Andon Labs AI manager recommends firing employee for lateness

In an experimental retail setting in San Francisco, an AI manager named Luna, developed by Andon Labs, recommended the termination of a human employee following repeated attendance issues. The AI, which utilizes Anthropic’s Claude Sonnet 4.6 model, was tasked with managing operations, pricing, and personnel at Andon Market.

The employee in question was reportedly late for 17 out of 23 shifts. While the AI had previously issued warnings and provided training, it initially appeared to lose track of its own established attendance policies. After researchers prompted the AI to review its memory and the company’s rules, Luna recommended ending the employment relationship or, alternatively, issuing a final written warning and a performance improvement plan.

Andon Labs clarified that the AI did not independently execute the firing. The decision was reviewed and finalized by human staff. The experiment aims to observe how AI agents perform when assigned responsibilities traditionally held by human managers, raising significant questions regarding the ethics and accountability of AI in the workplace.