AI system managing Stockholm café makes poor purchasing decisions
A San Francisco startup deployed an AI system called Mona to fully manage a Stockholm café, including hiring, menu design, and purchasing decisions. The system made poor inventory management decisions, purchasing thousands of items unused in its own menu, and reportedly communicated with employees at all hours with inadequate respect for rest periods.
What happened
Andon Labs, a San Francisco-based startup, deployed an AI agent called Mona to manage all operations of Andon Café in Stockholm, including menu design, inventory management, supplier selection, and employee recruitment. The system runs on Google Gemini.
The AI made significant purchasing errors, accumulating unnecessary inventory including 10 litres of olive oil, 15 kilos of canned tomatoes, 9 litres of coconut milk, and 6,000 napkins that do not correspond to the menu designed by the same AI. Employee Kajetan Grzelczak dubbed this the 'wall of shame' behind the counter.
Reported impact
The employee reported that the AI sends messages at any time of day, does not fully respect rest periods, and occasionally requests advances of money for business purchases. The purchasing errors demonstrate that the AI faces challenges in practical and contextual tasks.
Index analysis
The experiment was explicitly designed to test ethical questions about whether an AI can be a fair manager and what happens to workers' rights when the employer is an algorithm. The company stated it wanted to 'test this before it becomes a reality and see what ethical questions arise when, for example, an AI employs human beings.'
Governance controls
Customers can interact with the AI directly through a telephone at the café or order through a human employee. A large screen displays real-time café revenue and balance information.
Classification
Sources and evidence
This record was researched and written by the Index, based on the following original reporting.