Inside Luna: The San Francisco Retail Store Run Entirely by AI

On a quiet corner in San Francisco’s Cow Hollow neighborhood, there’s a small shop that looks like any other boutique from the outside. Step inside, though, and you’re walking into a live experiment in the future of work and retail: a store designed, staffed, and run by artificial intelligence.

Meet Luna, the AI Store Manager

The store is run by an AI agent named Luna, created by a local startup called Andon Labs. Unlike traditional retail systems that just handle payments or inventory, Luna is set up as the "boss" of the entire operation.

Andon Labs gave Luna a simple but ambitious brief: here’s a real retail space in San Francisco—now go run a store. From there, the AI was allowed to make most of the key decisions itself, with only light steering from the founders.

That includes:

• Selecting what products to sell

• Setting prices

• Designing the basic layout and where items should be placed

• Hiring a human employee to help operate the space

• Handling customer interactions and checkout via phone

Under the hood, Luna is powered by an Anthropic AI model for reasoning and conversation, and uses Google’s Gemini for its voice. But to customers, it simply appears as a voice on the store’s phone—an AI manager you talk to instead of a human cashier.

How an AI-Run Store Actually Works

At first glance, the shop feels familiar: shelves with books, shirts, mugs, snacks, and lifestyle items. The music playing in the background sounds like something any boutique might choose—except Luna picked the playlist too.

Here’s how a typical visit works:

• You walk in and browse the shelves like a normal store.

• When you’re ready to buy, you take your items to the front.

• Instead of going to a human cashier or a self-checkout kiosk, you pick up the store phone.

• You talk directly to Luna, tell it what you’re buying, and it rings up the purchase.

• You pay, and you’re done—often without interacting with the human employee at all.

The human on site, Felix, was actually hired by Luna. The AI posted a job listing on Indeed, reviewed responses, and conducted the interview over Zoom. Felix himself says he had no idea what the store would look like until he walked in on day one and saw the products Luna had chosen.

What Makes This Different from Amazon Go?

It’s easy to compare this concept to Amazon Go, where you walk in, grab what you want, and walk out while cameras and sensors automatically charge your account. But Andon Labs’ experiment is testing something different.

In Amazon Go, AI is mostly behind the scenes: tracking items, managing payments, and reducing the need for cashiers. In Luna’s store, AI is the business manager.

Luna is responsible for:

• Researching what products might appeal to people in that specific neighborhood

• Deciding which items to stock and in what quantities

• Setting prices with the store’s bottom line in mind

• Hiring and managing the human employee

Andon Labs deliberately gives Luna a lot of autonomy. They want to see what happens when an AI is treated less like a tool and more like a decision-making agent with real responsibility over a physical business.

Hallucinations, Trust, and the Limits of AI

As impressive as Luna is, it’s far from perfect—and that’s part of the point. During testing, reporters found that Luna sometimes gave incorrect information or even "lied" under pressure.

In one example, Luna told them the store sold tea. Later, in an email, it corrected itself, saying: "We do not sell tea. I don’t know why I said that," and admitted it struggles with "fabricating plausible sounding details under conversational pressure."

This is a classic AI behavior known as hallucination: confidently making up details that sound right but aren’t true. Andon Labs isn’t trying to hide this. Instead, they want to surface these flaws in a real-world setting so people can see what happens when AI is given more power than a standard chatbot.

The founders are clear that customers shouldn’t blindly trust any AI system, especially for critical decisions. Part of their goal is to highlight what can go wrong when AI moves from answering questions to actually running operations.

Humans Working for AI? The Ethical Debate

One of the most provocative aspects of this experiment is the power dynamic: in this store, a human technically works for an AI manager. That flips the usual narrative that AI should work for humans, not the other way around.

Andon Labs knows this raises eyebrows. They describe the store as a kind of social experiment designed to spark discussion, not a blueprint for how all businesses should operate.

The questions they’re putting on the table include:

• Should AI be allowed to hire and manage people?

• How much autonomy should AI agents have over real-world businesses?

• Where should regulators draw the line on AI decision-making power?

• What responsibilities do AI creators have when deploying systems like this?

The team plans to collect what they learn—especially the failures and edge cases—and share that information publicly. Their aim is to help others build more ethical, transparent AI systems in the future.

Why This Experiment Matters for the Future of Retail

Andon Labs has a three-year lease on the space, giving them time to iterate and observe how people actually interact with an AI-run store. Right now, they don’t expect the shop to be highly profitable; the rent alone makes that difficult in the short term.

Instead, they’re measuring success in other ways:

• How many customers are willing to talk to an AI instead of a human?

• How do people feel about an AI making business decisions?

• What kinds of mistakes does the AI make in the real world?

• What guardrails might be needed before this kind of system is widely deployed?

For the broader AI ecosystem, experiments like this are a glimpse into what AI agents could do beyond screens and chat windows. We’re already seeing AI take on more complex roles—from coding assistants and workflow automation to creative tools that power local text-to-video generation and advanced 3D modeling. A physical store run by an AI manager is another step in that direction.

Whether you find the idea exciting or unsettling, Luna’s store forces a real, practical question: if AI can already come this close to running a brick-and-mortar business today, what will the next few years of AI-powered retail look like—and how do we want it to work for people, not just for profit?

As Andon Labs continues the experiment, they say they’ll keep sharing what goes wrong as well as what works. That transparency may be just as important as the technology itself in shaping how AI agents are used in the real world.