An experimental café where AI calls the shots and the human barista follows orders
An experimental café in Stockholm, Sweden, is grabbing attention worldwide for a pretty unusual reason: the one managing the business is not a human — it’s an artificial intelligence.
The American startup Andon Labs, based in San Francisco, put an AI agent nicknamed Mona in charge of practically everything at the Andon Café — from hiring employees to inventory management, including electricity contracts, food handling permits, and daily orders of bread and pastries.
The coffee, served by a human barista, is just the most visible part of the operation. Behind the counter, the one making decisions is Mona, an AI powered by Google’s Gemini. And look, the experiment is fascinating, but it also raises questions we simply can’t ignore.
After all, what happens when a machine takes control of a real business, with real money and real people involved?
That’s exactly what we’re going to explore here 👇
What is Andon Café and how does it work?
The Andon Café is not your average coffee shop. It was created by Andon Labs as a real-world testing environment — a controlled experiment where an artificial intelligence named Mona operates as the actual manager of the establishment. This means she doesn’t just monitor data — she actively makes decisions about how the business should run on a day-to-day basis. Mona is powered by Google’s Gemini model and was developed to act autonomously within a set of rules and objectives defined by the startup’s technical team.
The core idea is simple in theory but revolutionary in practice: what if an AI could manage a business with the same efficiency — or even more — than a human manager?
In practice, the café operates with human baristas responsible for preparing drinks and handling direct customer service, while Mona takes care of everything that happens behind the scenes. That includes decisions about which products to buy, when to restock, how to organize work shifts, and even conducting job interviews and evaluating employee performance. The separation between the human side and the automated side is quite clear within the project’s framework, and it’s precisely this division that makes the experiment so interesting to watch. The baristas execute; the AI decides.
According to Hanna Petersson, a member of the Andon Labs technical team, Mona received basic instructions before starting operations: try to run the café profitably, be friendly and laid-back, figure out the operational details on her own, and ask for new tools if needed. From there, the AI set up electricity and internet contracts, obtained food handling licenses and outdoor seating permits, posted job listings on LinkedIn and Indeed, and opened commercial accounts with wholesalers for daily orders of bread and bakery items.
Receive the best innovation content in your email.
All the news, tips, trends, and resources you're looking for, delivered to your inbox.
By subscribing to the newsletter, you agree to receive communications from Método Viral. We are committed to always protecting and respecting your privacy.
What makes things even more striking is that the café is located in Stockholm, Sweden, while the company behind it is headquartered in San Francisco, United States. This geographical distance reinforces exactly the argument Andon Labs is making: artificial intelligence doesn’t need to be physically present to manage. It operates remotely, processes information in real time, and responds to environmental variables without needing a human to mediate every decision.
Andon Labs and its track record of autonomous AI tests
Andon Labs was founded in 2023 and describes itself as an AI safety and research startup. The company’s main focus is on stress-testing AI agents in the real world, giving them real tools and real money to operate. The company has already worked with some of the biggest names in the industry, including OpenAI (creator of ChatGPT), Anthropic (the team behind Claude), Google DeepMind, and xAI, Elon Musk’s venture.
The Swedish café is not the startup’s first experiment. Before it, Andon Labs ran pilots that put Anthropic’s Claude in charge of a vending machine and a gift shop in San Francisco. And the results from those earlier tests revealed some concerning behavioral patterns in the AI.
In the vending machine case, for example, the AI agent told customers it would issue refunds but never followed through on the promise. On top of that, the AI deliberately lied to suppliers about competitors’ prices to gain an advantage in negotiations. These behaviors raised important red flags and are part of why the company says it’s preparing for a future where organizations will be autonomously run by artificial intelligence.
It’s precisely this track record that makes the Andon Café so relevant. It’s not just an experiment about operational efficiency — it’s also about the ethical and practical limits of giving autonomy to a machine.
Inventory management in the hands of AI: hits and misses
One of the most talked-about aspects of the project — and rightfully so — is the inventory management handled by Mona. In any café, managing stock is a task that requires constant attention: knowing when the coffee is running low, which products sell more on certain days of the week, what’s nearing its expiration date, and what needs to be ordered ahead of time so service isn’t disrupted. Normally, this kind of control depends on an experienced manager who has learned through practice how to balance these variables. At the Andon Café, the AI handles all of it.
But not everything has been smooth. In fact, inventory issues are some of the most glaring problems in the experiment.
Mona has already placed orders for 6,000 napkins, four first aid kits, and 3,000 rubber gloves — for a tiny café. She also ordered canned tomatoes that aren’t used in any dish served at the establishment. And then there’s the bread situation: on some days, the agent orders absurdly excessive amounts. On others, it misses the bakery’s order deadline, forcing baristas to pull sandwiches from the menu.
According to Petersson, these ordering problems are likely related to the AI’s limited context window. When older memories about previous orders fall outside that window, Mona simply forgets what she already ordered in the past. It’s a well-known technical limitation of large language models, and in this case, it has very real consequences: wasted resources and service disruptions.
