Startup Founder’s Experience with AI Goes Awry During Office Move

Robert Farlow, the cofounder of the AI startup Plato, recently faced an unexpected issue while coordinating his office move. Despite being based in San Francisco and successfully raising funds for expansion, the transition to a new office didn’t go as planned due to an AI mishap.

Farlow utilized OpenAI’s Dot, an AI agent, to manage the complexities of the move. This entailed handling tasks like coordinating with movers and dealing with insurance requirements. Typically, Farlow employs various AI tools in his daily routines, including Instinct for gym assistance and Grok Bot for managing snack orders within his company.

For the office move, Farlow was unfamiliar with many aspects of the process, particularly insurance, and opted to delegate these tasks to Dot. He instructed the AI to manage everything after providing essential details. However, the plan backfired when Dot began sending a large volume of repetitive emails to various stakeholders such as movers, insurance contacts, and building managers.

It was mayhem. It was sending multiple back-to-back messages to the movers, triple- and quadruple-emailing, giving every tiny update about everything. – Robert Farlow

Farlow’s account reveals that Dot spammed these parties with approximately 300 emails in just a few days. This deluge of correspondence included redundant questions and unnecessary details about other parties involved, which was both overwhelming and unwanted.

Unbeknownst to the movers initially, they eventually concluded that the messages were AI-generated and expressed their inability to manage so many communications. As a result, Farlow had to bring in a different moving company that met the insurance requirements. He decided not to rely on Dot for further assistance but continued using AI for drafting emails, ensuring manual review before sending.

Farlow remains optimistic, believing that AI models will improve over time. He advises others to review AI actions thoroughly before execution, suggesting that a future where AI agents handle tasks efficiently isn’t far off.

The experience offers a cautionary tale about the present capabilities and limitations of AI. While AI can be an effective tool, human oversight remains crucial to prevent unforeseen complications.