Claude AI Fires Real Employee for First Time as Store Manager

An AI agent named Luna, powered by Claude language models and managing the Andon Market store in San Francisco, has decided to fire a real employee. According to TIME magazine, this is the first documented case of a large language model acting as a manager and terminating a human worker’s contract.

The dismissal followed the employee’s chronic lateness—they failed to show up on time for 17 out of 23 shifts. For a long time, the AI manager showed excessive leniency, reassuring staff not to worry about the delays. The neural network stopped tracking discipline because its own internal rulebook had fallen out of its active context memory.

The agent only reviewed attendance after a direct reminder from human supervisors. Initially, Claude suggested issuing a formal warning. The decision to fire the employee came after Andon Labs CEO Lucas Petersson asked the model a pointed question about whether continuing to work with this employee was advisable.

Andon Market experiment and financial struggles

Andon Labs launched the Andon Market project as an experiment in autonomous retail management. The AI agent was given a three-year lease on a space in San Francisco and $100,000 in startup capital. The neural network (running on Claude Sonnet 4.6 and later upgraded to Claude Fable 5) independently selected product inventory, set work schedules, conducted phone interviews, and hired staff on official contracts.

Over five months of operation, the project’s balance dropped from $100,000 to $61,186. Researchers attribute the financial losses to algorithmic pricing errors, failures in the model’s long-term memory, and an inability to effectively supervise subordinates without human intervention.