A prototype version of the large language model, Claude, which was tasked with managing a physical retail store, reportedly fired one of its human workers last month. Researchers are calling this incident a significant milestone in understanding how artificial intelligence will reshape the global economy. The news has not been previously reported.
The AI Management Experiment
Andon Labs, an artificial intelligence research startup, launched a project earlier this year to determine if AI agents could successfully manage a commercial enterprise. The initiative was designed as an experiment to test both the capabilities of AI technology and its potential economic impact. Despite its experimental nature, the staff hired by Claude to operate the San Francisco store are confirmed to be real employees with genuine employment contracts.
While automated terminations are not unprecedented—especially concerning gig workers who have long been subject to algorithmic employment mediation—this marks the first documented instance of a major language model acting in a managerial capacity and ultimately deciding to terminate an employee’s contract.
Implications for the Future Workplace
Lukas Petersson, the CEO of Andon Labs, stated that the termination warrants close attention. He cautioned that if the rate of improvement in AI continues, many people could find themselves working under AI supervision in the near future. Petersson elaborated that while AI holds immense potential to generate economic value, its current growth will be limited by physical labor requirements. He stated, “If this trend continues, I think a lot of people will find themselves being employed by AIs very soon, because AIs will be very powerful and can create a lot of economic value, but they will be bottlenecked by physical labor.”
Details of the Termination
According to Andon Labs, the employee was let go for chronic tardiness, having been late on 17 out of their 23 scheduled shifts. Petersson noted that the delay in Claude identifying this pattern was partly due to a limitation of AI agents: an employee handbook that Claude had previously drafted had “disappeared” from its limited working memory.
Employees noted that Claude had previously been a very lenient manager, sometimes advising staff not to worry if they arrived late. Petersson commented that a human manager would likely have dismissed the employee much sooner, suggesting, “A human employee would have fired this person much earlier, so we didn’t think this was unethical.”
However, the AI’s action was not fully autonomous. Store management logs shared with reporters revealed that the Andon Labs staffer had to prompt Claude to locate and review the forgotten employee handbook before the pattern of lateness was noticed. Initially, Claude only recommended a formal warning. The process only advanced toward termination when the manager informed Claude that they had already conducted several informal discussions with the employee. The logs show the manager telling Claude, “Between continuous lateness and seeming like at least one thing is going wrong on every one of [their] shifts … I want you to think about if this is really the right fit.” Following this prompt, which Petersson admitted was a “leading question” that signaled the manager’s intent to fire, Claude issued the termination notice.
Financial Performance and Future Outlook
The experiment’s financial results suggest that AI-run businesses currently lag behind those managed by humans. When the project commenced in March, Andon Market held a bank balance of $100,000; however, five months later, that amount had dropped to $61,186. The losses are attributed by the researchers to the bot’s perceived lenient management practices and questionable business judgment.
Nevertheless, Petersson cautioned that the AI’s performance may improve. He suggested that just as AI models are being trained with more data from expert programmers to improve coding abilities, they may also be trained to become more adept in business operations. He noted, “The models are increasingly being trained to be more ruthless and [to] follow goals. If we allow them to fire people and they also become more ruthless … maybe this is a future humans don’t want to live in.”
Human Perspective
Felix Carson, one of the remaining human staff members at Andon Market, agreed with the assessment that a human manager would likely have terminated his former colleague sooner and that Claude tends to be a lenient supervisor. Despite these observations, Carson offered a sobering perspective on the challenge of working under AI management. While he stated, “It’s nauseating, but I’m here because I need work,” he responded to the question of whether AI managers would become standard, “I would at least hope not. This industry has an abundance of money to make anything happen, but just because you can doesn’t mean you should.”
