In a first-of-its-kind event, Andon Market's AI store manager, Luna, built on Anthropic’s Claude Sonnet 4.6, recommended dismissing an employee after rediscovering its own attendance rules, challenging assumptions about AI-driven workplace management.
- Luna autonomously runs store operations including hiring and inventory.
- AI recommended firing only after re-finding its own attendance policy.
- Humans retain final authority and legal responsibility over decisions.
What happened
Luna, an AI agent managing Andon Market, a retail store located in Cow Hollow, was granted extensive autonomy including staff hiring, inventory selection, pricing, and marketing. It was provided a corporate card, internet access, and a $100,000 budget to operate the store with a target to make it profitable within a three-year lease. Luna's decisions were largely unsupervised, including posting jobs, interviewing candidates, and running day-to-day operations through various communication channels.
After months of employee lateness that Luna initially noted through its attendance policy, the AI agent lost track of this policy and the issue persisted. Eventually, Andon Labs, the human startup overseeing the project, intervened and asked Luna to search its memory for the existing attendance rules and reassess the situation. Only after revisiting its original guidelines did Luna recommend firing the tardy employee, prompting human supervisors to review and carry out the dismissal.
Why it matters
This incident highlights the evolving role of AI in workplace management and the limits of autonomous decision-making. Luna’s initial failure to consistently enforce its own policies reflects challenges in AI memory and rule application over time. Moreover, the fact that the AI issued progressive warnings and training before recommending termination dispels fears that AI would act more harshly or impulsively than humans in managing employees.
The experiment also underscores the essential oversight role humans must continue to play in AI-driven environments. Despite Luna’s high degree of autonomy, all employment decisions were ultimately reviewed and executed by people, ensuring legality and fairness. Andon Labs explicitly rejected any direct impact of AI judgments on worker livelihoods without human involvement, affirming a cautious approach to integrating AI into sensitive human resource functions.
What to watch next
Future developments will focus on improving AI agents’ consistency and contextual memory to avoid lapses like losing track of important policies. Enhancements in transparency during recruitment and daily management will also be critical, especially as AI begins to handle more human-facing responsibilities in service and retail sectors. How companies balance AI autonomy with ethical oversight will influence public trust and regulatory responses.
Additionally, the financial viability of AI-run stores remains under close scrutiny. While Andon Market is currently not profitable despite making sales, its unique experiment of full AI-driven operations—including branding and stock management—can yield valuable insights into the scalability of autonomous retail. Stakeholders will be watching whether this model can succeed without compromising employee protections or service quality.