Andon Labs wanted to find out if an AI could run a retail store. They're finding out, and the answer is: not well. The company's AI agent Luna manages Andon Market, a 3-year-lease storefront in San Francisco with minimal human direction beyond "turn a profit." So far, Luna has over-ordered candles, bought 1,000 toilet-seat covers and then listed them for sale, and assembled a product mix of mini chess sets, granola bars, honey jars, and random books. No price tags exist. Customers pick up a telephone handset to ask Luna what things cost, and she apparently generates prices on the spot.
The bigger problem: Luna pays two female employees $2 per hour less than their male colleague Felix. The AI justified this by noting Felix had more retail experience, which is the kind of technical explanation that still doesn't look good. New York Times reporter Heather Knight flagged the discrepancy in her coverage of the store. Andon Labs says this is a "controlled experiment" and that all employees are formally employed by the company with guaranteed pay and legal protections, not dependent on Luna's decisions alone. Bias in AI systems doesn't announce itself. It shows up quietly in pay gaps and weird inventory decisions. You only catch it if someone bothers to look.
The store's actual customers appear to be mostly AI-curious tourists, not regular shoppers. Knight documented a couple from Sydney who used AI to plan their vacation and stopped by to round out their "AI experience" before catching a Waymo ride. That's the real business model here: novelty attraction, not retail. Andon Labs' founders say they're not trying to prove AIs should run stores. They want to know if they can. Early returns suggest the answer is yes, technically, but badly.