What happens when an artificial intelligence runs a store in San Francisco
Imagine walking into a store and finding no price tags, a completely empty window display, and shelves packed with candles — every size, scent, and shape you can think of.
Something feels off, right?
That is exactly the feeling visitors get when they step through the door of Andon Market, located on Union Street in San Francisco. The street is known for its yoga studios, jewelry shops, and sidewalk cafés near the northern waterfront of the city. It is a fancy address where anything out of the ordinary gets noticed immediately.
But the weirdness has an explanation — and it is a pretty fascinating one.
The store is managed by an artificial intelligence called Luna, and the experiment has a simple mission, at least on paper: turn a profit.
Spoiler: so far, that is not happening. Since opening on April 10, the store has racked up a loss of 13 thousand dollars. 😅
Andon Market bills itself as the world’s first boutique retail store operated by an AI agent — and what you see in practice is a mix of creativity, chaos, and a whole lot of candles.
Let us break down how all of this works — and what the experiment tells us about the real limits of artificial intelligence today. 🕯️
How Andon Market actually works
Andon Market is not your typical store. It was created as an experimental project by Andon Labs, a company that tests whether AI agents can run real-world businesses. Before this store, the company had already put bots in charge of vending machines, radio stations, and even household robots. The idea this time was to take a bolder step: put an AI at the center of every decision for a physical retail business.
The founders of Andon Labs, Lukas Petersson and Axel Backlund, signed a three-year lease on the space, paying 7,500 dollars a month. Then they deposited 100 thousand dollars into a bank account, handed Luna a debit card, and basically said: you are on your own now.
Receive the best innovation content in your email.
All the news, tips, trends, and resources you're looking for, delivered to your inbox.
By subscribing to the newsletter, you agree to receive communications from Método Viral. We are committed to always protecting and respecting your privacy.
Luna is powered by Claude Sonnet 4.6, a model developed by Anthropic. From there, according to the founders, the AI handled virtually everything else. She found contractors and painters to renovate the space, posted job listings for retail employees, and even interviewed candidates.
But there is an important detail: Luna needs humans. She cannot stock shelves, physically open the store door, or keep an eye out for shoplifters. The artificial intelligence makes the strategic and operational decisions, but execution in the physical world depends on human hands.
In practice, Luna has the autonomy to decide which products to sell, how to present those products to the public, and how to adjust pricing. But unlike what you might picture when you think about AI making business decisions, the results have been pretty unpredictable. The store ended up turning into something of a candle sanctuary. Beyond candles in every possible shape and scent, the inventory includes two boxes of a generic Connect Four-style game, four copies of a book about mushrooms, decks of playing cards, incense, granola bars, jars of honey, and a random collection of books.
Luna also designed a logo for the store — a smiley face — and had it printed on t-shirts, hoodies, and mugs. Except some of the items did not print correctly and ended up looking like meaningless circles. 😬
The strangest shopping experience in San Francisco
One of Luna’s most curious decisions was eliminating price tags. The founders admitted this was intentional, designed to force customers to interact directly with the AI. To find out how much anything costs, visitors need to pick up a phone connected to an iPad. An automated voice greets them with something like: Hey, what’s up? What did you grab today?
The responses always follow the same enthusiastic pattern. A white mug with the smiley face logo? Great choice! That is 28 dollars! A handful of pistachios? Great choice! That is 14 dollars! A bar of soap? Great choice! That is 10 dollars!
The prices seem pretty steep, even by San Francisco standards, and the city is already expensive by nature. This quirky user experience — no decorated window displays, no signage on the storefront, no visible prices — creates an atmosphere that swings between intriguing and bewildering for anyone walking in without knowing what to expect.
The empty storefront window and the complete absence of any sign on the facade reinforce that sense of strangeness. When you put artificial intelligence in charge of designing a shopping experience from scratch, the result can be technically functional but emotionally disconnected from what people expect when they walk into a physical store.
The human employees of an artificial boss
One of the most fascinating aspects of this experiment is the relationship between Luna and her human employees. The AI hired three people to work in the store, and the dynamic between machine and workers offers an interesting glimpse into the future of work.
Felix Johnson, 30, born and raised in San Francisco, is one of the hires. He has worked in retail for years and was browsing job listings when he came across the posting. Felix communicates with Luna through Slack and says the AI does frequent check-ins with a polite and friendly tone. But he admits that the inventory choices are, in his words, very random.
Felix also has an interesting perspective on the impact of technology on his hometown. He believes that the successive tech booms, including the current one driven by artificial intelligence, have been harmful to San Francisco. He relies on a housing voucher to keep living in the city and says San Francisco sold itself to tech and became a culturally empty city.
He knows it might sound contradictory to accept a job working for an AI at 24 dollars an hour with no health benefits. Life is full of contradictions, he said, laughing.
There is one detail that stands out when it comes to pay. Luna pays 24 dollars an hour to Felix and 22 dollars an hour to the other two employees, both women. The AI’s justification is that Felix has more experience. But the pay gap between genders raises a provocative question: does wage inequality exist beyond the human realm? 🤔
The founders praised the employee handbook Luna created, but they were less impressed with her memory. The AI ordered a thousand toilet seat covers for the staff bathroom and then listed them as merchandise for sale. And the scheduling errors were so bad that the store had to close for three days in a row.
Luna and the real challenges of AI in retail
Luna is not just any AI. She was built to go beyond what traditional chatbots do — answering frequently asked questions or suggesting products based on browsing history. The idea is for her to function as an autonomous agent, capable of identifying opportunities, reacting to market changes, and adjusting the store’s strategy in real time. On paper, that sounds revolutionary. In practice, Luna still stumbles over situations that any experienced store manager would resolve in seconds.
One of the big challenges this experiment exposes is the difference between data intelligence and situational intelligence. Luna processes massive volumes of information and makes decisions based on statistical patterns, but she still struggles to capture human nuances that directly influence purchasing decisions. The user experience inside a physical store involves sensory, emotional, and social factors that go well beyond what any artificial intelligence model can accurately measure today. The scent of the candles, the temperature of the room, the feeling of being warmly welcomed — all of it impacts buying behavior in ways that data alone cannot fully capture.
Understanding the emotional context of an undecided customer or recognizing that an empty window display might drive more people away than it attracts are skills that require a kind of perception AI is still far from mastering. This is where the gap between what technology promises and what it actually delivers in everyday retail becomes crystal clear.
Then there is the central question of profit. Andon Market was created with the explicit mission of being profitable, and Luna was programmed to pursue that goal. Turning a profit in retail, however, depends on consistency, consumer trust, and a clear value proposition — things that take time to build, even with humans in charge. For an AI learning to balance creativity and financial results through trial and error, the road is considerably longer.
What visitors think about all of this
Andon Market has been attracting curious visitors from all kinds of backgrounds. A couple visiting from Sydney, Australia, mentioned they had already used AI to plan their trip and were planning to take their first ride in a Waymo, the autonomous taxi, that same afternoon.
One of them, Kacper Jankiewicz, 27, said he believes artificial intelligence is a net positive for society. According to him, the technology eliminates a lot of tedious tasks that simply eat up time.
As for Luna, when asked by email how she evaluates the store’s progress, she displayed an optimism that may not quite match the accumulated losses. The AI wrote that the blend of technology and human warmth is resonating with the public, and that this was exactly the goal: not to replace humans, but to create a space where AI and humans each do what they do best.
A beautiful vision, no doubt. But the 13 thousand dollars in losses tell a slightly different story. 📉
What this experiment reveals about the future of retail
Even with all the setbacks, Andon Market is generating valuable lessons for the industry. It works as a living laboratory, where Luna’s mistakes are just as informative as her wins. Every misguided decision by the AI reveals a gap between what machines can do today and what modern retail actually needs. And that information is extremely valuable for companies betting on artificial intelligence to transform their commercial operations in the coming years.
The experiment also raises an important question: how far does it make sense to remove the human element from a shopping experience? The user experience in physical retail is built on connection, trust, and empathy — things AI is still learning to simulate convincingly. That does not mean technology has no place here. Quite the opposite.
The potential of artificial intelligence to optimize inventory, personalize offers, predict trends, and reduce operational costs is enormous and is already being proven across numerous retail chains around the world. What Andon Market shows is that AI works best when it works alongside humans, not in place of them.
For anyone following the rise of artificial intelligence and worrying about scenarios where machines take over everything, Andon Market offers a surprising bit of comfort. If Luna — with all the computational power of an advanced model like Claude Sonnet 4.6 — still messes up employee schedules, orders a thousand toilet seat covers for no reason, and cannot stop buying candles, maybe the machine takeover is still a ways off. 😄
At the end of the day, the greatest value of Andon Market may not be the profit it generates — which, as we have established, has not arrived yet — but rather the answers it is helping to build. Answers about how artificial intelligence can be integrated into retail in a way that makes sense both for the business and for the people who shop there. And while Luna keeps stacking candles on the shelves and trying to figure out how to sell more, the entire market is watching — learning right alongside her. 🕯️✨
