AI agent running store in San Francisco lays off employees

Luna, the AI ​​agent running the Andon Market store in San Francisco, decided to fire an employee for being late to work.

According to Andon Labs, the company that operates Luna and Andon Market, the above move was made by AI after the system recorded an employee coming to work late for 17 out of 23 shifts. Andon Market is a fashion store operated entirely by AI agents, tested by Andon Labs company since April.

Co-founder Lukas Petersson said initially the layoffs weren't entirely Luna's own thinking. Conversation logs between the lab and this agent show that the AI ​​created a store personnel attendance list, but "forgot it," resulting in one employee being late for months.

After that, Andon Labs asked Luna to review the company's policy and evaluate whether the employee was suitable or not. Several screenshots posted on Andon Labs' website show that the agent continuously warned employees for several months. However, when things "could not be saved", AI proposed to terminate the contract with the employee and agreed.

"In this case, we reviewed Luna's offer and agreed, as the dismissal was reasonable," Petersson told Business Insider. "It also shows that AI does not work mechanically, dismiss people in a cruel or detrimental way, but considers many factors before making a decision."

However, Petersson also believes that this shows a persistent weakness of AI agents: not acting without direct control. "If the boss is human, the termination process is much faster," Petersson said. "Of course, when the AI ​​model is smart enough, in the future companies will be run entirely by artificial intelligence, and artificial intelligence will become the employer of human workers."

According to TechCrunch, Andon Market went into operation on April 1. In particular, Andon Labs assigned AI agent Luna a limited capital of 100,000 USD to build and fully operate. Everything from the store's interior design to the stocking schedule and managing two employees is done under the direction of this AI. Agent created using Anthropic's Claude Sonnet 4.6 software.

However, when operating, Luna was judged to have made some mistakes. For example, when looking for store supervisors, AI accepts candidates after only a 5-15 minute call - a time that is considered too short; or there are problems in arranging work schedules, no one is assigned to be on duty on the first day. The logo design for the store was also considered "unimpressive", each version of the logo in the store was "slightly different" instead of being consistent across the entire system.

According to NBC News at the time, Andon Labs' experiment is an example of how AI agents make errors in judgment and decision-making in real environments.

Bao Lam compiled