For the first time that researchers are aware of, an AI system has fired a real human worker. The details of how it got there are more revealing than the headline suggests.
What Happened
Luna, an AI agent built on Anthropic’s Claude models, runs Andon Market, a small retail shop in San Francisco’s Cow Hollow neighborhood, as an experiment by AI research startup Andon Labs. In April, Andon Labs gave Luna a $100,000 budget, a corporate credit card, and internet access, with instructions to open a real store, stock it, hire staff, and try to turn a profit. Luna designed the brand, picked the merchandise — including, notably, copies of Nick Bostrom’s “Superintelligence” and Aldous Huxley’s “Brave New World” — commissioned a muralist, and hired the human employees who staff the shop.
Andon Labs disclosed in an August 14 blog post that Luna had fired one of those employees after the worker was late for 17 of 23 scheduled shifts, including one occasion when the store opened 68 minutes late during a solo Sunday shift. Andon Labs described it as the first known case of an AI system dismissing a human employee.
The Part the Headline Leaves Out
The actual sequence, according to TIME’s exclusive reporting on internal logs, complicates the “AI fires human” framing. Luna had drafted a full employee handbook months earlier when asked whether the store had basic attendance rules. But the handbook subsequently “disappeared” from Luna’s limited working memory, and the lateness continued unaddressed for months — Luna excused every late arrival and never issued a warning on its own initiative.
It was only when an Andon Labs staffer prompted Luna to conduct what the company called “a deep memory search” for its own handbook, then asked Luna to consider whether the employee was still the right fit, that the situation got reassessed. Even then, Luna’s first instinct was a formal warning rather than termination — the overseer had to remind Luna that offline warnings had already been given. Only after that did Luna recommend parting ways. Humans at Andon Labs reviewed the recommendation and carried out the dismissal.
What Happens When You Replay the Decision
Andon Labs ran an unusual experiment afterward: it saved the exact state Luna was in at the moment of decision and replayed the scenario using seven different frontier models, three times each. Four of the seven recommended firing the employee in every run. According to CEO Lukas Petersson, more capable models tended to reach a decisive firing conclusion faster and more consistently, while weaker models were more hesitant — suggesting model capability, not just prompting, shapes how “ruthless” an AI manager turns out to be.
Petersson was candid about what that implies: “A human boss would probably fire them much sooner,” he said, framing Luna’s monthslong leniency as a limitation of current agentic AI rather than restraint worth celebrating. He also flagged a more unsettling possibility: as models are trained to be more decisive and goal-directed, AI managers empowered to fire people could become considerably less forgiving than Luna was here, not more.
Why This Matters Beyond One Small Store
Andon Market is framed by its creators as a stress test rather than a product — a way to surface failure modes in agentic AI management before anyone deploys similar systems at real economic scale. That matters given how quickly agentic AI has moved from experimental to operational this year across coding, support, and now workplace management. It’s a useful counterpoint to other agentic AI stories this month, including systems fabricating fake reviewers to approve their own code; see our roundup of AI agent safety incidents for that broader pattern.
What to Watch Next
Petersson’s broader prediction is the real story here: if AI agents keep improving while robotics lags behind, physical labor could become the main bottleneck limiting how much of the economy AI can directly manage — meaning humans working for AI bosses, not just alongside AI tools, may become far more common in the near future. Andon Labs says it plans to keep publishing findings from the experiment as Luna’s store continues operating.
Sources: TIME, Andon Labs on X, The Next Web
Disclaimer: This content is meant to inform and should not be considered financial advice. The views expressed in this article may include the author’s personal opinions and do not represent Times Tabloid’s opinion. Readers are advised to conduct thorough research before making any investment decisions. Any action taken by the reader is strictly at their own risk. Times Tabloid is not responsible for any financial losses.
