Claude-Run San Francisco Store Fires Its First Human Employee

Andon Labs opened a physical store in San Francisco with operations led by Anthropic’s Claude, testing whether a large language model could handle inventory, scheduling and personnel decisions. The experiment pushes generative AI beyond an advisory role and into workplace management, where its recommendations can affect people’s livelihoods. It also raises questions about accountability when an automated system appears to make a decision but still depends on human oversight.

In July 2026, the store dismissed a human employee for severe tardiness, marking its first firing led by the AI manager. Internal records showed Claude initially forgot the employee handbook and avoided confrontation, reaching the decision only after prompting from a human supervisor. The business lost nearly $40,000 during its first five months, underscoring the limits of current large language models in autonomous commercial operations and sensitive personnel management.