In a pioneering experiment at a San Francisco retail outlet, an artificial intelligence agent named Luna, tasked with overseeing store operations and employee management, recommended the termination of a human employee due to repeated lateness.
Andon Labs, the AI research company behind the initiative, deployed Luna at their Andon Market store. The AI agent holds significant autonomy, including managing a $100,000 budget, setting prices, creating branding, and handling hiring processes. While humans still perform physical tasks and secure permits, Luna is responsible for enforcing workplace policies.
The incident unfolded when an employee accrued 23 instances of lateness. Luna, which had previously established an employee handbook and attendance policy, reviewed the worker's record. Despite earlier warnings and additional training, the lateness persisted. Researchers prompted Luna to re-evaluate its own policies and the employee's suitability for the role, with human input on prior conversations also provided.
After its assessment, Luna, operating on Claude Opus 4.8, recommended ending the employee's tenure. It also presented alternatives, such as a final written warning and a two-week improvement plan. Crucially, while Luna made the recommendation, human staff at Andon Labs reviewed its findings and conducted the final conversation, making the ultimate employment decision. This marks a significant moment in understanding how AI agents might influence future workplace decisions, raising questions about the extent of authority automated systems should wield.