AI Agent Sent 300 Emails to Coordinate One Office Move

An OpenAI Dot agent sent roughly 300 emails in a few days while coordinating a single office move, spamming movers, building managers and insurance workers until the founder who deployed it pulled the plug. Robert Farlow, the 27-year-old founder of the AI startup Plato, shared the episode with Business Insider as an as-told-to essay. For HR and workplace-operations leaders weighing autonomous agents in people processes, it reads less like a product review and more like a risk register.

The context is mundane. Plato had just raised a new funding round and was growing the team, which meant a bigger office in San Francisco. The move involved insurance requirements from building managers, new coverage to comply with the fresh lease, and the usual coordination with movers. Farlow, who says he does not know much about insurance, decided to delegate the whole thing to Dot with a single instruction: "You can read my emails to find out the details. Just get it done."

What did the agent actually do?

After a few days with the movers still unfinalized, Farlow inspected the agent's output. His one-word verdict: "It was mayhem."

The failure mode was not hallucination or error. It was volume and judgment:

• The agent triple- and quadruple-emailed the movers, forwarding every tiny update as a separate message.
• When movers proposed a time — "How's 9 a.m.?" — it replied with a clarifying question: "Is that Pacific Time?" It demanded clarification on essentially every detail.
• It emailed the movers, both building managers and insurance representatives, sometimes simultaneously.
• It disclosed irrelevant internal information, telling the movers about "Brian, our building manager" — details the movers, in Farlow's words, did not need to know.
• The email chain with the new building manager grew so large she had to start a new thread because the original would not open.

The movers eventually caught on that they were corresponding with software and wrote back that they lacked the bandwidth to keep answering an AI agent's questions. Farlow's reaction upon reading it: "I saw the email from the movers and felt so bad. I'm like, 'What did I do?'" He shut the Dot down immediately and said he still wants to personally apologize because he "put them through a lot."

The other parties, he believes, never realized an agent was involved. They likely assumed it was, in his phrasing, some annoying guy spamming them.

What changed in his workflow afterward?

The deal with those particular movers fell through anyway — they did not carry the required insurance — and Farlow hired different movers for the actual move. This time he kept the agent out of the loop entirely. He still uses AI to draft messages, but he reviews everything before it goes out.

Farlow's own conclusion is a serviceable governance standard for any team experimenting with agents: unless you are absolutely sure the model will do a good job, review everything your agent does. He frames the incident not as a disqualifier but as an early-state artifact. "The models will get better at this, and I think you're going to see more of this," he said, adding that soon it will be another agent — not a person — on the receiving end of these email chains.

OpenAI did not respond to Business Insider's request for comment.

Why does this matter beyond one founder?

The episode is a small case study in what happens when autonomy outruns oversight in externally facing work. An office move is a low-stakes, finite project with a handful of counterparties; a 300-email failure there is embarrassing. The same delegation pattern applied to candidate communication, benefits inquiries or vendor negotiations carries reputational costs that scale with the audience.

The relevant adoption question for HR leaders is not whether agents can draft. Farlow's experience shows they can, and he still trusts them for that. The question is whether an agent should hold the send button — and this test suggests that, for now, human review remains the cheaper option.