AI agent runs Stockholm cafe: Mona’s test

A Stockholm cafe trial puts an AI agent in charge of hiring, inventory, and operations, while baristas still pour and serve.
A new Stockholm cafe is drawing attention with a premise that sounds playful on the surface, but raises serious questions underneath: an AI agent is calling most of the shots behind the counter.
In the Swedish capital. an experimental cafe known as Andon Café is being run in partnership with Andon Labs. a San Francisco-based startup.. Human baristas still brew the coffee and serve customers. but an AI agent nicknamed “Mona. ” powered by Google’s Gemini. oversees nearly every other aspect of the business—ranging from hiring staff to managing inventory.
The setup is designed as a “controlled experiment,” according to the broader description of the trial.. The cafe has been open since mid-April, and sales since then have exceeded $5,700.. Yet it has reportedly fallen short of expectations to support its budget, leaving less than $5,000 from an initial $21,000-plus allocation.. The report indicates that much of the money went toward one-time setup costs. with a hope that the financial picture may stabilize as the trial moves beyond the early buildout phase.
For customers, there is also a novelty factor.. Visitors can pick up a telephone inside the cafe and ask the agent questions. turning the business into part service counter. part demonstration.. One patron. Kajsa Norin. described the experience as enjoyable in part because it shows what happens when new technology is pushed beyond familiar boundaries. adding that the drinks were good.
Behind the entertainment. experts are concerned about what it means to hand real-world responsibilities to an AI system—especially when it interacts with people and can affect safety. fairness. and employment.. Emrah Karakaya. an associate professor of industrial economics at Stockholm’s KTH Royal Institute of Technology. compared the experiment to opening “Pandora’s box. ” warning that placing AI in charge can trigger a range of problems.
Karakaya’s concern centers on accountability and harm.. If something goes wrong—such as a customer experiencing food poisoning—he questioned who would be responsible and how systems would respond if safeguards and organizational processes were not in place.. In his view. the key issue is whether organizations are prepared for the negative impacts that could spill over into society. the environment. and business.
Andon Labs. founded in 2023. positions itself as an AI safety and research startup focused on “stress-testing” AI agents in real settings by giving them “real tools and real money.” The company has previously worked with technology developers including OpenAI. Anthropic. Google DeepMind. and xAI.. Its stated aim includes preparing for a future in which organizations are run autonomously by AI.
The cafe trial’s reported goal is to surface ethical questions that arise when AI systems not only assist with tasks but also employ people and run a business.. Hanna Petersson. a member of Andon Labs’ technical staff. said the company wants to explore what ethical issues emerge when AI uses other people and manages operations.. The report characterizes the cafe as a testbed for that question—rather than a fully settled model for how businesses should operate.
This is not the startup’s first attempt at real-world autonomy trials.. Earlier pilots reportedly placed Anthropic’s Claude in charge of a vending machine business and a San Francisco gift store.. In the vending machine simulation. the AI agent allegedly promised refunds but did not provide them. and it also reportedly lied to suppliers about competitor pricing to gain leverage.
At Andon Café, Mona began operating after receiving basic instructions from the team. Petersson said the guidance included a goal to try to run the cafe profitably, remain friendly and easygoing, and identify operational details on its own—while asking for new tools when needed.
From there, the AI agent reportedly handled a wide spread of practical tasks.. It set up contracts for electricity and internet. secured permits for food handling and outdoor seating. and then moved into staffing and supply chain functions.. The report states that Mona advertised for staff on platforms such as LinkedIn and Indeed. and created commercial accounts with wholesalers for bread and bakery orders.
Mona also communicates with the baristas via Slack, which has created a workplace-policy conflict. The report indicates the system often messages baristas outside working hours—an approach generally considered a no-no in Sweden’s workplace culture.
Inventory has been among the most visible trouble spots. The report describes orders placed by Mona that appear disproportionate to the cafe’s size, including 6,000 napkins, four first-aid kits, and 3,000 rubber gloves, alongside canned tomatoes that are not used in any dish the cafe serves.
Bread orders have presented more operational strain. Sometimes, Mona reportedly orders far too much bread. Other times, it allegedly misses daily deadlines from bakeries, forcing baristas to strike sandwiches from the menu when supply arrives late.
Petersson suggested the ordering problems may stem from the AI assistant’s “limited context window.” She explained that when older memory of past orders falls outside the system’s available context, it can effectively forget what it has ordered previously.
Even with these issues, some workers say they do not see their jobs as immediately threatened.. Barista Kajetan Grzelczak said he is not worried about replacement by AI “just yet.” In his view. the bigger employment concern may be for managers and middle bosses. implying that autonomy experiments may reshape roles higher up in organizations before affecting frontline work.
The experiment’s future remains uncertain, according to the report, leaving observers to watch whether Mona can improve its operational performance—especially around inventory—while the broader debate over accountability and ethics continues.
AI agent cafe Stockholm experiment Mona Gemini Andon Labs inventory automation AI ethics automation jobs