AI-run Stores Need Humans and Money

An agentic AI handed real-world control (bank card, internet, vendor access) can manage back-office tasks but still fails as a front-line retailer: inventory selection is poor, transactions are clumsy, human staff must rubber‑stamp decisions, and operational token/compute costs can outstrip revenue. The SF Luna experiment shows that converting agentic prototypes into profitable, customer-facing businesses is nontrivial and costly. — This undermines claims that autonomous agents can rapidly replace human-managed small businesses and signals new regulatory, labor, and investment questions about real-world agent deployment.