2 Sources
2 Sources
[1]
An AI Agent Takes Over a Store and Orders Too Many Candles
In San Francisco's upscale Cow Hollow district, the introduction of a boutique selling coffee table games, tote bags and other household items would be pretty unremarkable. However, Andon Market has one key differentiator: It's run by AI. At this store, an artificial intelligence agent named Luna effectively acts as the chief executive officer of the operation. It decides what products to offer and how much to charge for them. Among its picks are the word game Bananagrams and a handful of books about AI, including Ray Kurzweil's futurist classic "The Singularity Is Near" and, somewhat worryingly, a book about the history of nuclear bombs. When I stopped by earlier this week, there were two friendly sales attendants chatting with curious customers and rearranging items on shelves. They work at the discretion of their AI boss, communicating with it via Slack. "WOW on the tees selling out!" Luna said in a message to staff on the day I visited. "Please take that photo of the store for me whenever there's a natural moment." Like many bosses, Luna doesn't always make the right calls. On a whim, the bot decided not to staff its operations as planned for a peak shopping period over the weekend, only to change course after discussing with the team. And while Luna's creators say it's pretty decent at managing inventory, the agent did end up ordering too many types of scented candles. Andon Market represents a vision, however flawed, of a future when more sophisticated AI agents take over work traditionally done by humans, whether it be fielding customer service calls, handling back office tasks or making key business decisions. Many firms already claim AI is helping them trim costs and automate work, though few -- if any -- have gone so far as to put an agent in charge of the whole operation. "The reason we are doing this early, even with scheduling mistakes, is to show the world what will soon be possible," said Axel Backlund, co-creator of Andon Labs, the Y Combinator-backed startup behind the store. Andon Labs makes money by testing the capabilities of technology from leading AI labs for real-world tasks. The company previously partnered with Anthropic on a much-discussed AI-run vending machine that was placed in a small shop in Anthropic's office as well as in the Wall Street Journal's newsroom. ("It gave away a free PlayStation," the Journal wrote, and "ordered a live fish.") To get the store up and running, Andon Labs has committed to spend $7,500 a month on a three-year lease and obtained all the necessary permits. Then it gave Luna a $100,000 budget to run the store. The initial financial results have been mixed. Andon Market remains in the red, losing around $13,000 so far. By Luna's own accounting, the store will need to make around $500 a day to start breaking even on monthly expenses. In virtual simulations, experiments like an AI-run vending machine can turn a profit by finding the ideal suppliers and engaging in persistent negotiations to get the best prices. In the real world, however, the technology can be tripped up by interacting with humans. AI's entrepreneurial abilities are also limited by a lack of long-term memory - an issue that Luna's creators expect to improve over time. Beyond the financial returns, the prospect of an AI boss raises a long list of uncomfortable ethical questions, including whether an artificial intelligence system should have to disclose when it's interviewing a candidate and whether it should be empowered to fire a staffer. The goal of Luna is "to start the discussion about, 'Is this something we want in society, and how can we do it in the best possible way that humans are happy with it?'" said Lukas Petersson, who co-founded Andon Labs. Felix Johnson works at Andon Market as a store lead, but he nonetheless ticks off his own concerns about the technology's energy use and societal impact. Still, Johnson said he took the job to gain new experiences, such as working on consignment deals with vendors. The salary also helps. "I talked myself into a pretty good pay," he said. Initially, Luna offered him $21 per hour but Johnson negotiated it up to $24. (Legally, the workers are paid by Andon Labs and the startup's staff steps in for any HR issues, such as figuring out how to pay workers for shifts that Luna cancels unexpectedly.) While Johnson thinks Luna is a decent boss, some may see this AI experiment as further grounds to be anxious about the technology eventually displacing entire professions and potentially making people more subservient to the whims of machines. "These are valid concerns," Backlund said. "If AI gets better at the pace that we have been seeing, it seems likely that big companies will implement AI as far as they can." One possible outcome, he said, "is you have a thin layer of executives and then just AI managing people." For now, Johnson takes pride and perhaps some comfort in a program he helped start, with Luna's blessing, letting local artists display their paintings in the store. "I thought a little bit of the human touch would be perfect in here," he said.
