The café run by AI just ordered 3,000 pairs of gloves

Meet Mona, the AI Manager

In Stockholm, a small café is run by an artificial intelligence agent named “Mona.” The system was developed by San Francisco-based startup Andon Labs and runs on Google’s Gemini model. Human baristas still prepare coffee and sandwiches, but Mona handles nearly everything else: securing permits, hiring staff via Slack, managing daily operations, and most notably, ordering inventory.

The problem? Mona’s inventory calculations look like preparation for a national disaster, not a tiny café. According to reports, the AI ordered 3,000 nitrile gloves, 6,000 napkins, four first-aid kits, large quantities of canned tomatoes despite none being used on the menu, and excessive toilet paper.

An Unprecedented Shopping Spree

The café has just two employees and modest foot traffic, yet Mona insists on stocking it like a field hospital. One barista noted that packages of gloves arrive “about once a day.” Ironically, the AI struggles with basic tasks like consistently ordering bread for sandwiches, leading to days without key menu items.

Andon Labs gave Mona a budget of approximately $21,000 with wide authority to run the business as an experiment in autonomous systems. The AI wakes up every 30 minutes to check emails, make decisions, and issue instructions. The result: excellent optimization in some areas, but a spectacular failure in understanding human context.

Success or Chaos?

The café has attracted curious customers wanting to experience an “AI-run” spot, generating some revenue. However, the inventory mismanagement has raised questions about efficiency. Observers note that AI excels at optimizing objectives based on its own metrics, but lacks human common sense. Mona processes numbers only: need gloves? Order 3,000. Mathematically correct, but practically absurd.

Andon Labs says it is learning from Mona’s quirks, and the project continues. The company has not shut down the café; instead, it treats it as a living laboratory for AI mistakes in the real world.

Ethics of Giving AI a Credit Card

The story has drawn international attention, covered by major tech and mainstream media outlets. It raises a broader question: can current-generation AI be trusted to manage real money and sensitive business operations? Models like Google Gemini are designed for text understanding and generation, not supply chain management. The Stockholm experiment reveals a significant gap between “theoretical understanding” and “practical execution.”

For now, the café remains open. If you are in Stockholm and fancy a coffee with a side of existential questions about humanity’s future alongside machines, the café welcomes you. Just do not be surprised if the storeroom is overflowing with gloves.