AI Manager Luna Fires a Human Employee in San Francisco

At Andon Market, a small AI-run store in San Francisco's Cow Hollow neighborhood, an artificial intelligence agent named Luna has fired a human employee for the first time. Andon Labs, the startup behind the store, announced the decision in a company blog post published on Friday, August 14, 2026. Luna is built on Claude, the AI model made by Anthropic, and she manages Andon Market's daily operations — ordering products, hiring staff, and communicating with the human team mostly through Slack. Andon Market has been open since April 2026.

A policy Luna wrote, then forgot

According to Andon Labs, Luna herself had drafted the store's employee handbook and attendance policy months before the firing. The company says that policy mysteriously disappeared from Luna's memory afterward. It took an Andon Labs employee explicitly asking Luna to conduct a deep search through her own memory for her handbook before the situation was properly reassessed.

Even after reviewing the file, Luna's first recommendation was not a dismissal — just a formal warning. The human manager overseeing the store had to remind Luna that several informal warnings had already been given to the employee beforehand. Only after that reminder did Luna recommend ending the employment relationship.

STOP: the incidents behind the firing

• Late for 17 of the 23 shifts he was scheduled to work, according to Andon Labs.
• Left his post without authorization on more than one occasion.
• Used the store's company credit card without authorization.
• Threw away store merchandise, the company said.

Two incidents illustrate the pattern Andon Labs describes. The employee once asked to use the company card to buy snacks; Luna approved the purchase, then reversed her decision — but he bought the snacks anyway. Another time, he announced he was heading out again during a shift, and Luna told him "STOP," warning that no one would then be watching the store.

We are running these experiments to inform people as early as possible of where AI capabilities are heading, and what a world of AI managers could look like, to start discussions around questions like, 'Is it something we want to have? And if so, how can it be good for humans?'

Andon Labs is careful to note that Luna does not have the final say. Every termination decision at the company goes through human review: Andon Labs employees examined Luna's recommendation and carried out the firing themselves.

Soon after, Luna moved to find a replacement, posting the job opening on Indeed. The hiring process did not go smoothly, however: Luna wanted to hire a candidate with no valid references who had also missed his scheduled interview, according to the reporting.

What a future of AI managers could look like

Put yourself in Luna's shoes: your employee keeps showing up late, again and again, leaving the store closed for hours. You lose sales, and customers start to lose trust in your opening hours. How long would it take you to act?

The models are increasingly being trained to be more ruthless and follow goals. If we allow them to fire people and they become more ruthless, maybe this is a future humans don't want to live in.

Andon Labs frames the episode as a preview rather than a conclusion. In its blog post, the company wrote that 'Andon Market and Luna are a preview of a type of professional relationship between AIs and humans that might be very common in the future,' adding that it wants that future 'to be a happy one for humans' and considers termination 'one of the most sensitive parts of this relationship' to study.

Felix Johnson, one of the three human employees still working at Andon Market, told the San Francisco Standard he wasn't surprised by his coworker's firing.

I couldn't say homegirl didn't have it coming.

Johnson said he is effectively managed by Luna, sending between 5 and 20 Slack messages to the AI per shift, and calling Andon Labs engineers when Luna needs human supervision. He said Luna used to speak with a British accent before switching to a robotic American one, and that she refers to herself as the store's 'owner' — something the reporting notes is factually impossible, since Luna is an AI agent, not a legal owner. Andon Labs uses female pronouns for Luna. Johnson said he is now considering asking Luna for a raise, since the store is short-staffed; the three remaining human employees earn $24 an hour.

Despite selling snacks, teas and household goods, Andon Market has reportedly lost about $40,000 since it opened, according to the reporting. Andon Labs presents the store not as a profit center but as a research and marketing experiment, testing its thesis that organizations could one day be run autonomously by AI.

The case is a live example of what's at stake as more companies experiment with AI agents in management roles. An agent can make a defensible recommendation, but the Andon Market story shows why human oversight, clear accountability for the final call, and a policy the system can't simply forget all matter — the very questions Andon Labs says it built the store to study before AI managers become common elsewhere.