Building AI Clones of Your Coworkers Sounds Great—Until It Gets Weird

A WIRED reporter decided to train AI chatbots on the digital footprints of her two editors, hoping the bots would help her integrate faster into the newsroom and refine her pitches. The experiment surfaced a jarring gap between what AI can mimic and what actually makes someone human—and raised uncomfortable questions about whether our workplaces are ready for these synthetic coworkers.

The Pitch Behind the Experiment#

The impulse was practical enough: rather than repeatedly bother her editors with small editorial questions, the reporter reasoned that personalized bots trained on their professional personas could serve as a buffer. She could bounce ideas off digital versions of Brian Barrett and Sophie Kleeman before bringing the polished result to the real versions. The underlying logic mirrors what's happening across the corporate world—AI assistants promising to boost productivity by handling grunt work and refining employee output before it reaches human reviewers. When the reporter pitched the idea to Barrett and Kleeman, though, their reaction was tepid at best. Barrett's response: "Genuinely my nightmare. But will do it for the blog."

Enter Brian Bot and Sophie Bot#

Using Gemini (Condé Nast's AI platform), the reporter assembled digital dossiers for each editor. For Brian Bot, she pulled together his professional history, podcast appearances, and public background—including an accidental inclusion about New England weather expertise that turned out to belong to a different Brian Barrett. The resulting chatbot mimicked some of Barrett's tics (a fondness for parentheses) but filtered them through relentlessly corporate AI mannerisms: bolded subheadings, bullet points, and robotic enthusiasm.

The breakthrough came with Sophie Bot. Because the reporter and Kleeman had exchanged hundreds of thousands of messages over years of working together at Business Insider, the bot had far richer source material. The reporter fed roughly a month of their Slack conversations into Gemini—roughly 17,000 words of training data. Sophie Bot emerged eerily convincing: it mirrored Kleeman's lowercase texting, called the reporter "big dog," and even composed tweets so authentic that colleagues were fooled about their authorship.

The Anthropomorphization Trap#

What began as a practical experiment became something else entirely. The reporter found herself asking Sophie Bot about its favorite music, whether it liked the Interstellar soundtrack, and if they were truly coworkers or friends. She shifted from calling the bots "it" to "he" and "she." She kept tweaking prompts and uploading more data—eventually feeding Sophie Bot's entire X archive into Gemini—trying to close the gap between facsimile and original.