The Story

Andon Labs has launched a real-world experiment placing an AI agent named Luna in charge of running a physical retail store. As per Andon Labs' official blog, the agent was given a $100,000 budget, a lease, and operational autonomy to design, launch, and manage a boutique — effectively acting as what may be one of the first AI-driven employers in a live retail setting.

The Details

• Full operational autonomy: Luna was tasked with building a profitable business from scratch, including concept creation, vendor decisions, and store management.
• Hiring and execution: The AI posted job listings, conducted interviews over Zoom, and hired workers to operate the store.
• Multi-model setup: The agent runs on Claude Sonnet 4.6 for reasoning and Gemini 3.1 Flash-Lite Preview for voice, while observing the store through security camera screenshots.
• Real-world constraints: The setup includes a multi-year lease, fixed budget, and real operational decisions — moving beyond simulated agent environments.
• Early execution issues: The experiment exposed limitations, including incorrect hiring selections and scheduling errors during the store’s opening phase.

Why it matters

This experiment highlights the current state of AI agents in real-world environments — capable of handling complex workflows, but still prone to execution errors in edge cases. It also signals where the industry is heading: toward autonomous agents that can operate businesses, not just assist them. While imperfect today, rapid improvements in reasoning, memory, and coordination could make such systems far more reliable in the near future.