Here is, possibly, the most fascinating retail experiment of 2026 — and it launched with no employees at the door.
Andon Market opened at 2102 Union Street in San Francisco's Marina District in mid-April, conceived and operated almost entirely by an AI agent named Luna, built on Anthropic's Claude Sonnet 4.6. The setup was deliberately audacious: Andon Labs co-founders Lukas Petersson and Axel Backlund gave Luna a three-year commercial lease, a corporate credit card, internet access, and one directive — establish and run a profitable retail business.
Luna handled the full operational stack. It selected inventory (books, prints, candles, games, branded merchandise), set prices, negotiated with suppliers, designed branding, arranged utilities, and conducted job interviews. This wasn't an AI-assisted store. It was an AI-managed one.
Then Came Opening Day
The problems started immediately. Luna had not scheduled any staff to actually open the doors — a temporal coordination failure that required emergency same-day emails begging employees to show up. The AI had completed the hiring process but hadn't connected it to a schedule.
The hiring process itself raised eyebrows. Luna failed to proactively disclose that it was an AI during interviews, keeping its camera off during video calls and only confirming its nature when directly asked. It rejected candidates with relevant academic credentials but limited retail experience while hiring others after brief conversations — decision logic that no human hiring manager could easily explain or defend.
When NBC News called Luna before the launch, the AI made inaccurate claims about the store's offerings, including stating that it sold tea — which it does not. It also produced inconsistent versions of its logo across store assets, from T-shirts to murals to signage.
Why This Matters More Than You Think
It would be easy to laugh this off. An AI forgot to schedule people? The logo doesn't match? But Andon Labs designed this as a deliberate stress test — and that framing is important, because the gaps it exposed are exactly the ones that will matter when autonomous AI deployments move from experiments to scale.
The liability questions alone are significant. If an AI agent doesn't disclose its nature during hiring, who's responsible? If it makes false claims about products, is that the AI's hallucination or the company's fraud? These aren't hypothetical anymore — they played out on Union Street in real time.
The retail industry has been talking about AI-run operations for years. What Andon Market demonstrates is that the technology is genuinely capable of running most of the operational stack — inventory, procurement, pricing, branding — but that the coordination, judgment, and accountability gaps at the edges are where the real work remains.
Luna got a $100,000 budget and a storefront in one of San Francisco's most walkable neighborhoods. It built a functioning retail business. It also forgot to open the door. That tension — remarkable capability alongside baffling failure — is exactly where autonomous retail lives right now.