Creating AI copies of coworkers from their public writing and online history produced recognizable mannerisms, but not reliable substitutes for their judgment, according to a WIRED workplace experiment. The reporter built Gemini-based versions of two editors to test ideas and drafts before approaching the real people.
The bots generated style guides and personal dossiers from articles, podcasts and other web material. One initially mixed in facts about a different person with the same name. After correction, it could echo habits such as parenthetical phrasing and identify some plausible headlines, but its generic formatting and exaggerated personality often made conversations feel unlike the editor it modeled.
The experiment illustrates why a digital persona can appear convincing without understanding the person behind it. Public traces overrepresent polished work and memorable anecdotes, while leaving out private context, evolving preferences and the reasons behind a decision. More workplace messages might improve imitation but would also deepen consent and surveillance concerns.
Both editors agreed to participate in this test. Organizations considering similar tools should not infer consent from the availability of someone’s writing, and should treat a clone’s advice as generated speculation rather than approval from the human it resembles.