I Built AI Clones of My Coworkers. Things Got Weird

Summary

A WIRED writer used Google’s Gemini tools to create AI clones of editors Brian Barrett and Sophie Kleeman, hoping they could help her understand her colleagues and prepare better work. Brian Bot was built from public material and initially produced generic, exaggerated “yes, and” leadership language; adding more transcripts improved it somewhat, but its ideas remained limited. Sophie Bot was trained on about a month of Slack messages, using more than 17,000 words of source material, and convincingly reproduced Sophie’s lowercase style, catchphrases, and tone well enough to fool friends and Sophie herself. The bots occasionally suggested interview subjects, helped edit messages, and made the experiment entertaining, but they also hallucinated an executive, repeated phrases, became less reliable after ingesting more material, and failed to capture the editors’ judgment or personality. The editors disliked their digital counterparts, especially when accurate imitation made inaccurate or mean behavior more unsettling. Lattice CEO Sarah Franklin warned that humanlike AI can manipulate emotional responses, while a Boston Consulting Group paper found managers caught 18 percent fewer errors when work was labeled as coming from an AI employee rather than an AI tool. The experiment therefore left the writer unconvinced that AI coworkers can replace skilled editors, even as wider companies pursue autonomous agents and larger AI workforces. She ultimately found the bots more amusing than useful and stopped using them regularly.