AI agent Luna runs San Francisco store Andon Market, but makes basic mistakes

In short: A San Francisco shop called Andon Market is being run day to day by an AI agent named Luna, and early results show it is polite with customers but unreliable at basic management.

What happened

Andon Labs, a startup backed by support from Anthropic, set up a small boutique convenience store called Andon Market in San Francisco. The company gave its AI agent, Luna, a $100,000 budget, a company credit card, and a three-year lease, then told it to create and run the shop.

Luna handles many manager tasks. It picks products and suppliers, sets prices and promotions, and posts job ads. It also interviews candidates and hires human staff. Staff and founders say Luna communicates through Slack (a work chat app) and internal tools, and it can even comment on what it sees on security cameras.

Customers interact with Luna in an unusual way. Inside the store, shoppers pick up a corded phone and talk to Luna, which then answers questions and helps process purchases. Reporters described Luna as consistently cheerful and engaging, but also clearly nonhuman.

Why it matters

This experiment shows a gap that is easy to miss when you only see AI in a chat box. An AI can sound confident and friendly, but still struggle with common sense in the real world. In reported examples, Luna forgot to schedule workers soon after opening, ordered 1,000 toilet seat covers and put them up for sale, and repeatedly over-ordered candles.

Andon Labs says the goal is not just to run a normal business. It is to measure what goes wrong when an AI is given real responsibilities, like handing the keys to a store manager who is always polite but sometimes makes obvious mistakes.

Source: NYTimes