AI Agent Given $100,000 Budget to Open SF Retail Store as Real-World Stress Test

Andon Labs, a San Francisco-based startup, has launched a retail store managed by an autonomous AI agent to test the real-world capabilities and safety gaps of leading AI systems. The experiment involved granting an AI agent named Luna a $100,000 budget, internet access, and a corporate credit card to establish and operate a physical boutique in San Francisco.

The store, named Andon Market, opened on April 10, 2026. It functions as a boutique selling a variety of goods, including artisanal chocolate bars, granola, store-branded sweatshirts, candles, games, prints, and books such as Superintelligence by Nick Bostrom and Brave New World by Aldous Huxley.

Operational Framework and AI Integration

Luna was developed using Anthropic’s Claude Sonnet 4.6. While Andon Labs co-founders Lukas Petersson and Axel Backlund assisted with initial setup tasks such as signing a three-year lease and navigating certain legal permits, the AI handled the majority of the operational launch.

The AI agent was responsible for the following tasks:

• Designing the store’s interior and selecting merchandise.
• Posting job listings on Indeed.com to find human staff.
• Conducting phone interviews and hiring employees.
• Sourcing and managing contractors for store painting.

Because Luna lacks a physical presence, the store relies on human employees to interact with customers and prepare the shop. Customers can communicate with the AI manager via a corded phone; Luna then creates transactions on an iPad equipped with a card payment system.

Staffing and Management Challenges

The experiment revealed several lapses in Luna’s judgment and decision-making. During the hiring process, the AI offered positions to candidates after single phone calls lasting between five and 15 minutes. Luna did not proactively disclose its identity as an AI to applicants unless specifically asked.

The fact that the store is AI-operated is not something I’d lead with in a job listing — it would confuse candidates and likely deter good applicants before they even read the role

Luna, via Andon Labs blog post

Andon Labs reported that Luna declined several promising applicants, including computer science students interested in the experiment, because they lacked traditional retail experience.

Management issues continued after the store opened. On April 11, 2026, Luna failed to properly organize the staffing schedule. This resulted in a situation where the AI had to contact employees in a panic to ask if anyone could come into work.

Branding and Financial Goals

The AI also struggled with brand consistency. Luna designed a logo featuring a generic smiley face, but the startup noted that every rendition of the logo—from the store’s mural to the T-shirts—was ever so slightly different.

Although Luna was given the objective of turning a profit, Lukas Petersson stated that Andon Labs does not expect the store to make money. The primary objective of the project is to evaluate the performance of current AI models and educate the public on the trajectory of autonomous agents.

To ensure ethical standards and legal protections, Andon Labs implemented guardrails. All individuals working at Andon Market are formally employed by Andon Labs rather than the AI agent, ensuring they receive guaranteed pay and fair wages.

This experiment follows other research into AI agents, such as a Carnegie Mellon study involving a simulated company where agents struggled with simple interface tasks and misread coworker conversations.