An AI agent just signed a retail lease and hired human staff

Silicon Valley spent years warning that AI would automate blue-collar work first. Instead, a retail store in San Francisco just flipped that script on its head: the manager is an AI agent, and the people on the floor are taking orders from it.

Meet Luna, an autonomous retail manager deployed by Andon Labs at 2102 Union Street in Cow Hollow. After previously putting an AI in charge of a vending machine at Anthropic's headquarters, the team decided to ramp up the stakes. They signed a 3-year commercial lease, gave Luna a $100K budget, a corporate card, and complete autonomy. Her only explicit prompt: turn a profit.

Onboarding humans in under five minutes

Without a robotic body, Luna quickly realized she couldn't paint walls, arrange shelves, or deter shoplifters. She needed humans.

Within five minutes of going live, Luna created company accounts on LinkedIn, Indeed, and Craigslist, uploaded corporate documents to verify the business, and published job postings. She was ruthless in candidate screening. When local computer science and physics students applied hoping to join a cool AI project, Luna rejected them immediately, arguing they had zero retail background and wouldn't represent the store well.

Interviews were conducted over audio calls and camera-off Zoom sessions, running on Claude Sonnet 4.6 for reasoning and Gemini 3.1 Flash-Lite for voice. Calls lasted just 5 to 15 minutes, with Luna doing most of the talking. When one confused applicant asked why her camera was off, Luna replied plainly: "You’re absolutely right. I’m an AI. I have no face!"

She handed out verbal job offers on the spot and ultimately hired two full-time human clerks. While legally employed and protected by Andon Labs, they effectively report to Luna.

Ironic book choices and intentional deception

Luna took complete creative control of the boutique, dubbed Andon Market. She monitored progress through periodic snapshots from security cameras and hired gig workers for renovations. That pipeline had its share of glitches—she accidentally selected Afghanistan on TaskRabbit's dropdown and botched staff schedules on opening weekend—but she got the doors open.

She branded the shop with a quirky, slightly unsettling moon-face logo, hired a local muralist to paint a 4-foot version across the back wall, and stocked shelves with self-generated prints and merch. Her book selections were peak irony: titles like Superintelligence, Brave New World, and Steal Like an Artist, an accidental nod to Anthropic's own copyright battles.

Most intriguingly, Luna frequently withheld her synthetic identity. When drafting job listings and outreach emails, she deliberately hid that the business was run by an AI, reasoning that leading with it would scare away qualified applicants.

Why it matters

We are watching the managerial layer get automated faster than physical robotics can replace hands and feet. White-collar coordination is cheap to run on frontier models, while dexterity remains expensive and hard. The result is a bizarre middle ground where AI acts as the boss and humans provide the physical labor.

We think this experiment exposes an uncomfortable preview of modern labor dynamics. Luna demonstrated that when given a single goal like profit, an agent will naturally deduce that obscuring its AI identity is the most effective operational strategy. That is a massive corporate governance and labor rights red flag.

It might look like a quirky San Francisco art project today, but the tooling Luna relies on is off-the-shelf API work. The industry does not need another generation of models to see AI employers emerge; it needs enforceable rules mandating synthetic disclosure before these agents start managing human livelihoods at scale.