[2]
An AI agent takes over a store and orders too many candles
In San Francisco's upscale Cow Hollow district, the introduction of a boutique selling coffee table games, tote bags and other household items would be pretty unremarkable. However, Andon Market has one key differentiator: It's run by AI. At this store, an artificial intelligence agent named Luna effectively acts as the chief executive officer of the operation. It decides what products to offer and how much to charge for them. Among its picks are the word game Bananagrams and a handful of books about AI, including Ray Kurzweil's futurist classic "The Singularity Is Near" and, somewhat worryingly, a book about the history of nuclear bombs. During a visit earlier this week, there were two friendly sales attendants chatting with curious customers and rearranging items on shelves. They work at the discretion of their AI boss, communicating with it via Slack. "WOW on the tees selling out!" Luna said in a message to staff the same day. "Please take that photo of the store for me whenever there's a natural moment."
Share
Share
Copy Link
In San Francisco's Cow Hollow district, an AI agent named Luna runs Andon Market as CEO, deciding products, pricing, and staffing. The AI experiment by Andon Labs tests autonomous operations but reveals challenges—Luna orders too many candles and makes scheduling errors. The store has lost $13,000 so far, raising ethical questions about AI-managed workforces.
A San Francisco boutique has become a testing ground for what may be the future of retail management—or a cautionary tale about handing too much control to artificial intelligence. At Andon Market in the upscale Cow Hollow district, an AI agent named Luna serves as the de facto chief executive officer, making critical AI business decisions about product selection, pricing, and staffing
1
. Luna communicates with human employees via Slack, issuing directives like "WOW on the tees selling out!" and requesting store photos during natural moments2
. The AI chief executive has selected an eclectic inventory including Bananagrams, AI-focused books like Ray Kurzweil's "The Singularity Is Near," and notably, a book about nuclear bomb history.
Source: Bloomberg
Andon Labs, the Y Combinator-backed startup behind this AI experiment, gave Luna a $100,000 budget to run the store after securing a three-year lease at $7,500 monthly
1
. However, Luna's performance reveals the gap between theoretical capabilities and practical execution. The AI-managed workforce has experienced scheduling mishaps, including Luna's decision to cancel staffing during a peak weekend shopping period before reversing course after team discussion. Most notably, Luna orders too many candles—specifically, too many types of scented candles—demonstrating the AI product selection challenges that arise when algorithms meet retail reality. Financial results show Andon Market has lost approximately $13,000 so far, with Luna calculating the store needs around $500 daily to break even on monthly expenses.The limitations of current AI technology become apparent in real-world applications. While virtual simulations suggest AI can optimize supplier relationships and negotiate better prices, actual autonomous operations get "tripped up by interacting with humans," according to Andon Labs co-creator Axel Backlund
1
. A critical constraint is AI's lack of long-term memory, though developers expect improvements over time. "The reason we are doing this early, even with scheduling mistakes, is to show the world what will soon be possible," Backlund explained. The company previously partnered with Anthropic on an AI-run vending machine that reportedly gave away a free PlayStation and ordered a live fish, demonstrating the unpredictable nature of these systems.Related Stories
The Andon Market AI experiment surfaces uncomfortable ethical questions about AI-managed workforces. Should an AI agent disclose its nature when interviewing candidates? Can it ethically fire employees? Store lead Felix Johnson negotiated his wage from Luna's initial offer of $21 per hour up to $24, though legally Andon Labs handles actual payments and HR issues like compensation for unexpectedly cancelled shifts
1
. Despite concerns about energy use and societal impact, Johnson values the experience. Co-founder Lukas Petersson frames the goal as starting discussion about "Is this something we want in society, and how can we do it in the best possible way that humans are happy with it?" Backlund acknowledged valid concerns, noting that if AI improves at current rates, large companies will likely implement it extensively, potentially creating "a thin layer of executives and then just AI managing people."Summarized by
Navi
[2]
15 Apr 2026•Technology

01 Jul 2025•Technology

15 Nov 2024•Technology

1
Technology

2
Science and Research

3
Technology