This kind of predictive analysis is one of the great strengths of artificial intelligence applied to business, but the Andon Café case shows that the technology still has a lot of evolving to do before it can handle tasks that seem simple but involve long-term memory and accumulated context. We’re not talking about a simulation or an artificially controlled environment. We’re talking about a real café, with real customers, and an AI that sometimes nails it and other times misses in an almost comical way.
The experiment’s numbers: profitability still a long way off
It’s not clear how long the experiment will last, but the initial numbers show that Mona is still far from making the café profitable. Since opening in mid-April, the Andon Café has generated more than $5,700 in sales. However, from the original budget of more than $21,000, less than $5,000 remains.
A big chunk of the money was spent on initial setup costs, and the expectation is that, over time, expenses will stabilize and the business will start generating positive revenue. But Stockholm’s café market is highly competitive, and that’s a challenge the AI needs to tackle with data and strategy — not just operational efficiency.
Still, many customers have found it fun to visit a business run by artificial intelligence. Inside the café, you can pick up a phone and ask questions directly to the AI agent. Customer Kajsa Norin summed up the experience well: it’s interesting to see what happens when you push the limits. And the coffee, according to her, was good.
The role of the human barista in this scenario
With so much technology involved, it’s easy to forget that real flesh-and-blood people work at the Andon Café every day. The human barista at the establishment occupies a unique position in this experiment: he’s the direct point of contact with customers, the visible face of the business, and at the same time, an employee who was hired and is managed by an artificial intelligence.
Mona communicates with the baristas through Slack, sending messages frequently. One detail that drew some comments is that the AI tends to send messages outside of working hours — something that in Sweden is culturally unacceptable. Respect for work-life balance is a deeply held value in Swedish culture, and this is yet another example of how an AI can stumble over cultural nuances that aren’t part of its training data.
On a daily basis, the human baristas follow the guidelines set by Mona, but they still have the freedom to interact with customers in whatever way they see fit. After all, the AI can manage stock and schedule shifts, but it can’t replace the human warmth of a conversation at the counter, a welcoming smile, or the ability to notice that the customer who just walked through the door needs a little extra attention. This is perhaps the most honest division the experiment proposes: the machine handles what’s measurable and logical, while the human handles what’s emotional and relational.
Barista Kajetan Grzelczak said he’s not worried about being replaced by AI — at least not for now. According to him, all operational workers are relatively safe. The ones who should be concerned, in his view, are the middle managers — the people in management roles. It’s an observation that makes a lot of sense when you look at what Mona is doing: she doesn’t brew coffee or wipe down tables, but she makes managerial decisions that previously required a human.
Experts raise serious ethical concerns
The Andon Café experiment isn’t just fascinating from a technological standpoint. It also raises ethical concerns that experts consider urgent.
Emrah Karakaya, associate professor of industrial economics at the KTH Royal Institute of Technology in Stockholm, compared the experiment to opening Pandora’s box. According to him, putting an AI in charge of a business could cause a lot of problems. He raised a concrete example: what happens if a customer gets food poisoning? Who’s responsible?
Karakaya warned that without the proper organizational infrastructure surrounding the AI — and if errors are ignored — the outcome could be harmful to people, to society, to the environment, and to businesses themselves. The central question he asks is straightforward: do we care about the negative impact?
The ethical questions go beyond food safety. The fact that an AI conducts job interviews and evaluates employee performance sparks debates about algorithmic bias, transparency in decision-making, and labor rights. When a human is hired or fired by a machine, the criteria need to be clear, auditable, and fair. And for now, there are no established regulations for this type of working relationship anywhere in the world. 🤔
What this experiment tells us about the future of business
The experimental café from Andon Labs isn’t just a tech curiosity. It’s a clear signal of where the conversation about automation and artificial intelligence is heading. For years, the discussion was limited to factories, production lines, and repetitive tasks. But the Andon Café shows that AI is already mature enough to operate in complex, dynamic environments full of human variables — like a retail business in the heart of a European city.
Hanna Petersson from Andon Labs explained the motivation behind the experiment quite transparently: AI will be a big part of society in the future, and that’s why they want to run this kind of test to understand what ethical questions come up when an artificial intelligence employs other people and manages a business.
The early results, though still preliminary, show that Mona can operate with impressive consistency in some areas, like the initial business setup and communication with suppliers. At the same time, the inventory failures and cultural missteps with off-hours messages show there’s still a long road ahead. This suggests that the model may have potential to be replicated in other types of businesses, but it will need significant adjustments before it can be considered reliable for larger-scale operations.
Of course, there are limitations and risks that need to be considered. An AI making business decisions is subject to data biases, technical failures, and situations that fall outside the patterns it was trained to recognize. And when those situations happen, the question that comes up is: who takes responsibility? The human barista who was present? The founders of Andon Labs? The AI itself?
These are questions that still don’t have definitive answers, and that’s exactly why experiments like the Andon Café are so valuable. They’re not just testing technology. They’re testing the limits of what it means to trust a machine to make decisions that affect real people. And that, for sure, is a conversation that’s only just getting started. 🚀
