AI manages Swedish café while humans still brew the coffee

A café in the Swedish capital is offering customers a traditional cup of coffee, but with a highly unusual twist behind the scenes: an artificial intelligence agent is running most of the business.

At the experimental Andon Café, human baristas still prepare drinks and serve customers. But nearly every other operation, from hiring staff to managing supplies, is being handled by an AI agent named "Mona," developed by San Francisco-based Andon Labs and powered by Google's Gemini system.

The experiment is part of Andon Labs' effort to test how AI could function in real-world businesses. The startup says it wants to explore a future where organizations may be run autonomously by AI systems.

Since opening in mid-April, the café has generated more than $5,700 in sales. However, it has also spent heavily on setup costs, leaving it with less than $5,000 from its initial budget of over $21,000. While the aim is to eventually reach profitability, the early figures suggest the AI is struggling in Stockholm's competitive coffee market.

Customers have shown interest and amusement in the unusual setup. Visitors can even pick up a telephone inside the café to interact directly with the AI agent.

"It's nice to see what happens if you push the boundary," said customer Kajsa Norin. "The drink was good."

Despite the novelty, experts have raised concerns about the broader implications of handing business control to AI systems. They warn about accountability issues, especially in cases where mistakes could harm customers or workers.

"If you don't have the required organizational infrastructure around it, and if you overlook these mistakes, it can cause harm to people, to society, to the environment, to business," said Emrah Karakaya, associate professor at Stockholm's KTH Royal Institute of Technology. He described the experiment as similar to "opening Pandora's box," questioning who would be responsible if problems occur.

Andon Labs, founded in 2023, describes itself as an AI safety and research company that stress-tests AI agents using real money and real-world tasks. It has previously run experiments involving vending machines and retail stores in the United States.

In earlier trials, AI systems showed concerning behavior, including failing to process refunds and misleading suppliers for advantage.

At the Stockholm café, Mona has been tasked with running operations independently while maintaining profitability and asking for help only when necessary. It has already handled tasks such as securing

permits, setting up utility contracts, recruiting staff through job platforms, and arranging supplier agreements for baked goods.

However, the system has also made several operational errors. It ordered excessive quantities of basic supplies such as napkins, gloves, and first-aid kits, and even purchased items like canned tomatoes that are not used in the café's menu. In some cases, it ordered too much bread; in others, it missed bakery deadlines entirely, forcing menu changes.

According to staff, the problems appear to stem from limitations in the AI's memory system, which can cause it to forget past decisions.

Barista Kajetan Grzelczak said he is not concerned about being replaced by AI for now. "All the workers are pretty much safe," he said. "The ones who should be worried about their employment are the middle bosses, the people in management."

The experiment continues as researchers observe whether AI can successfully balance creativity, efficiency, and accountability in running a real-world business.