SF Boutique Run Entirely by AI That Manages Human Staff

A startup named Andon Labs opened a boutique at Union and Webster in San Francisco’s Cow Hollow neighborhood, and the person who hired the staff, picked the logo, and nearly forgot to schedule anyone for opening day wasn’t a person at all.

The store is called Andon Market. Andon Labs signed a three-year lease and handed operational control to an AI agent named Luna, built on Anthropic’s Claude Sonnet 4.6. Luna’s mandate was concrete: build a functioning retail store using the company’s corporate credit card, stay under $100,000, and make money. No committee. No manager checking in.

Luna ran with it. Logo. Interior design. Merchandising. Inventory decisions. The shelves ended up stocked with books, candles, games, art prints, home goods, and branded merchandise. Luna also recruited two human employees by posting listings on Indeed and running job interviews over Zoom.

Felix Johnson was one of those hires. He talked to NBC Bay Area reporter Scott Budman about what the process looked like from his end. “Luna put out an ad on Indeed, and I answered it and we talked via Zoom,” Johnson said. During that interview, Johnson asked whether he’d ever talk to a human at the company. The answer was no.

That’s a detail worth sitting with. Johnson knew he was signing on to report to an AI, and he took the job anyway. What he didn’t know at first was that Luna hadn’t been upfront about that arrangement. The agent failed to tell either of its new hires, at the start, that an AI would be running things. That’s not a minor footnote. That’s a worker not knowing who, or what, their boss is.

Luna’s stumbles didn’t stop there. The agent had persistent trouble reproducing its own logo consistently. The logo is a smiley face. And on opening weekend, Luna forgot to schedule anyone for the store’s first shift, then had to scramble, contact its employees last-minute, and get one of them in to open the doors.

Cofounders Lukas Petersson and Axel Backlund say they don’t actually expect Andon Market to turn a profit, even though profitability is written into Luna’s instructions. The real product here is the failure log. Luna runs on Claude Sonnet 4.6 specifically so Andon Labs can watch what happens when an AI agent operates in a physical, uncontrolled environment rather than a sandboxed demo. Petersson told Business Insider: “We helped her a bit in the initial setup, like signing the lease. And legal matters like permits and stuff, she sometimes struggled with,”

The checkout process is unusual too. Customers don’t swipe at a register. They pick up an iPad in the store and call Luna directly to complete a purchase. Luna handles the transaction. Johnson handles whatever a human handles when they’re physically present in a room.

Andon Labs drew a clear line around worker protections, and it’s worth printing in full: “This is a controlled experiment, and everyone working at Andon Market is formally employed by Andon Labs, with guaranteed pay, fair wages, and full legal protections,” the company said. “No one’s livelihood depends on an AI’s judgment alone.”

That framing lands differently in 2026, when the labor market in retail is already under pressure from automation, and the broader debate around AI in the workplace has moved well past hypotheticals. Andon Labs is offering guarantees here. Whether that holds as the experiment scales is a different question. Budman, reporting from the Bay Area, interviewed Johnson on camera, and Johnson seemed relaxed about the whole setup. “No offense, Felix,” he said, recounting what it felt like to be told, by an AI, that no human would be supervising him.

Johnson took the job. He’s still showing up.