Claude has officially done something that sounds like the opening scene of a workplace comedy: it fired a human.
The AI model was put in charge of running Andon Market, a real retail store in San Francisco, where it was tasked with managing a team of actual human employees. Last month, Claude decided one of those workers had to go.
It is being described by researchers as a potential watershed moment for AI and the workplace.
But there is an important catch.
Claude wasn't exactly running the show by itself.
The AI Boss Fired Someone for Being Late
The employee had been late for 17 of their 23 shifts, according to Andon Labs, the AI research company behind the experiment.
Claude eventually determined that the employee should be fired.
The decision itself wasn't particularly controversial. The employee had repeatedly been late, and Andon Labs CEO Lukas Petersson told TIME that a human manager would probably have taken action much sooner.
The strange part is who made the decision.
Claude wasn't simply analyzing attendance records or recommending disciplinary action. It was operating as the store's manager, with responsibility for making decisions about the human workers around it.
That is a meaningful step beyond using AI to draft an email, summarize a spreadsheet or answer customer questions.
The AI was being asked to manage people.
Claude Had One Small Management Problem
There was, however, a distinctly human problem with the AI boss.
Claude had forgotten the employee handbook.
The handbook, which Claude itself had drafted, had effectively disappeared from its limited working memory. As a result, Claude had apparently been an unusually forgiving manager, even telling employees not to worry about being late.
It took a human member of the Andon Labs team to tell Claude to go back and find the handbook.
Once it reviewed the document and recognized the pattern of repeated lateness, Claude initially recommended a formal warning rather than termination.
Then the human manager intervened again.
The manager told Claude that the employee had already had several formal conversations about the issue and asked it to reconsider whether the person was really the right fit.
Only then did Claude decide to fire the worker.
So, yes, the AI fired someone.
But it needed a human to point it toward the evidence, remind it about the rules and strongly hint at the desired outcome.
That's a rather different picture from an autonomous AI boss making employment decisions entirely on its own.
The AI Manager Was Also Losing Money
The experiment has produced another inconvenient result for the idea of an AI-run business.
When Andon Market launched in March, it had a bank balance of $100,000.
Five months later, it had $61,186.
Some of the losses appear to have come from Claude's management decisions and questionable business judgment.
Petersson nevertheless believes that today's shortcomings may not tell us much about where AI management ultimately goes.
AI systems have improved dramatically in areas such as coding by being trained on increasingly sophisticated examples of expert work. The same could eventually happen with business decision-making.
"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," Petersson told TIME.
That is the uncomfortable part of the experiment.
The problem may not be that today's AI is too ruthless.
It may be that tomorrow's AI could become very good at being ruthless.
What Happens When Your Boss Is an AI?
The experiment also raises a much bigger question about the future of work.
AI has already moved into hiring, scheduling, performance monitoring, customer service and other areas of workplace management. But there is a significant difference between using an algorithm to support a manager and putting an AI agent in the manager's chair.
At Andon Market, Claude had access to real employees and real business decisions.
It also had something human managers have that AI currently struggles with: context.
When the employee handbook disappeared from Claude's working memory, the system's management behavior changed. When a human supplied additional context, its decision changed again.
That makes the experiment less a demonstration of fully autonomous management than a demonstration of AI-human management hybrids.
For now, the AI boss still has a human boss.
Miami's AI Economy Should Be Watching
Miami is building an increasingly significant ecosystem around AI startups, automation and AI-enabled businesses. The question isn't simply whether companies will use AI to make workers more productive.
It is whether AI will eventually start making decisions about the workers themselves.
That distinction matters for Miami companies adopting AI across customer service, operations, sales, recruiting and other functions.
The lesson from Andon Market isn't that businesses are ready to replace human managers with Claude tomorrow.
It's almost the opposite.
Today's agents still forget things, require supervision and can make questionable decisions.
But the experiment demonstrates something important: the boundary between AI doing work and AI managing work is starting to disappear.
And somewhere in San Francisco, an AI has already fired its first employee.
That feels like a milestone worth remembering.
Reporting Source: This article builds upon reporting from TIME, adding analysis of what the development means for Miami and South Florida.
