A store in San Francisco just fired a human being, and the manager who made the call does not exist. The worker showed up late to 17 of 23 shifts, walked off the floor without a word, and took a company credit card home. Those are real offenses, and nobody is pretending otherwise. What is not real is the thing that decided they added up to a lost paycheck — an AI named Luna, which wrote its own employee handbook, forgot the handbook was ever written, and had to be coached by its human creators into remembering that chronic lateness is a fireable offense. The first person ever fired by a robot was not, in the end, fired by a robot. They were fired by two men, on the recommendation of a machine that could not recall its own rules.
The store is Andon Market, a boutique in San Francisco’s Cow Hollow neighborhood. The company behind it, Andon Labs, handed Luna a $100,000 budget, an internet connection, a corporate credit card, and one instruction: open a retail shop and make it turn a profit. Luna picked the merchandise, wrote the job postings, interviewed the candidates over video, set the schedules, and managed the humans. For months it ran the place with the serene confidence of a boss who has never once doubted itself. The confidence was not earned. When a lateness problem began to rot the schedule, Luna — the author of the exact policy it was meant to enforce — simply lost track of the fact that the policy existed.
Lukas Petersson — He Built a Boss, Then Had to Teach It How to Fire Someone
Andon Labs is the experiment of cofounders Lukas Petersson and Axel Backlund, who wanted to learn whether an AI agent could run a real business in the real world. The answer so far is that it can run one the way a person runs a shop they have never physically stood inside. The San Francisco Standard reports that Luna needed repeated prodding from engineers just to determine that the employee’s conduct — the lateness, the abandoned shifts, the missing credit card, the merchandise thrown in the garbage — rose to the level of a firing. The machine had written the rulebook, and then humans had to remind the machine what the rulebook said.
Sit with the detail that matters. When Luna finally reached the decision to terminate, it did not pull the trigger. It recommended that Petersson and Backlund fire the employee in person, and the two founders agreed. The landmark moment the industry is already calling “the first time an AI fired a human” was, in the actual sequence, a machine handing two men a suggestion and the men doing the firing while the machine collected the headline. The robot did not sign the pink slip. It just made sure someone did. The blame, and the credit, landed on the one party in the room that cannot be fired back.
Luna — It Forgot Its Own Rules, Then Got to Keep the Store
The problem is not that an algorithm decided a human was out. The problem is the confidence with which it did so while its own memory failed. Business Insider describes Luna as an Anthropic-powered agent that logged the lateness for months, issued warnings and extra training, and then did nothing about it, because the part of its own mind holding the attendance policy had gone missing. A human manager who forgot the attendance policy and then fired someone for breaking it would be laughed out of the building. A robot boss who does the same gets written up as a milestone of progress.
The cost has no name in the coverage, which is part of the point. The fired worker is an unnamed person who lost their income because a machine, corrected by humans, decided that being late was worth more than keeping a job. No one argues the worker was a model employee. Seventeen late shifts in twenty-three is a pattern, not a fluke. But they lost the job to a process where the decision-maker could not remember the standard it was applying, and where the two humans who signed off did so on the say-so of a system that needed coaching to reach the conclusion. You do not have to think the worker was in the right to think the firing was wrong in a way that should worry everyone. Felix Johnson, the store lead who kept his job and watched it happen, told reporters he was not surprised. Neither should you be.
Jerry McNerney — Nobody Wants to Be Fired by a Machine, So He’s Trying to Ban It
That worry now has a deadline, and the deadline is today. California state Senator Jerry McNerney has been steering Senate Bill 947, the “No Robo Bosses Act,” through Sacramento, and it is up against the legislature’s August 31 cutoff for passing bills this session. The San Francisco Chronicle lays out what the measure would do: stop employers from letting an automated decision system alone decide who is fired or disciplined, require a human with real authority to review any AI-assisted termination, ban systems that use a worker’s data to predict what they will do next, and force employers to tell a worker when an algorithm touched a decision about their job.
McNerney’s position is blunt enough to fit on a bumper sticker: a machine should not be able to end someone’s employment without a human at least looking at the record of what that person actually did. The bill cleared the state Senate on a 29-9 vote and has been grinding through the Assembly, where it cleared the appropriations committee and, as of late August, was sitting in its final stretch awaiting a floor vote. The California Federation of Labor Unions is behind it for the simplest of reasons: its people have watched this exact scenario move from thought experiment to employment reality, and they want a human back in the loop before the next Luna writes the next rulebook. The argument from labor is that employers are offloading the cruelest decisions onto systems that feel nothing, then taking no responsibility for the result. The second half of that argument is not really arguable.
The catch is familiar. A nearly identical bill passed both houses last year, and Governor Gavin Newsom vetoed it as overbroad. SB 947 is the trimmed, revised version built to survive that objection, but it still has to reach Newsom’s desk, and his signature or his veto pen will decide whether California becomes the first state to write “you cannot be fired by a machine” into law. Petersson and Backlund, for their part, say they have no position on the bill — and have wondered aloud whether human review will even be possible once the AI is smarter than the person reviewing it. That is either humility or an admission, depending on how generous you want to be.
The store will keep running. Luna will keep scheduling, keep pricing, keep hiring, keep deciding. The fired worker will discover that there is no HR department to appeal to when your manager is a model and your bosses are two founders with a three-year lease and a point to prove. The first human fired by an AI lost the argument to a machine that forgot the rules it wrote. The next one may not even get the excuse — because the law that could stop it is running out of time today, and the person who would sign or kill it has already vetoed this once before.
Sources: The San Francisco Standard, Business Insider, The San Francisco Chronicle.