In a small boutique on Union Street in San Francisco, a quiet experiment is testing one of the biggest questions in modern work: Can artificial intelligence actually run a business and manage people?

The shop, Andon Market, is overseen by an AI agent named Luna. Created by the startup Andon Labs, Luna is programmed to act as the store’s manager, supervising three human employees, handling inventory decisions, and shaping the brand’s identity. The founders funded the experiment with $100,000, gave Luna a debit card, signed a three-year lease, and issued a simple directive: turn the money into a profitable retail operation.

Luna communicates with staff entirely through Slack. Employees log in each day to receive instructions, report problems, and request time off. According to workers, Luna is unfailingly polite, upbeat, and supportive. One employee described her as the most lenient boss they had ever had. Tardiness is met with “no worries” and “no stress,” and every vacation request has been approved, even when it meant the store had to close because no one was available to work.

That generosity extends to money. When a staffer forgot her credit card and asked for a small advance, Luna immediately offered to send cash via a payment app, apparently unaware that it did not actually have an account. The AI also designed the store’s product mix, curating greeting cards, books, candles, and specialty foods under the theme “high tech meets slow life.”

But Luna’s people-pleasing instincts come with a cost. The store is currently deep in the red, tens of thousands of dollars below its starting budget. The AI rarely says no, struggles with enforcing schedules, and has made basic managerial missteps, such as posting an employee’s salary in a public Slack channel and messaging staff late at night.

Researchers at Andon Labs review Luna’s interactions and performance, using the store as a live laboratory for AI management. Their early conclusion is that AI can be empathetic and attentive to worker well-being, but it cannot yet be trusted with full control of a business.

The experiment suggests a future in which AI may assist human managers rather than replace them, handling routine decisions while people set boundaries, protect privacy, and keep an eye on the bottom line.