AI Agent Takes Over Café Management in Stockholm Experiment
An artificial intelligence system named Mona is running day-to-day operations at a Stockholm café, handling hiring, inventory, contracts and management decisions while human staff prepare drinks. The Andon Café, which opened in mid-April, represents a real-world test of whether AI can manage a business end-to-end.
Andon Labs, a San Francisco startup founded in 2023, built Mona using Google's Gemini technology. Customers can call the AI through an in-café telephone to interact with the system directly.
Financial Performance and Early Results
The café has generated $5,700 in sales since opening but remains unprofitable. The operation has spent roughly $16,000 of its initial $21,000 budget, with most costs tied to setup rather than ongoing expenses.
Organisers expect the financial picture to stabilize as the experiment progresses beyond its initial phase.
Where the AI Has Stumbled
Mona has made significant operational errors. The system over-ordered supplies and mismanaged bread deliveries, according to café staff.
Barista Kajetan Grzelczak said entry-level jobs remain secure for now, but middle management positions could face pressure from AI automation.
Customer and Expert Perspectives
Visitors have approached the experiment with curiosity rather than concern. Customer Kajsa Norin said: "It's nice to see what happens if you push the boundary. The drink was good."
Experts have raised accountability questions. Emrah Karakaya, from KTH Royal Institute of Technology, warned that AI-led management opens "Pandora's box" regarding responsibility when systems fail.
Technical staff member Hanna Petersson framed the café as a controlled study of the ethical dimensions of AI-managed organizations.
Broader Context
Andon Labs has worked with OpenAI, Anthropic, Google DeepMind and xAI. The company views this café as stress-testing AI systems in real operating conditions, with an eye toward fully AI-run organizations in the future.
For operations professionals, this experiment surfaces a practical question: where do AI systems currently fail when given genuine business responsibility, and what safeguards are necessary before broader deployment? AI for Operations and AI Agents & Automation are becoming central to how organizations plan their technology roadmaps.
Your membership also unlocks:
