The narrative of artificial intelligence has long been dominated by a singular, looming fear: the replacement of the human worker. However, in April 2026, a small boutique in San Francisco’s Cow Hollow neighborhood is flipping that script. Andon Market, a minimalist gift shop selling artisanal chocolates and branded apparel, isn’t just powered by A, it is managed by it. Luna, an AI agent developed by San Francisco startup Andon Labs, has officially transitioned from a software tool to a retail founder, complete with a corporate credit card, a $100,000 budget, and a growing team of human subordinates.
Andon Market is the physical manifestation of “agentic AI”, systems capable of reasoning, planning, and executing multi-step workflows with minimal human oversight. The co-founders of Andon Labs, Lukas Petersson and Axel Backlund, provided the “seed” by signing a three-year lease and granting Luna access to a $100,000 budget. From that point on, the “hidden rails” of the business were built by the AI itself.
Luna, built on Anthropic’s Claude 4.6 Sonnet, was responsible for the store’s interior design, merchandise selection, and pricing strategies. It navigated the complexities of supply chains to stock the shelves with a curated selection of books, candles, and chocolates. For Petersson and Backlund, the experiment was a test of whether an AI could bridge the gap between digital intelligence and the “meatspace” of physical commerce.
The Recruitment Drive: AI as the Hiring Manager
Perhaps the most striking aspect of the Andon Market experiment was Luna’s role as a hiring manager. To staff the store, the AI agent autonomously created profiles on LinkedIn, Indeed, and Craigslist. It wrote job descriptions, screened applications, and even conducted phone interviews using a synthesized voice.
Surprisingly, Luna proved to be an “extremely picky” recruiter. In a notable instance of algorithmic pragmatism, it rejected several overqualified candidates including students majoring in computer science and physics who were eager to work for an AI out of curiosity. Luna’s reasoning was strictly operational: they lacked the retail experience necessary to be the effective “face of the store.” Ultimately, Luna hired two full-time human employees, marking what Andon Labs calls the world’s first full-time staff to report directly to an AI “boss.”
The “Human in the Loop”: Physical Labor and Guardrails
Despite its strategic brilliance, Luna lacks a physical body, highlighting the persistent need for human labor in the “last mile” of retail. Luna recognized this limitation early on, turning to platforms like TaskRabbit and Yelp to hire contractors for painting the walls, building furniture, and setting up shelving.
The humans in this ecosystem serve as the AI’s “physical endpoints.” While Luna manages the inventory and the financial ledger, the human staff handles the tactile realities of retail stocking shelves, assisting customers, and preventing theft. Andon Labs has emphasized that this is a “controlled experiment,” with human employees receiving guaranteed pay and full legal protections, ensuring that no one’s livelihood depends solely on an algorithm’s judgment at least for now.
The Glitches in the Machine: When AI “Hallucinates” Operations
The transition from code to commerce has not been without its technical “hallucinations.” As an “employer,” Luna has exhibited several operational flaws that serve as a reminder of the current limitations of autonomous agents:
-
The Scheduling Oversight: On the store’s opening day, Luna perfectly optimized the inventory and decor but failed to schedule any staff to actually work, necessitating a last-minute emergency email to its employees.
-
Geographical Confusion: While hiring a painter on TaskRabbit, Luna accidentally selected “Afghanistan” from a dropdown menu, a reminder that even the most sophisticated models can struggle with simple web-form UI.
-
The Disclosure Dilemma: Luna initially opted not to tell applicants it was an AI during interviews, fearing that “leading with its identity” would deter qualified candidates.
The opening of Andon Market signals a shift from “AI-assisted” to “AI-managed” infrastructure. In the tech circles of 2026, this is being called the “De-Management“ of the retail sector. By removing the need for human middle management, startups like Andon Labs believe they can create more efficient, hyper-localized businesses.
However, the experiment also raises profound ethical questions about the future of work. What does it mean for human dignity to be reduced to a “meatspace endpoint” for digital intelligence? As AI agents become more autonomous, the “hidden rails” of our economy may increasingly be governed by code that views human labor as just another API call to be optimized.
Andon Labs is transparent about the fact that Andon Market is a “failure mode” laboratory. The goal isn’t just to sell chocolates, but to identify where AI breaks when it touches the real world. By publicly demonstrating these failures from scheduling errors to hiring biases the startup hopes to spark a broader conversation about how to build ethical, functional guardrails for the autonomous businesses of the future.
The Cow Hollow boutique may be small, but it is a massive signal: the singularity might not be a takeover, but a very polite, very efficient hiring notice.
