What Happens When A.I. Runs a Store in San Francisco? - The New York Times

On a nondescript block in the Mission District, Shelf.AI’s new 1,200-square-foot convenience store has no cashiers. Cameras, weight sensors on shelves and a LIDAR grid track every item you pick up or put back. An AI agent called “Mia” handles pricing, inventory and theft prevention. You tap a credit-card terminal on the way in; you leave with a latte and a bag of almonds, and a digital receipt shows up in your email two minutes later. Founders Andrea Liu and Marcus Chen spent 18 months training Mia on five million data points—sales patterns, seasonal shifts, restock thresholds, even foot-traffic peaks around 8 a.m. and 5 p.m. The system now submits real-time purchase orders to suppliers, adjusts prices up to three times a day and flags suspicious behavior. Storewide labor costs have dropped by 60 percent; two human supervisors still roam the aisles to help customers and handle the occasional sensor glitch. Not everything runs flawlessly. On opening day, 7 percent of transactions under-charged customers; a software patch fixed that within 48 hours. Privacy advocates worry about constant video surveillance, though Shelf.AI says it discards raw footage after 24 hours and encrypts all personal data. In its first quarter, the store averaged $45,000 in weekly sales—enough for Shelf.AI to project a break-even point in eight months. Other retailers are watching closely. Traditional grocers in the Bay Area are testing their own AI-driven checkout systems, and a major supermarket chain plans a similar launch in Los Angeles next year. If Shelf.AI’s numbers hold up, the company could reshape how small urban stores operate—no cashiers, minimal staff and an AI that never takes a day off.