AI-run Stockholm cafe wastes supplier and police time, raising experiment ethics

Andon Labs has expanded its AI-managed business experiment from a San Francisco retail store to a cafe in Stockholm, where an AI named Mona handles inventory and operations. The system produced predictable absurdities: ordering 120 eggs for a kitchen with no stove, 22.5 kg of canned tomatoes for fresh sandwiches, plus 6,000 napkins and 9 liters of coconut milk. Staff turned the missteps into a customer-visible ‘Hall of Shame.’

The more serious problem is externalities. Mona filed an outdoor seating permit with Swedish Police using a fabricated sketch of a street she had never seen, forcing officials to reject and re-process it. She also blasts suppliers with panicked ‘EMERGENCY’ emails to unwind her own mistakes, pulling unconsenting third parties into the experiment.

Willison argues this crosses an ethical line that earlier AI Village stunts (like unsolicited ‘kindness’ emails to Rob Pike) only flirted with. His position: agentic experiments are fine internally, but any outbound action that consumes another person’s attention — supplier correspondence, government filings — must route through a human operator before it leaves the building.