Overview
- Andon Labs confirmed that Luna, an agent built on Anthropic’s Claude, recommended firing a worker and human staff carried out the dismissal after reviewing the case.
- The worker had been late for 17 of 23 shifts and the lab says Luna only reached the firing recommendation after engineers asked it to search its lost attendance policy.
- Logs show the agent required regular human steering, initially issued warnings and training over months, and first recommended a formal warning rather than termination.
- Andon Market began with a $100,000 budget and is running at a loss, with the project’s bank balance reported at $61,186 after five months, highlighting operational risks of AI-run businesses.
- All employees are formally employed by Andon Labs with guaranteed pay and legal protections, and the company says the case raises questions about future workplace rules, accountability, and how more goal-driven models could change management.