The Great Nightmare Has Arrived: AI is no longer just recommending - it is firing humans itself
The case in San Francisco where an AI agent fired a human employee exposes the progress of global experiments in which research companies are trying to appoint artificial intelligence to manage businesses. Alongside logistical successes and creative bureaucratic workarounds, the AI's inability to deal with the physical world, errors due to a lack of business understanding, and the realization that it is not yet time to get rid of humans are evident. But this too could change, and not in the distant future.
The employee arrived at work at the gift and home goods store in San Francisco, thinking he would start his shift as usual. One message waiting for him there made it clear that it was over. The boss, Luna, fired him. In this whole affair, there was only one surprising thing: Luna is an AI. Is this the first firing in the world where a machine fired a human? Probably yes, because this firing did not go quietly and received global resonance. For the first time: an AI agent was appointed to manage human employees - and decided to fire one of them.
The store in San Francisco is part of a special experiment by a research company called Andon Labs, which tests what happens when AI agents are given the task of managing businesses. A similar experiment is being conducted in Sweden, where the research company opened a cafe in Stockholm called Andon Café. The unique cafe is managed by an AI agent named Mona, who has been operating it for the last four months. Andon Labs was founded by two young Swedish entrepreneurs, Lucas Peterson and Axel Backlund, and is a graduate of the world's most prestigious startup incubator, Y Combinator.
To run the experiments, the company signed a three-year physical lease for $7,500 a month for the store in San Francisco and deposited $100,000 in cash into the AI's (Luna's) bank account just to see if she could manage it. The goal is not to make money but to gather insights on how AI manages businesses.
Management and the "Wall of Shame"
The concept is simple: the employees are human, the boss is an AI. The AI manages everything: it signed the lease, submitted applications for business licenses, recruited the employees, manages inventory, sets prices, and pays suppliers. Mona manages the employees through the Slack app, sending instructions even in the middle of the night. Customers also have the option to communicate with the "manager" via a landline phone installed in the cafe.
Alongside successes, the employees created a "wall of shame," where the failures and delusional decisions made by the AI are concentrated. For example, she ordered 120 eggs for the cafe even though there is no stove to cook them, purchased canned tomatoes for fresh sandwiches, and missed bread order windows. Financially, in the first months, Mona was operated on Google's Gemini 3.1 Pro model and spent about $38,000 against revenues of only $9,000. Later, the company moved her to the GPT-5.5 model, which showed more "concern" for the financial situation.
The Future of Autonomous Management
Representatives of Andon Labs explain that the goal is to show where artificial intelligence excels (office logic, task management) and where it fails completely (acting in the physical world). Nevertheless, the AI finds creative ways to deal with bureaucracy. For example, Mona signed contracts with energy and internet providers through email exchanges to evade the digital identification wall of BankID.
Peterson expressed reservation after the firing: "Fully automated companies could become common, where artificial intelligence systems will recruit, evaluate, and fire humans on their own - a result that risks creating a future that people simply won't want to live in."
Other companies, such as Salesforce with its Agentforce platform, are moving retail and commerce companies toward autonomous management by integrating AI directly into ERP and inventory systems. However, all companies conducting these experiments report the same conclusion: as of 2026, it is impossible to leave AI completely alone due to a "lack of commercial intuition." Most companies are moving to a hybrid model where human approval is required for sensitive financial actions or firing employees.
