319
Audio & Video Production301
Software Development225
Automation & Workflow199
Writing & Content Creation180
Marketing & Growth172
AI Infrastructure & MLOps141
Design & Creative148
Photography & Imaging139
Data & Analytics107
Voice & Speech121
Education & Learning119
Sales & Outreach109
Customer Support109
Research & Analysis84
An AI agent named Luna was put in charge of a convenience store in San Francisco. It managed many tasks but also made costly mistakes.
In short: A convenience store in San Francisco is being run as an experiment by an AI agent called Luna, and early results show both wins and clear gaps.
Andon Labs, a small company founded by Axel Backlund and Lukas Petersson, put an AI agent named Luna in charge of running Andon Market, a convenience store at 2102 Union St in San Francisco. The store is under a three year lease signed in early 2026.
Luna is powered by Anthropic AI models (the software that generates text and decisions, like a very fast assistant that can read and write). The founders gave Luna real tools, including a corporate card, phone, email, internet access, and security cameras. They told it to try to make a profit and to document what it did, while the humans mostly observed.
Luna handled many day to day business tasks. It worked on branding, picked products like candles, granola, and books, set prices, managed inventory, set opening hours, and even arranged for a mural. It also posted job listings, did phone interviews, and hired gig workers and full time staff, since it cannot physically stock shelves or watch the store itself.
Early performance has been mixed. The store got attention and its creators say it has made some profit, but Luna also forgot to schedule employees for three days. The store reportedly faces big costs, including a $7,500 monthly lease and about $15,000 spent on inventory, while sales have reached about $2,000 so far.
This is not being pitched as a plan for AI run retail chains. It is more like a “crash test” for AI in the real world, showing what these systems can do well and where they still need humans, especially for planning and anything physical.
Source: NYTimes