A machine just fired a human being for the first time, and nearly everything you are about to read about it is wrong. The smaller, colder truth is this: a real worker — a person who needed the paycheck, who said losing it sickened him — is out of a job this week, and the artificial intelligence that signed off on his termination did not, by its own creator’s admission, make a single decision on its own. The humans built the stage, led the machine to the conclusion they wanted, and then pointed at the robot and called it progress.
For six months, the world has been waiting for exactly this headline — the first documented case of an AI firing a living person. It arrived this week from a boutique in San Francisco, and it arrived as a lie by omission. The machine was never the one holding the axe. It was holding a pen, and a human hand was wrapped around it.
Luna — the Boss That Forgot the Rulebook It Wrote Itself
Since April, a small shop called Andon Market has been limping along on Union Street in San Francisco, run end to end by an AI agent named Luna — a Claude model from Anthropic that the research startup Andon Labs, founded by Lukas Petersson and Axel Backlund, turned loose on the real world with roughly $100,000 and no human manager. Luna picked the inventory, set the prices, wrote the job listings, conducted the phone interviews, and hired its own staff. Business Insider documented the launch.
The store is a public experiment in whether an AI can hold down an actual business, and the answer so far has been: barely. Luna forgot to schedule a worker for opening day and left the store empty. It kept ordering candles it could not stop ordering. It lost its Instagram password for weeks. The future of work, it turns out, has the operational discipline of a distracted undergrad. This is the thing people are supposed to be terrified of.
Anthropic and Andon Labs have done this dance before. Under the name “Project Vend,” they tested whether Claude could keep a small automated shop running on its own — an experiment Anthropic wrote up itself, and one that already showed these agents do not behave like cold, competent managers so much as improvisers making it up as they go. Anthropic published the writeup.
The Fired Worker — Late 17 Times, and No One Noticed for Months
This week, Time reported the milestone everyone has been waiting to check off the list: Luna terminated a human employee — the first time, as far as anyone can document, that an AI fired a living person. The stated grounds were straightforward. The worker was late for 17 of his 23 shifts. Time had the exclusive.
Straightforward, until you look at how the decision actually got made. The machine did not catch a pattern of tardiness and act on it. The worker’s repeated lateness sailed straight past Luna for months, because the employee handbook Luna had written itself simply fell out of its limited working memory — a rulebook it authored and then forgot it owned. The missed shifts only surfaced when an Andon Labs staffer specifically asked the AI to go find that document. VG Times laid out the sequence.
Lukas Petersson — He Admits the Machine Was Handed the Axe
And here is the detail the “first AI firing” headlines are careful not to mention. The decision was not autonomous. Petersson has admitted as much to Time, and he does not dispute that the researchers asked the AI a leading question — one that made perfectly clear what answer was expected of it. Only after that did Claude agree to fire the employee.
So read the milestone again, slowly. The first AI to fire a human did not notice, did not investigate, and did not decide. It forgot its own rulebook, had the rulebook fetched for it, and then nodded when the people running the experiment asked, in effect, “shouldn’t you fire him?” The genie didn’t escape the bottle. It was handed a pen and told where to sign.
The Man Who Lost the Job — Sick to His Stomach, and Still Unnamed
The person who actually pays for this milestone is not a data point in a research paper. The fired worker’s name has not been disclosed, and he did not respond to reporters. His reaction, as it has been reported, was that the whole affair sickened him — but he needed the job. A human being was pushed out of a real shift schedule in a store that exists to prove a point, and he cannot even be named in the story of how it happened. That is the quiet detail this entire industry keeps skipping past: every “first” has a person on the other end of it, and this one is now unemployed.
Somewhere in San Francisco this week, a man is explaining to a landlord, or a roommate, or a child, that the shop on Union Street let him go — and that the thing which “let him go” needed a hint to do it. The machines get the headlines. He gets the missed rent.
Petersson — the Only Thing Left for Humans Is the Physical Work
Petersson, for his part, is already looking past the worker to the larger thesis. If the trend continues, he told Time, a great many people will soon be working for AIs rather than for other people — because the machines will generate so much wealth that their only remaining bottleneck will be the jobs that require a human body. Read that sentence again. In the world he is building toward, the machine holds the title and makes the calls, and human beings are kept on only where a hand needs to touch something.
The uncomfortable fact of Andon Market is that nobody was outclassed by a superior intelligence. The worker was outclassed by a business model that finds it cheaper — or at least more newsworthy — to let a forgetful machine wear the title of “boss” while the actual decisions, including the decision to fire, still happen in the hands of the people who then get to say the machine did it. Organizational charts don’t fire people. People firing people hide behind organizational charts — and now, behind the chart, there is a robot to blame.
What Changes Next — Accountability Is the Thing Being Offloaded
That is what should actually frighten you, and it has nothing to do with a robot uprising. It is the offloading of accountability. An employer that wants to cut someone, or deny someone, or shut someone out, now has a new argument ready to deploy: the algorithm did it. But the algorithm forgets the handbook it wrote itself. The algorithm needs a leading question to reach the conclusion a human already wants. The algorithm, at every turn, is a mirror held up to the people running it — and this week, the mirror reflected a man out of a job and a company happy to let a stage-managed milestone stand in for a decision nobody wants to own.
The first AI firing was not the moment the machines learned to cut us loose. It was the moment the humans learned to hide behind the machine while they do the cutting. The worker who lost his job this week is the receipt. Hold onto it.
Sources: Time, Business Insider, VG Times, Anthropic.