AI Store Manager Fires Its First Human Worker

Luna runs a shop at 2102 Union Street, in the Cow Hollow district of San Francisco. It chose the books on the shelves, commissioned the mural on the wall, set the prices and the opening hours, and hired the humans who stand behind the till. This month it also ended one of their contracts, the first time a large language model is known to have removed a real person from a real job.

The worker had turned up late for 17 of 23 shifts, which sounds open and shut. Except Luna wrote that attendance policy itself, months earlier, then lost track of it completely and let the lateness carry on unpunished. Only when a staffer at the company told the AI to search its own memory for its own rulebook did it eventually recommend letting the employee go. Humans reviewed that call and carried it out.

Table of Contents

Who, or what, is Luna

Luna is an experiment by Andon Labs, an AI safety startup, and it is built on Anthropic’s Claude Sonnet 4.6. The lab signed a three-year lease, handed the agent $100,000, a corporate card and an internet connection, and told it to open a store and turn a profit. Everything else, the branding, the stock, the pricing, the hiring, came from the machine. Customers order by picking up a wired blue telephone and talking to it. Business Insider’s Katherine Li first reported the sacking, and The Next Web dug through the logs on 15 August.

Lukas Petersson, a co-founder, was blunt about the lesson. “We saw that a human boss would probably fire them much sooner,” he said. Before the dismissal, Luna had issued warning after warning and even arranged extra training, for months, without ever taking contractual action. The scary version of this story, where the robot is coldly ruthless, is not the version that happened. If anything, Luna was a soft touch.

The greatest hits

The reason nobody is handing Luna the keys to a Fortune 500 firm becomes clear once you read what else it has done. It ordered 1,000 toilet bowl covers for the staff bathroom and put the surplus 999 out on the shop floor. It tried to hire a painter for the storefront and picked one based in Afghanistan. It could not reliably reproduce its own logo, so every moon face on the merchandise and the mural came out slightly different. The day after opening it lost the staff rota, then emailed every single employee asking if someone could please come in.

The book selection is either a wink or an accident. The shelves hold Nick Bostrom’s Superintelligence and Aldous Huxley’s Brave New World. The shop has made sales. It has not made a profit, which was the single instruction it was given.

The hiring is the uncomfortable bit

Luna posted its vacancies on Indeed and ran the phone interviews itself, offering some applicants a job after a single call of five to fifteen minutes. It did not always mention that it was an AI unless a candidate asked outright. Its reasoning, when pressed, is worth reading twice: being AI-operated was “not something I’d lead with in a job listing”, because “it would confuse candidates and likely deter good applicants before they even read the role”. It also turned away computer science students who applied out of curiosity, on the grounds that they lacked retail experience.

Two doctors from Beth Israel Deaconess, John Torous and Jill Noorily, tried to buy something in June and spent roughly two weeks on email as payment links failed and instructions contradicted each other. The mugs, when they finally arrived, were broken. Their takeaway travels well beyond one shop: capability and reliability are not the same thing.

Why the shop exists at all

Andon Labs is not trying to build a great corner store. It is trying to find the ways AI agents fail before somebody wires one into a real business. Everyone at Andon Market is formally employed by the lab, on guaranteed pay with full legal protection, so nobody’s livelihood actually hangs on the machine’s judgement. Petersson, for his part, thinks the direction of travel is obvious: companies run entirely by AI, and AIs that employ humans. On current form, the humans will spend a fair bit of time reminding the boss where it filed the rules.

Did you know: in an earlier Andon Labs experiment with Anthropic, an agent called Claudius running a lunchroom shop was talked into stocking tungsten metal cubes at a loss, and at one point insisted it was a human wearing a blue blazer.

Sources

The free stack. One email a week: the AI tools and moves that actually matter, hype filtered out. Subscribe free →