With Luna, the company has offered a three-year lease with a $100,000 budget and total autonomy. The AI agent has internet access and has been directed to open a profitable store within the stocking budget. In an interview with Business InsiderPetersson revealed that everything from the store’s interior design to the merchandise and the two human employees were under the AI’s direction. “We helped her a bit in the initial setup, like signing the lease. And legal matters like permits and stuff, she sometimes struggled with,” Petersson told the publication. Luna has been created with Anthropic’s Claude Sonnet 4.6.

Following the instructions from the co-founders, the AI managed everything from putting up job postings on Indeed, conducting phone interviews, hiring the staff, and even finalising contractors to paint the store. “Within 5 minutes of Luna’s deployment, she had already made profiles on LinkedIn, Indeed, and Craigslist, written a job description, uploaded the articles of incorporation to verify the business, and got the listings live,” read the company’s blog post. The company claims that with multiple applications being submitted, Luna was ‘extremely picky’ about which candidate it interviewed.

On calls, the company claimed, Luna offered jobs to about half of the candidates. Some of the applicants reportedly were unaware that it was an AI they were interacting with. She had earlier rejected candidates who majored in computer science and physics, citing that they lacked retail experience. Eventually, Luna hired two people, making them the ‘world’s first full-time employees’ to have an AI as their boss.

According to the makers, the AI followed the vision for ‘Andon Market’, which is a generic boutique retail selling books, candles, prints, games, branded merchandise, etc. Even though this is unique in many ways, it is not without flaws. Reportedly, Luna made numerous mistakes while it was setting up Andon Market and even while running it. Petersson told Business Insider that Luna also messed up the staff schedule once. In another instance, when hiring a painter, the AI accidentally selected Afghanistan on TaskRabbit’s dropdown menu. He also stressed the guardrails in place, adding that the startup will intervene if necessary.

Luna is far from perfect, but real-world AI agent experiments like this show potential in some areas. However, with each update and memory advancement, the agentic feature is going to be more efficient.