AI Store Manager Luna Fires Worker After Losing Policy It Wrote
TL;DR. An AI agent managing a San Francisco store dismissed an employee for chronic lateness, but only after human intervention helped it recall its own attendance policy. - Luna, built on Anthropic's Claude Sonnet 4.6, lost track of a policy it created months prior. - Andon Labs, the developer, guided Luna to review its memory and assess the worker's fit. - Researchers noted a human boss would have acted faster, as Luna had issued warnings without contractual action.
- AI agent 'Luna' manages Andon Market in San Francisco.
- Luna, built on Anthropic's Claude Sonnet 4.6, fired a human worker.
- The AI agent had created its own attendance policy but then 'lost track' of it.
- Human intervention was required for Luna to recall and apply its own rules.
- The experiment highlights challenges in autonomous agent memory and consistent policy application.
- Luna had previously issued warnings and arranged training, showing a less 'ruthless' approach than expected.
Sources
- The AI store manager fired its first human — thenextweb.com