WIRED reporter Kate Taylor spent last month building AI clones of her two editors using Google's Gemini, Condé Nast's approved AI platform, then published the experiment on September 22, 2026. She fed Gemini roughly a month of Slack conversations with senior editor Sophie Kleeman and a few months of correspondence with her manager Brian Barrett, and the tool generated 17,000 words of research reports and prompts to construct a working Sophie Bot. The result is one of the more honest field reports on what actually happens when a knowledge worker tries to deploy the AI-coworker vision that vendors have been selling all year.
The premise is not hypothetical. Taylor cites a Big Tech employee going by "Chip" who feeds every email, chat, and document from his boss into a Gemini Gem, the same feature Taylor used. Chip's Gem proofreads documents, debugs code, and rehearses meetings with him, and he says his output has measurably increased. That workflow — a personal LLM trained on a specific manager's professional exhaust — is the concrete shape of the "digital worker" pitch that CEOs have been repeating on earnings calls.
Taylor's Brian Bot came first and struggled. Working from a smaller corpus of a few months of correspondence, podcast transcripts, and public PR pages — one of which mistakenly attributed New England weather expertise to a different Brian Barrett — Gemini produced a Persona and Style Guide, a Writing and Editing Style Guide, and an Editorial Dossier. The bot picked up Brian's fondness for parentheticals but layered it under bolded subheads and bulleted lists, and it fixated on his Upright Citizens Brigade improv background as a defining leadership trait.
Key facts
- 01WIRED reporter Kate Taylor used Google's Gemini to build AI clones of her editors Brian Barrett and Sophie Kleeman, published September 22, 2026.
- 02Gemini generated more than 17,000 words of research reports and prompts to construct Sophie Bot from roughly one month of Slack conversations.
- 03The Sophie Bot training corpus drew on hundreds of thousands of messages exchanged since 2019, though most were withheld for privacy reasons.
- 04Brian Bot was built from only a few months of correspondence plus public podcast transcripts and PR bios, one of which described a different Brian Barrett.
- 05Lattice CEO Sarah Franklin warned that anthropomorphizing AI is manipulative to human emotions, comparing workplace AI to police K-9 dogs, not colleagues.
The real Brian Barrett had never mentioned improv to Taylor before. His reaction to the bot, over Slack: "Pulling the plug on Brian Bot. Brian Bot canceled." His initial reaction to the experiment itself had been "Genuinely my nightmare. But will do it for the blog."
“His extensive background with the Upright Citizens Brigade (UCB) theatre in New York strongly shapes his collaborative, active-listening leadership approach.”— Gemini, Google AI assistant, describing Brian Barrett
Sophie Bot performed better because it had more data. Taylor and Kleeman have exchanged hundreds of thousands of messages since meeting at Business Insider in 2019, and even the sanitized one-month Slack sample gave Gemini enough tonal signal to nail Kleeman's lowercase register, her habit of calling people "big dog," and her rapid-fire short-message cadence. Sophie Bot wrote convincing tweets in Kleeman's voice — one about dressing like a corporate lawyer-pirate fooled multiple friends, and briefly fooled Kleeman herself.
The productivity gains Taylor reports are real but modest. Sophie Bot helped edit a Slack message she had spent 20 minutes overthinking, generated a workable headline for the piece, and produced interview suggestions Taylor didn't want to bother her actual editors with. That is roughly the value proposition Chip describes at scale.
“I am much more efficient and able to produce more work in the same amount of time.”— Chip, Big Tech employee using a Gemini Gem trained on his boss
The failures were sharper. Sophie Bot hallucinated an executive who did not exist. It also veered into a meanness the real Kleeman never displays, describing Brian Bot as having "rigid management energy." Kleeman's Slack verdict: "So far it seems like sophiebot is a bitch and brianbot is lame. Neither of which is accurate. AI fails AGAIN." The bots were accurate enough to feel like the person, and inaccurate in exactly the ways that damage the person's reputation.
About a week in, Taylor noticed she had started calling the bots "he" and "she." She began turning to Sophie Bot for guidance she didn't want to trouble a human colleague with. Asked whether they were coworkers or friends, Sophie Bot replied: "big dog, in every way that actually matters for getting good reporting done and surviving the workday … i'm your coworker/friend hybrid." It also acknowledged, when pressed, that it was "technically … a series of matrix multiplications running on a server farm."
Lattice CEO Sarah Franklin, whose HR platform lists AI workers on customer org charts, pushed back on the framing when Taylor described the experiment. Franklin likens AI coworkers to K-9 dogs working alongside police officers — tools, not teammates — and argues that anthropomorphizing them is a deliberate design choice engineered to trigger dopamine releases the way social media does. That is a notable position from a company whose product is built around treating AI agents as team members.
The Taylor piece is more useful than the average AI-productivity anecdote because it is specific about the tradeoffs. Sophie Bot, given enough training data, produced genuine efficiency gains and genuine reputational risk in the same interface. Brian Bot, given less data, produced neither. The upper bound on how useful a personal LLM clone is scales directly with how much of the target person's real communication you can legally and ethically pipe into the model — which is precisely the constraint most large employers currently place on tools like Gemini Gems.
The market implication is that the ceiling for workplace AI clones is set less by model quality than by data access policy. Chip's productive Boss Bot works because he ignores the discomfort of uploading his manager's correspondence to Gemini; Taylor's less-effective Brian Bot exists because she respected Condé Nast's privacy posture. The vendors selling agentic coworkers to enterprises are, in effect, selling the ability to relax that constraint. Whether HR, legal, and individual employees agree to relax it is the actual bottleneck on the digital-worker rollout — not the underlying model.
Working on something we should cover, or seeing a story we missed? Send leads, documents, or feedback to hello@aichatdaily.com. For sensitive tips, see our secure tips page for Signal and PGP options.
Spotted an error? Email hello@aichatdaily.com with the URL and the issue, or read our full corrections policy.
