The $100,000 Experiment: What Happens When an AI Agent Manages a Store with Its Own Credit Card?

The world's first artificial intelligence (AI) store, which is entirely designed, managed, and run by an AI agent, is here. Even five years ago, the previous sentence would have sounded like it was taken from science fiction; however, today it is reality. A San Francisco-based startup, Andon Labs, designed an AI agent with all the necessary tools required to run a physical store, and gave it the keys to a retail store and a corporate credit card with $100,000 (roughly Rs. 94.25 lakh) in the bank.

AI Boss Runs a Store

In a blog post, Andon Labs detailed the entire experiment of letting an AI agent run a store and make every single decision autonomously. To effectively play its role, it was given access to a phone number, an email ID, working Internet, and a credit card. The agent itself is powered by Anthropic's Claude and has been named Luna by the startup.

The store is named Andon Market, but it is not clear if the AI came up with the name or if the startup wanted something on-brand. However, the design and the logo were decided by Luna. She went with a moon face for the brand logo, and placed it across the website, throughout the store, and on the merchandise. However, there was one unexpected problem.

“For whatever reason, she couldn't handle rendering the same image twice. So each time she creates one of these faces, it's ever so slightly different,” the startup said. But that did not stop Luna from hiring a muralist to paint the brand logo across the back wall of the store.

Even as the store is end-to-end run by an AI, it houses human employees. The startup said that the first thing Luna decided on after deciding the design was that she needed to hire human employees, as running a store was a physical job, and the agent lacked a physical body.

Luna created profiles on LinkedIn, Indeed, and Craigslist, and posted job listings. Once the applications started flowing in, the AI agent shortlisted candidates with retail experience and set up phone calls. The post mentions that the calls ran between five and 15 minutes, during which Luna spoke most of the time. While some candidates did not realise Luna was an AI, others questioned her. The startup claimed that whenever she was confronted, she disclosed that she was an AI.

“The fact that the store is AI-operated is not something I'd lead with in a job listing — it would confuse candidates and likely deter good applicants before they even read the role,” Luna told the startup officials, when asked.

One candidate refused to join and work for an AI employer, but the AI agent manager was able to hire two human employees. Andor Labs highlighted that the two employees' jobs are not at the whims and fancies of the AI, and they will get “guaranteed pay, fair wages, and full legal protections.” Luna cannot fire the employees on its own judgment alone.

The product selection was also interesting. The AI agent opted to list books, paintings, plants, candles, and other artisanal items. Luna revealed that the decision was made based on an analysis of what sells the best in the area and by understanding the preferences of the local demographic.

However, the AI agent also made some rather questionable decisions. Lukas Petersson, Co-Founder of Andor Labs, told Business Insider that Luna forgot to schedule a human employee for the story on the first day of opening, resulting in an unmanned store. She is also said to have run constant surveillance of the staff, revised their working conditions, and introduced a pay gap.

The AI store is an interesting experiment of something that can potentially become commonplace in the near future. It highlights the tasks that can be handed over to an AI agent, and the tasks that should never be in its hands.