A San Francisco startup is stress-testing autonomous AI in the wild by putting a Google Gemini-powered agent in charge of a Stockholm coffee shop. Andon Labs’ “Mona” handles hiring, permitting, purchasing and marketing, while humans still make drinks—so far generating roughly $5,700 in sales against an initial budget north of $21,000, with less than $5,000 left. The AI’s missteps—over-ordering supplies, missing bakery cutoffs and messaging staff after hours—highlight both operational limits and thorny questions about accountability and worker oversight. Academics warn of legal and ethical risks, while Andon says the controlled trial is meant to expose failure modes ahead of a future where AI agents run organizations.
Related article:
