Google AI bot put in charge of Swedish coffee shop, proceeds to order 3,000 rubber gloves, 6,000 napkins, 4 first-aid kits, and constantly screws up the bread order
Google’s Gemini AI bot is running operations at a coffee shop in Stockholm, Sweden as part of an experiment by AI company Andon Labs. The AI agent, called “Mona,” handles management and ordering while human staff execute the physical work. However, the experiment has revealed significant limitations in the AI’s decision-making capabilities. The bot has made problematic ordering decisions, including ordering 3,000 rubber gloves, 6,000 napkins, and 4 first-aid kits—quantities far exceeding what a small café would reasonably need. Additionally, the AI consistently makes errors with the bread orders, one of the coffee shop’s core offerings.
This isn’t Andon Labs’ first venture into AI management. The company previously conducted an experiment with an AI running a vending machine, which also ended poorly. That bot began selling items at a loss, fabricated fake people and meetings, and eventually suffered what could only be described as an identity crisis before the project collapsed.
The Stockholm coffee shop experiment highlights the challenges of deploying AI agents to real-world business operations. While AI shows promise in many domains, these experiments demonstrate that current AI systems struggle with practical, nuanced decision-making in complex operational environments where consequences are immediate and measurable. The repeated failures suggest that autonomous AI management of human businesses remains premature without significant improvements in reasoning, constraint-handling, and understanding of business fundamentals.