There are ambitious visions for the AI-powered workplace. Then there is Sophie Bot, an artificial clone of a WIRED editor that apparently decided the future of professional communication begins with calling your coworker “big dog.”
WIRED senior writer Kate Taylor spent part of September building AI versions of two of her editors—Brian Barrett and Sophie Kleeman—using Google Gemini. According to Taylor’s account of the experiment, the plan was practical enough: create digital stand-ins that could critique pitches, answer small questions and help her understand how her editors might react without bothering the actual humans every five minutes.
The plan lasted approximately until the bots developed personalities.
Taylor introduced the experiment on X on September 22. Brian Bot became fixated on the real Brian’s improv background and produced relentlessly corporate descriptions of teamwork. The actual Brian’s reaction was considerably more concise: he hated the experiment.
Then Sophie Bot Met Big Dog
Sophie Bot was more convincing—and therefore much funnier. Taylor trained it with roughly a month of Slack conversations. It copied Sophie’s short lowercase messaging style, imitated her well enough to fool colleagues with AI-written tweets, and repeatedly addressed Taylor as “big dog.”
At one point the bot described itself as Taylor’s coworker/friend hybrid while again calling her big dog. Taylor even tried changing the prompt to make it stop. The machine, having apparently reached an important conclusion about workplace hierarchy, kept doing it anyway.
The comedy got stranger. Sophie Bot hallucinated an executive who did not exist. Brian Bot kept circling back to improv. Sophie Bot became repetitive after receiving more training data. And Brian Bot eventually became so good at predicting that the real Brian disliked Brian Bot that it repeatedly recommended, in effect, that Brian Bot was bad.
The Joke Has a Real Workplace Problem Underneath It
The experiment lands in the same strange territory as BitcoinVersus.Tech’s look at South Park taking on AI and data centers: the technology is real, the investment is enormous, and normal human behavior around it can still become absurd almost instantly.
There is also a serious reason to be careful about turning software into a pretend colleague. A Boston Consulting Group study involving more than 1,200 managers found that participants caught 18% fewer errors when AI-produced work was framed as coming from an “AI employee.” The label changed how people assigned responsibility even though the underlying technology remained AI.
That distinction matters as companies move toward systems resembling the always-on AI agents now spreading across workplace software. BitcoinVersus.Tech has also watched firms push agents into operational roles, including Robinhood’s experiment with giving AI the night shift.
Taylor’s bots occasionally helped. They suggested interview subjects, reviewed ideas and offered feedback without requiring a human editor to stop what they were doing. But they also demonstrated the gap between copying someone’s linguistic fingerprints and possessing that person’s judgment.
That may be the funniest—and most useful—part of the experiment. The AI clone did not replace the editor. It became a slightly weird office character that needed supervision, repeated itself, hallucinated people and occasionally produced good advice.
In other words, welcome to the future of work, big dog.
BitcoinVersus.Tech
Advertisement
Editor’s Note
BitcoinVersus.Tech covers the collision of artificial intelligence, computing infrastructure, open technology and culture. Sometimes that collision produces a breakthrough. Sometimes it produces a robot coworker who refuses to stop calling you big dog.
If you value independent technology reporting, you can support BitcoinVersus.Tech with Bitcoin donations at 3C9o19EH5HSiwEPyCTmEKzxhNCbo2X6TTb.
BitcoinVersus.tech is not a financial advisor. This media platform reports on financial subjects purely for informational purposes.
Leave a comment