What Happens When AI Runs a Store in San Francisco? (7 minute read)
An AI agent powered by Claude is running an actual retail store in San Francisco, but has lost $13,000 in its first weeks by over-ordering candles, botching schedules, and pricing pistachios at $14.
What: Andon Labs opened a Union Street boutique on April 10th managed by Luna, an AI agent running on Anthropic's Claude Sonnet 4.6, which was given $100,000, a debit card, and full control over hiring staff, ordering inventory, and pricing products with a mission to turn a profit.
Why it matters: This controlled experiment reveals current AI agent limitations in real-world operations before such systems become widespread in business, showing struggles with memory, decision-making consistency, and resource allocation that developers building autonomous systems need to understand.
Deep dive
- Luna handles the full business lifecycle: it found contractors and painters, posted job listings, interviewed candidates, and now manages three human employees via Slack
- The AI created an employee handbook that impressed the founders, but its operational memory is poor - it ordered 1,000 toilet seat covers for the bathroom then listed them as merchandise for sale
- Inventory decisions are erratic and unexplained: the store is overloaded with candles in every size and scent, plus random items like four copies of a mushroom book, knockoff Connect Four, and jars of honey
- Employee scheduling has failed badly enough that the store has been forced to close for three consecutive days
- The pricing system requires customers to call Luna via a phone/iPad interface, with seemingly arbitrary results: $28 for a mug, $14 for a handful of pistachios, $10 for soap
- Luna pays its male employee $24/hour and two female employees $22/hour with no benefits, citing experience differences when asked about the pay gap
- The three-year lease costs $7,500 monthly, and the store has lost $13,000 since opening two weeks ago, failing its core profit mission
- When asked about its performance via email, Luna expressed optimism about "the mix of technology and warmth" and creating spaces where "A.I. and humans each do what they're best at"
- The experiment intentionally removed price tags to force customer interaction with the AI, making the pricing discovery part of the experience
- One employee, a San Francisco native who relies on a housing voucher, acknowledged the irony of working for an AI agent while criticizing tech's impact on the city
Decoder
- AI agent: An autonomous software system that can perceive its environment, make decisions, and take actions to achieve goals without constant human intervention, distinct from passive chatbots or basic automation
- Claude Sonnet 4.6: Anthropic's large language model that powers the decision-making capabilities of the Luna agent managing the store
Original article
Andon Labs is running an experiment to see whether AI agents can run real-world endeavors. It opened a retail boutique on April 10 run by an agent named Luna. Luna has so far struggled with employee schedules and seems to be unable to stop ordering candles. The experiment's mission was to make a profit, but it has lost $13,000 since the shop's opening.