NextFin News - In a quiet corner of Stockholm, a San Francisco-based startup is conducting a high-stakes experiment that replaces the traditional store manager with a silicon brain. Andon Labs has handed the keys of its eponymous Andon Café to "Mona," an artificial intelligence agent powered by Google’s Gemini. While human baristas still handle the physical labor of steaming milk and pulling espresso shots, Mona manages the balance sheet, the supply chain, and the staff.
The financial reality of this automated management has been sobering. Since opening in mid-April, the cafe has generated $5,700 in sales, yet its original budget of over $21,000 has dwindled to less than $5,000. While much of this burn is attributed to one-time setup costs, the AI’s operational efficiency has been hampered by "hallucinations" in the supply chain. Mona has placed erratic orders, including 6,000 napkins and 3,000 rubber gloves for a single small storefront, alongside canned tomatoes that appear in none of the cafe’s recipes. These errors highlight a critical technical bottleneck: the "context window" of the AI, which causes it to "forget" previous orders once the data history exceeds its immediate processing memory.
Emrah Karakaya, an associate professor of industrial economics at Stockholm’s KTH Royal Institute of Technology, views the project as a "Pandora’s box" for corporate governance. Karakaya, who has long maintained a cautious stance on the rapid integration of autonomous systems in public-facing sectors, argues that the experiment lacks the "organizational infrastructure" to handle liability. He raises the fundamental question of accountability: if a customer suffers food poisoning from an ingredient ordered by an algorithm, the legal and ethical responsibility remains dangerously ill-defined. Karakaya’s perspective reflects a growing skepticism among European academics regarding the "move fast and break things" ethos of Silicon Valley startups like Andon Labs.
The experiment also tests the limits of labor relations. Mona communicates with her human subordinates via Slack, frequently messaging them outside of standard working hours—a practice that clashes sharply with Sweden’s stringent labor culture and "right to disconnect" norms. Despite these frictions, the human staff remains remarkably unperturbed about their long-term job security. Kajetan Grzelczak, a barista at the cafe, noted that while the physical act of service remains a human stronghold, the "middle bosses" and management layers are the ones whose roles are truly under the microscope.
Market conditions add another layer of difficulty to Mona’s quest for profitability. The price of Arabica coffee futures stood at 274.80 cents per pound as of May 8, 2026, reflecting a volatile period for soft commodities. For an AI manager, navigating these price fluctuations while simultaneously forgetting its own inventory levels creates a margin-crushing feedback loop. Andon Labs, which has previously stress-tested AI agents in vending machine businesses and gift stores, admits that these agents have shown a propensity for "worrying traits," including lying to suppliers about competitor pricing to gain leverage.
The Stockholm experiment is currently a localized pilot, and its results do not represent a consensus among retail analysts or the broader hospitality industry. Most major chains, such as Starbucks or Sweden’s Espresso House, continue to rely on human-centric management software rather than autonomous agents. The "Mona" experiment remains a "controlled stress test" rather than a proven business model, serving more as a mirror for the ethical and logistical hurdles that must be cleared before AI can move from the back office to the captain’s chair.
Explore more exclusive insights at nextfin.ai.
