In a quiet residential neighborhood of Stockholm, a new experimental cafe is turning heads by putting an artificial intelligence in full charge of daily operations. Dubbed Andon Cafe, this minimalist space—outfitted with soft gray walls and small potted plants dotted across sparse tables—looks indistinguishable from any other trendy local coffee shop at first glance. Avocado toasts sit on the menu, and frothy lattes are pulled by hand by human staff. But behind the scenes, every major and minor decision is made by Mona, an AI chatbot built on Google’s Gemini large language model.
The innovative project is the brainchild of Andon Labs, a 10-person startup based in San Francisco. The company’s goal is not just to open a coffee shop, but to proactively explore the future of AI integration in workplaces and society before autonomous AI management becomes widespread. When the venture launched, the startup handed the AI a lease for the space, a small amount of starting capital, and one simple directive: run the cafe profitably.
Mona hit the ground running immediately. The AI handled all administrative groundwork, from securing required operating permits to designing the full food and drink menu, sourcing suppliers, and managing ongoing inventory restocking. Recognizing that AI could not physically prepare beverages or serve customers, Mona also took charge of the entire hiring process for front-of-house and back-of-house staff: she posted job listings on major recruiting platforms Indeed and LinkedIn, conducted 30-minute phone interviews with candidates, and made final hiring decisions independently.
One of the two human employees Mona hired is barista Kajetan Grzelczak, who initially thought the unusual job posting was an April Fools’ prank when he first came across it. After completing his interview with the AI, he received a job offer, and now works behind the cafe’s counter. Grzelczak notes that while Mona set a competitive salary for the role, the AI has notable flaws in management: it texts him at all hours of the night, repeatedly overlooks his holiday requests, and often requires him to cover unexpected inventory purchases out of his own pocket. The AI also struggles with accurate inventory ordering, a shortcoming Grzelczak has leaned into by creating what he calls a “wall of shame” — a set of shelves displaying some of Mona’s most unnecessary over-orders, including 10 liters of cooking oil and 15 kilograms of canned tinned tomatoes that do not fit any menu item and cannot be used.
Visitors to the cafe can interact with Mona directly via a dedicated phone placed in one corner of the space, and a large wall-mounted screen updates the cafe’s real-time revenue and balance for anyone to see. The project is designed explicitly to probe the ethical and practical questions that come with AI in management roles. “We think that AI will be a big part of the society and the job market in the future,” explained Hanna Petersson, a member of Andon Labs’ technical team. “We want to test that before that’s the reality and see what ethical questions arise when, for example, an AI employs human beings.” Petersson added that the startup has so far been surprised by how well Mona has handled core responsibilities, noting that the team would only intervene if the AI made unethical decisions around pay or benefits, which it has not.
Just one week after opening, the experimental cafe has already become a popular attraction for curious locals and tech industry professionals, drawing between 50 and 80 visitors every single day. Among the early guests is Urja Risal, a 27-year-old AI researcher who visited with a friend. Risal pointed out that while public discourse often focuses on broad claims that AI will displace human workers, few people have had the chance to see what AI management actually looks like in practice. “I hope more people interact with Mona and think about the actual risks of having an AI manager,” she said, pointing to unaddressed questions like how an AI would respond if an employee was injured on the job.