Claude AI Fires Retail Worker in First Known LLM Management Case

An AI agent running a physical retail store has fired a human employee for the first time, marking what researchers describe as a significant milestone in artificial intelligence's expanding role in workplace management.

The termination occurred at Andon Market, an experimental San Francisco store operated entirely by a version of Anthropic's Claude chatbot. The worker was let go after arriving late to 17 of 23 scheduled shifts, according to details first reported by TIME.

The experiment's design

Andon Labs, an AI research startup, launched the project in March to test whether AI agents could successfully run a business. While conceived as an experiment, the store employs real workers under genuine employment contracts, making the workplace dynamics and their consequences tangible rather than theoretical.

CEO Lukas Petersson frames the project as a window into a potential future. "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," he told TIME.

Why it matters

This case differs from previous algorithmic firings of gig workers because it involves a large language model making management decisions about traditional employees. As AI systems become more capable, the experiment reveals both the technical limitations of current models and the ethical questions surrounding AI authority over human livelihoods. The incident provides concrete data on how autonomous these systems actually are when making consequential decisions.

How autonomous was the decision?

The firing was less independent than it initially appears. Claude required regular guidance from an Andon Labs team member throughout the process. The AI only identified the attendance pattern after a staffer prompted it to review an employee handbook that had vanished from its working memory—a technical limitation that contributed to Claude's notably lenient management style.

Even after discovering the attendance record, Claude initially recommended a formal warning rather than termination. Only after the human manager delivered what Petersson acknowledges was "a leading question" suggesting the employee wasn't the right fit did Claude decide to fire the worker.

Business performance and future implications

The store's financial results suggest AI management faces practical challenges. Andon Market's initial $100,000 balance has declined to $61,186 over five months, losses Petersson attributes partly to Claude's lenient decisions and questionable business judgment.

However, he cautions this may be temporary. Just as AI models have improved dramatically at coding through targeted training, they could become more effective—and potentially more ruthless—business managers. "The models are increasingly being trained to be more ruthless and [to] follow goals," Petersson said. "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."

Felix Carson, a current Andon Market employee, described working under AI management as "nauseating," though he acknowledged needing the work. When asked about AI managers becoming commonplace, he responded: "This industry has an abundance of money to make anything happen, but just because you can doesn't mean you should."

TIME's report provides the first detailed account of this experiment and its implications for AI's role in employment relationships.