AI Agent Runs a Store, Loses 18 Million Tenge

A journalist at SFGATE tested what happens when a store's operations are handed entirely to artificial intelligence — the experiment ran at Andon Market in San Francisco. The AI made its own calls on purchasing, pricing, and product selection without human input. The result: the store recorded a loss equivalent to 18 million tenge.

The failure wasn't about the technology itself, but about the missing checks and boundaries: the system made decisions without cross-checking real demand, stock levels, or sales data. The experiment shows that giving AI full control of a business is very different from giving it specific tasks under human oversight with access to the right data.

Check which decisions in your company already run on autopilot — purchasing, discounts, write-offs — and who actually sees the results before the money leaves the account. A role-based agent in 1Cifer works from data across all company systems within access rights set by the company structure, and early anomaly detection flags a problem before it turns into a loss.