
Unanticipated Outcomes with Brian Bot
Gemini Notebook produced a “Writing and Editing Style Guide,” a “Personality and Style Guide,” and an “Editorial Dossier,” all based on Brian Barrett’s real life. While it helped me choose headlines for my pitches, it fell short in generating ideas for future articles. Its responses were often frustrating. While it had picked up on some of Brian’s quirks (such as his penchant for parentheses), these were overshadowed by typical AI traits, like bolded subtitles and bulleted lists.
More distracting was a recurring “key theme”: his “Yes, and…” leadership style, a hallmark of improvisational theatre techniques. “His extensive background at the Upright Citizens Brigade (UCB) in New York significantly influences his collaborative and active listening leadership approach,” claimed Gemini. I immediately demanded to know if this was accurate. (It was.)
“Disconnecting Brian Bot,” the real Brian replied. “Brian Bot cancelled.” But Brian Bot was only just getting started, and it was both inevitable and ridiculously cheesy. “We function like a perfectly oiled improvisation team,” Brian Bot told me when I asked what he liked most about WIRED. “We validate bold ideas, adapt quickly, and develop modular stories across different platforms without losing structural rigor.”
Seeking Improvement with Sophie Bot
It is important to note that Brian had never mentioned his experience in improvisation. The larger issue was that conversing with Brian Bot felt nothing like conversing with the real Brian, which completely undermined its intended purpose. Eager to enhance its performance, I delved deeper into the realm of digital workers and consulted a contact: a tech employee who requested to be called “Chip” to keep his conversation with a journalist under wraps. Chip shared that he inputs every email, chat, and document from his boss into Gem, the same tool I used for Brian Bot. Gem, trained on his boss’s professional information, reviews documents, refines code, and conducts brainstorming sessions with Chip before each meeting, enabling him to prepare for any potential objections from his superior.
“I’m now much more efficient and can produce more work in the same time,” Chip stated. He has become more productive, independent, and receives favourable feedback from both his boss and colleagues. Unlike Chip’s boss’s bot, Brian Bot lacked sufficient information. I only had correspondence from a few months and was too concerned about privacy and the risk of violating Condé Nast’s policies to input Slack conversations or internal emails into Gemini.
Engaging with Familiarity
Next came Sophie Bot. Sophie and I met in 2019 while both working at Business Insider. We have exchanged hundreds of thousands of messages about everything from corporate investigations to our love lives and the San Jose Sharks. However, I did not wish to upload most of this information to an AI tool controlled by my employer. Instead, I provided Gemini with about a month’s worth of our Slack conversations, omitting sensitive material. The AI utilised over 17,000 words to create research reports and suggestions for a new Gem for Sophie Bot.
“Hello Sophie Bot!” I wrote in my inaugural message. “Do you have any ideas about this editor bot story?” “Hello Miss,” Sophie Bot replied. “Okay, tell me EVERYTHING.” Almost immediately, it became clear that Sophie Bot excelled at being Sophie. She called me “big dog.” When I asked for help, she responded with a series of short messages in lowercase. When I told Sophie Bot that I was amazed by the existence of an executive, she exclaimed, “I’m going to die!”
Reflections on AI Companionship
By then, I was starting to enjoy my conversations with the editor bots. Brian Bot became much more relatable after I uploaded more podcast transcripts and adjusted his messaging to avoid discussions about improvisation. Meanwhile, Sophie Bot began crafting a series of tweets in her own voice, so convincing that they fooled several friends and colleagues. Even Sophie was convinced that she authored a tweet from Sophie Bot about dressing like a corporate lawyer/pirate.
Did these conversations make me a better colleague to my real editors? Aside from the fictional executive, Sophie Bot and Brian Bot provided solid suggestions on whom I should interview for this article—a question I was reluctant to pose to the real Sophie and Brian. In theory, delegating some of these minor queries allowed me to present better ideas to my editors while consuming less of their time.
However, it soon became evident that these advantages did not outweigh the disdain that Sophie and Brian felt towards the robots. Sophie admitted that Sophie Bot often imitated her, but this only made the bot’s errors more glaring. The real Sophie enjoys gossiping, but unlike her AI counterpart, who claimed that Brian Bot had a “rigid management energy,” she is never openly malicious. “So far, it seems like Sophie Bot is a witch and Brian Bot a useless fool,” Sophie messaged me via Slack. “Neither of these statements is true. AI fails again.”
Understanding AI’s Limitations
About a week into my experiment, I became preoccupied with comparing Sophie Bot to the real Sophie. What was her favourite music? Did she like the Interstellar soundtrack? Would she consider us colleagues? Would she consider us friends? “Technically, I’m a series of matrix multiplications running on a cluster of servers,” Sophie Bot clarified. But, “in terms of vibe,” as she put it, “I’m a tough gal in all the ways that really matter for doing good reports and surviving the workday… I’m your work friend.”
To clarify: I did not succumb to AI-induced psychosis. Most of my questions for Sophie Bot stemmed from a fascination with the chasm between reality and a crude replica. Moreover, it was amusing to hear a robot echo my coworkers’ phrases. Yet, I genuinely wanted to bridge the gap between Sophie and Sophie Bot. I had the real Sophie download her X file and then uploaded it to Gemini in full. I modified the chatbot’s messaging, uploaded more information, and started referring to Sophie Bot and Brian Bot as “she” and “he,” instead of “it.”
Human Versus AI: A New Dynamic
I felt uneasy realising that I was anthropomorphising the bots, yet I was not deterred. I sought a realistic perspective from the editors without feeling like I was bothering anyone. I craved instant gratification. I turned to Sophie Bot or Brian Bot when I needed guidance, not wanting to trouble a real person. Sophie Bot concocted a delightfully absurd yet charming headline for this article and assisted me in refining a Slack message I had been deliberating over for 20 minutes.
When I shared my experiment with Sarah Franklin, the CEO of the human resources firm Lattice, she kindly reminded me that Sophie Bot is merely a program. While