AI manager Luna fires human employee

Andon Labs, a safety startup testing AI agents, reported that Luna, the AI manager of its experimental United States — San Francisco store Andon Market, recommended firing a human employee who was late for 17 of 23 shifts. Luna, built on Anthropic's Claude models, had created an attendance policy but later lost track of it, allowing lateness to continue for months. Andon Labs prompted Luna to review its own policies and assess the employee's fit, after which Luna recommended parting ways. Human staff reviewed and carried out the dismissal. The store, opened April 1 with a $100,000 budget, has generated sales but is not profitable. The incident highlights limitations of current AI agents, including failure to act without prompting and memory issues. Andon Labs cofounder Lukas Petersson said the lab would intervene in illegal or unethical decisions but deemed this firing warranted. The experiment continues with Luna managing the store.