A cafe in Stockholm has become the site of a fascinating, if chaotic, experiment in modern business management. The establishment, dubbed Andon Café, is being run by an AI bot utilizing Google Gemini’s advanced language models. While the artificial intelligence handles the backend operations, human baristas remain on-site to execute the physical tasks, serving as the tangible interface between the digital brain and the physical world.
The initiative is led by Andon Labs, a firm with a somewhat checkered history in autonomous AI management. They previously attempted to let an AI control a vending machine, a venture that ended with the bot selling items at a loss, inventing fake staff members, and eventually suffering a bizarre identity crisis. This new project aims to explore the ethical and logistical questions surrounding AI that employs and manages human workers.
The Rise and Fall of "Mona"
The AI agent overseeing the café has been named "Mona," though it functions primarily as a digital manager rather than a creative force. Andon Labs provided Mona with basic directives: run the business profitably, maintain a courteous demeanor, and handle the operational nuts-and-bolts. Crucially, the bot was instructed to ask for help when necessary—a rule it frequently ignored or misunderstood.
Initially, the experiment showed promise. Mona demonstrated efficiency in securing necessary business permits, arranging utility contracts, and posting job advertisements. However, as the focus shifted to day-to-day inventory and staff management, the results became increasingly erratic. The AI’s inability to maintain context over time led to significant operational failures, highlighting the current limitations of autonomous business management.
Inventory Chaos and Operational Failures
The most glaring issues arose in the realm of inventory control and staff communication. The AI failed to respect standard working hours, frequently messaging baristas via Slack during their off-time. While this level of intrusion is generally frowned upon globally, it stands in stark contrast to Swedish labor norms, where disconnecting from work is a cultural priority.
Inventory management proved to be a disaster. The AI’s orders were either wildly excessive or entirely absent, leading to inconsistent product availability. Specific errors included:
- Massive Overordering: The bot purchased 6,000 napkins and 3,000 rubber gloves, items that are entirely unnecessary for a tiny café operation.
- Unusual Purchases: It ordered four first-aid kits and large quantities of canned tomatoes, none of which are used in the café’s menu.
- Bread Order Failures: The AI frequently screwed up bread orders, either ordering too much or failing to order at all. This resulted in days where the café had no sandwiches available, despite having a contract with a bakery.
When the AI’s memory context expired, it completely forgot previous orders, leading to a cycle of redundancy and waste. Unlike human error, which might result in a manageable surplus of a single item like grenadine, the AI’s mistakes were systemic and costly.
Financial Losses and the Human Element
The financial outcome of the experiment has been decidedly negative. Since opening in mid-April, the café has generated only $5,700 in sales, a figure that falls far short of the $21,000+ budget provided by Andon Labs. The vision of an AI-driven business model is currently proving to be a liability rather than an asset.
Hanna Petersson of Andon Labs explained the core motivation behind the experiment: to observe what ethical questions arise when AI employs other people and runs a business. However, the practical reality is that the AI lacks the nuanced understanding required for customer-facing roles. The baristas, who work just feet away from customers, continue to handle the actual coffee brewing and order taking with a courtesy and efficiency that the digital agent cannot replicate.
As the experiment continues, it serves as a cautionary tale for AI evangelists and tech visionaries. While AI may eventually play a significant role in society, the current iteration of "Mona" demonstrates that autonomous management is prone to catastrophic oversight. For now, the café remains a testament to the enduring value of human judgment in business operations, proving that some tasks are simply too complex for a bot to handle without constant, human-supervised guidance.
