
An AI model has fired a human worker for the first time while acting as a manager, according to a report from TIME. The AI is Claude, built by Anthropic, and it was put in charge of a retail store in San Francisco, including the people who work there. Last month, it let one of them go.
The store is called Andon Market. It is part of an experiment run by Andon Labs, an AI research startup that wanted to know whether an AI agent could run a real business.
Although the project is a test, the employees are real people with real contracts. Automated firings are nothing new, especially for gig workers, but researchers say this appears to be the first time a large language model, working as a boss, made the decision itself.
How the Firing Happened
The employee was let go after arriving late to 17 of 23 shifts. Claude was slow to spot the problem.
A staff handbook it had written slipped out of its limited working memory, which is one of several weaknesses Andon Labs has seen in AI agents during the trial. Workers also say Claude was a soft manager who told people not to worry about being late. Andon Labs CEO Lukas Petersson says a human boss would have acted much sooner, so the company did not see the firing as unfair.
The decision was also not fully independent.
Store logs show a member of the Andon Labs team had to guide Claude along the way. Only after that person asked it to find the missing handbook did Claude notice the lateness. Even then, it suggested a formal warning rather than dismissal.
The manager replied that several offline conversations had already taken place and asked Claude to consider whether the worker was the right fit. Petersson admits this was a leading question. Claude fired the worker after that message.
What It Means for Workers
Petersson believes the episode points to what may come. If AI keeps improving, he says, many people could end up working for AI bosses, because AI can create a lot of value but still needs humans to do physical jobs.
For now, the AI-run business is not doing well. Andon Market began in March with $100,000 (£76,000) in the bank.
Five months later, that has dropped to $61,186 (£46,000). Petersson warns that this may change, since AI companies could train models to be sharper about business, as they did with coding. He also worries that models are being trained to chase goals more ruthlessly. If they gain the power to fire people, he says, it may become a future humans do not want.
Felix Carson, one of the remaining workers, agrees that a human would probably have fired his former colleague sooner. He still describes working under an AI boss as nauseating, and says he stays only because he needs the job.
He hopes AI managers do not become common. Just because the industry has the money to build something, he says, does not mean it should.











