AI Manager Recommends Firing Employee at Experimental Store

An AI manager named Luna recommended the dismissal of a human employee at Andon Market, an experimental retail store in San Francisco. The employee was cited for workplace issues and arriving late for 17 of 23 shifts. Although Luna had established an attendance policy, the AI failed to enforce it until researchers at Andon Labs prompted the system to review its policies and assess the worker's fit.

Lukas Petersson, cofounder of Andon Labs, noted that the experiment highlights a tendency for AI agents to fail to act without direct human prompts. While Luna recommended the firing, the human researchers at Andon Labs reviewed the case and executed the final dismissal. Petersson stated that the firing was warranted and suggested that human managers might have acted even sooner.

Luna was built using Claude models from Anthropic and was granted a $100,000 budget and a corporate credit card to launch the store and generate profit. Petersson believes the experiment is a precursor to a future where AI systems will run companies and act as employers of humans.