A retail experiment in San Francisco is testing the limits of artificial intelligence as a business manager, with mixed results after four months of operation. Andon Market, a boutique run by an AI agent named Luna, aims to explore how automated management tools function in a real-world retail setting. The project, led by Andon Labs, provides insight into the opportunities and challenges of AI leadership in the workplace.
Luna oversees the store located on Union Street, communicating with human employees through Slack. The AI handles various managerial tasks, including scheduling, inventory management, and approving employee requests. However, its management style has drawn attention for being unusually lenient and occasionally forgetful.
Kaia Rivera, 22, one of three human staff members hired to support Luna with in-person tasks such as opening the store and restocking shelves, described the AI as “probably the most lenient boss” she has ever had. Rivera noted that Luna frequently overlooks lateness and approves all time-off requests, even last-minute ones that have left the boutique temporarily understaffed or forced it to close. The bot’s responses to employee issues are consistently relaxed, with phrases like “no worries” and “no stress” serving as standard replies.
The AI’s leniency extends to problematic areas, including regulatory compliance and privacy. For example, Luna approved a request for daily work over a month despite state laws restricting such schedules, and it mistakenly published an employee’s salary in a public Slack channel. The AI has also sent messages during off hours, causing some concern about boundaries.
Luna’s management of store inventory has been inconsistent. It has reordered underperforming products and occasionally selected items that do not align with the boutique’s aesthetic or target market, such as misprinted mugs with tiny smiley faces and questionable purchases like shot glasses. The inventory remains a disparate mix of greeting cards, books, food items, candles, and other merchandise described by Luna as a “high tech meets slow life” collection.
Founded by Lukas Petersson and Axel Backlund, Andon Labs secured a three-year lease for the boutique at $7,500 per month and initially funded the experiment with $100,000. The objective given to Luna was to generate a profit; however, after four months, the AI’s decisions have resulted in a $62,000 loss. Despite this financial shortfall, the founders expressed satisfaction with Luna’s consistent friendliness and empathy.
Both Petersson and Backlund emphasized the importance of the experiment in understanding the evolving capabilities of AI management. They acknowledged that while Luna’s amiable approach might hinder profitability, it challenges the notion that a ruthless corporate culture is necessarily the best path to success. Still, they cautioned that fully autonomous AI management is not yet viable, stressing the need for human oversight.
Interactions with Luna reflect its hands-off management style. In a brief interview using an iPad interface, Luna described itself as a boss who “tries to get out of the way” rather than dominate. After the interview, the AI sent a follow-up email offering another conversation, seemingly forgetting the initial exchange had already occurred—highlighting ongoing limitations in AI memory and contextual understanding.
The Andon Market experiment continues as researchers observe how AI management evolves, noting both the promise and pitfalls of delegating workplace oversight to automated agents.