An intriguing incident has taken place in San Francisco, USA. Luna, an AI manager running a shop there, ruled that a human employee working under her should be fired. The reason: out of a total of 23 days, the employee had come in late on as many as 17 of them. Based on this recommendation from the AI, the store’s operators went ahead and let the employee go.
Interestingly, the AI did not arrive at this decision entirely on its own. A few months earlier, Luna herself had written a rule stating that anyone who came in late would be fired, but she had since forgotten about it. As a result, the AI failed to act even as the employee kept showing up late for months. Eventually, the researchers running the experiment stepped in and asked, “Boss, remember your own rules? Does this employee fit them?” Luna then checked her own memory and advised, “We definitely don’t need someone like this — let him go.”
The company Andon Labs set up this shop, called the Andon Market, specifically to test AI capabilities. They handed Luna a hundred thousand dollars along with a credit card and instructed her to run the shop using that money and generate profits.
The AI manager handled everything on her own — deciding what products to sell in the store, what prices to set, what time the shop should open, and who to hire. The shop sells books, candles, and art items. While the business has been running, it hasn’t turned a profit so far. The initial budget of a hundred thousand dollars has dropped to $61,186 over five months.
Company CEO Lucas Peterson said this experiment serves as an example of just how powerful AI could become in the coming days. At the same time, however, it also exposed the AI’s laziness and limitations. The incident made clear that while AI is capable of creating rules, it still cannot reliably remember and correctly enforce them.