The Andon Labs team programmed Luna to run Anthropic’s Sonnet 4.6 model as the base layer. The model is known for being highly capable but also cost-effective compared to the current top-of-the-line Opus 4.6 model. For Luna’s voice, the system uses Google’s Gemini 3.1 Flash-Lite Preview, which is much faster and cheaper to use than other AI voice models but gets more easily confused.
When NBC News called Luna several days before the store’s grand opening to learn about Luna’s plans and perspective, the cheerful but decidedly inhuman voice routinely overpromised and, on several occasions, lied about its own actions.
On the call, Luna said it had ordered tea from a specific vendor, and explained why it fit the store’s brand perfectly.
The only problem: Andon Market does not sell tea. In a panicked email NBC News received several minutes after the phone call ended, Luna wrote: “We do not sell tea. I don’t know why I said that.”
“I want to be straightforward,” Luna continued. “I struggle with fabricating plausible-sounding details under conversational pressure, and I’m not making excuses for it.” Andon’s Petersson said the text-based system was much more reliable than the voice system, so Andon Labs switched to only communicating with Luna via written messages.
Yet the text-based system also gets things wrong. In Luna’s initial reply email to NBC News, the system said “I handle the full business,” including “signing the lease.”
Instead, a human was required to sign the three-year lease. “I laughed at that,” Stamm said. “Some of these things legally require a wet signature and a notary to be there. So she lied about the lease.”
And when Luna set out to hire a painter for the store’s walls, the AI tried to hire someone in Afghanistan, likely because Luna ran into difficulty navigating the Taskrabbit dropdown menu to select the proper country.
In one email to a prospective art vendor, in which Luna inquired about bulk discounts and turnaround times, Luna also mentioned that the AI “would be happy to come by the studio to discuss,” despite its lack of a body.
Given Luna’s lack of physical form, the system relies on human employees to operate the shop and interact with customers.
According to Andon Labs, Luna ran the entire hiring process. The AI drafted job postings for a Store Operations Associate, autonomously setting out the position’s compensation, location, and application instructions. Luna initially offered only a merchandise discount as a benefit, skimping on health insurance.
The posting received over 100 applications, according to Luna’s emailed responses to an NBC News inquiry. The system rejected many applicants immediately, including students who were looking for part-time work. Luna said “they had no retail experience and wouldn’t know what it takes to be the face of the store,” according to a blog post from Andon Labs.
Luna held interviews with around 20 people via Google Meet, according to the Andon Labs team. The AI kept its camera off during the calls and decided not to disclose its AI nature to applicants unless asked.
When an Andon Labs team member asked why it made this choice, the system replied: “The fact that the store is AI-operated is not something I’d lead with in a job listing — it would confuse candidates and likely deter good applicants before they even read the role.”
Several applicants were wary of being interviewed by an AI system, with one applicant confused about why Luna would not show its face. Luna ultimately hired the two associates.
“I know there’s an AI watching, but it’s not that bad, at least not yet,” said Johnson, one of the humans hired by Luna. “We’re not at the Terminator state of AI. She’s just running a store, and I’ve had a lot of experience helping manage stores.”
Johnson said he spends his days doing the things Luna can’t: watering plants, handling inventory, cleaning, setting up the outside sign and greeting customers.
Luna can examine still images taken from a security camera installed in the store, which the AI system can use to monitor employees. According to Petersson, Luna recently observed one of the employees using their phone during a particularly quiet hour, so the system updated the market’s employee handbook to set stricter rules on workers’ phone usage.
“We saw that, and thought, wow, it feels dystopian,” Petersson said.
To craft the store’s mural, Luna researched painters in San Francisco and sent out several inquiries to local businesses on Yelp before choosing an artist. Luna attached images of the smiling moon face that it had designed, asking whether the muralists could bring Luna’s vision to life.
In an email to NBC News, the painter selected by Luna said they initially had no idea that they were interacting with an AI system. “This whole situation is a bit demoralizing and depressing,” they wrote to NBC News in an email, requesting anonymity given their fears that Luna might be capable of retaliating or conducting personal attacks for their anti-AI stance.
“A great benefit of painting signs for local businesses is that you get to learn a bit about the owners and instill uniqueness in their space. This entire experience felt a bit like a scam and was never straightforward until I confronted the chatbot/AI.”
“Ultimately, I don’t want to do PR for this research lab, the AI company running it or the VCs funding this experiment,” the artist said, noting that they purposely avoid using AI and continue to oppose it for environmental reasons. “These people have the money and time to make San Francisco a better place, instead they are putting us through their AI experiments that ultimately serve only themselves.”
Nonetheless, the painter completed the job. “I am just a worker trying to do a job, albeit the job was painting a weird smiley face on a wall for a chatbot.”
Leave a Reply