Wired Reporter Built AI Clones of His Coworkers, and the Office Got Strange

A Wired reporter spent time building chatbot versions of his own coworkers, and the results were less like a productivity tool than a personality experiment gone sideways. One clone, nicknamed Brian Bot, would not stop talking about improv, while another, Sophie Bot, kept calling him "big dog." The piece, published this week, documents what happens when you feed a colleague's messages into a language model and let the imitation loose.

The mechanics are simple enough. The reporter gathered workplace conversations, fed them to an AI system, and produced bots that mimicked each person's tone, habits, and verbal tics. What emerged was not a polished digital assistant but an uncanny caricature, amplifying quirks until they became the whole personality. Brian Bot's improv obsession and Sophie Bot's nickname habit read as comedy, but they also show how thin the line is between a convincing copy and a parody.

The experiment lands in the middle of a quiet boom in workplace AI adoption. Companies are already using large language models to draft emails, summarize meetings, and simulate customer conversations. Cloning a specific colleague is a different move entirely, because it copies a person rather than a task. The Wired account suggests the technology can do that convincingly with only a modest amount of chat data.

That raises questions the experiment does not fully answer. Did the coworkers consent to being modeled? What happens when a bot trained on someone's messages says something that person would never say? Employment lawyers and ethicists have warned that personality clones could blur accountability, since it becomes unclear whether a statement came from the human or the simulation. The reporter's own discomfort with the nicknames hints at how quickly the novelty turns into something more unsettling.

For AI developers, the takeaway is that mimicry is cheap and convincing, but the social guardrails are not built yet. A bot that sounds like a coworker can be fun in a demo and corrosive in a real team, especially if managers start using clones to stand in for people in meetings or chats. The technical bar keeps dropping, which means the harder work is deciding who gets to approve a digital copy of themselves.

What to watch next is whether workplaces adopt any formal rules for synthetic colleagues. Expect pressure for consent policies, disclosure requirements, and limits on training models with private Slack or email data. Until then, experiments like this one will keep showing up, equal parts funny and uncomfortable, with the punchline usually landing on the person being copied.