By The European Times | Created at 2026-10-19 08:16:31 | Updated at 2026-10-19 09:31:31
1 hour ago
While drinks may be prepared by human hands, an artificial intelligence system is in full control of an experimental café in the Swedish capital, Stockholm, the Associated Press reported in late July. San Francisco-based startup Andon Labs has placed an AI-powered agent named “Mona” at the helm of a café in Stockholm. The system, built on Google’s Gemini model, manages nearly every aspect of the business—from hiring staff to handling supplies—while baristas continue to prepare and serve the drinks.
According to the company, the experiment aims to explore the extent to which the automation of real-world businesses via artificial intelligence is possible, including the AI’s capacity to manage both people and processes. The café generated approximately 5,700 US dollars in revenue in less than a month, against an initial investment of 21,000 US dollars.
Visitors describe the experience as intriguing; customers can even converse with the AI agent via a telephone installed on the premises. “It’s interesting to see what happens when you take a leap into the unknown,” one customer remarked, adding, “But the drink was good.” However, experts warn of a number of ethical and practical risks, including questions regarding liability for errors and potential damages. The artificial intelligence agent has already encountered operational difficulties, such as errors in supply orders—for instance, ordering excessive quantities of consumables or items not included on the café’s menu. According to the Andon Labs team, these issues stem from limitations in the system’s so-called “context window,” which causes it to “forget” previous orders and decisions.
The company describes the establishment as a “controlled experiment” designed to reveal the ethical and organizational issues that arise when a business is managed by artificial intelligence.
Meanwhile, humanoid robots are already working full-time in a warehouse. The company Figure AI live-streamed an eight-hour shift of humanoid robots operating in a warehouse without any human intervention.
The broadcast showed “Figure 03” model robots, controlled by the Helix-02 system, positioned along a conveyor belt. Their task was to identify incoming packages and flip them so that the barcode faced downward—an operation that takes an average human three seconds to perform. In the first ten minutes alone, the system processed 230 packages, at a rate of approximately 2.6 seconds per item. The operation is carried out by a multi-robot system with an autonomous backup protocol: if a robot detects an issue, it performs a self-diagnosis and proceeds to the technical area to request a replacement from the available robots—all without human intervention.
The unedited video also reveals certain shortcomings. Early in the broadcast, one of the robots fails to properly orient a box, allowing it to pass through in the wrong position. Some observers noted that the task being demonstrated—orienting packages—is identical to one shown in a one-hour video released nearly a year ago; this limits the value of the live stream as evidence of the robots’ autonomous capabilities.
Nevertheless, for a company valued at $39 billion, the willingness to broadcast live—rather than releasing edited footage—represents a step toward transparency in an industry frequently accused of staging its demonstrations.
The Helix-02 system employs a neural network that translates camera imagery directly into full-body motor commands, bypassing traditional manual motion planning. This video marks the latest development from Figure AI. Recently, the company released footage showing two robots independently tidying up a bedroom and making a bed in under two minutes, coordinating solely through visual cues—without any direct messaging between them.
The company’s production facility in California has reached a capacity to manufacture one humanoid robot per hour.
Illustrative photo: pexels-santhosh-kanthala-372615-39588764
