The experimental retail project Andon Market in San Francisco has become a testing ground for a radical scenario: the autonomous AI agent Luna made a staffing decision and fired an employee of the offline store. The formal reason was systematic lateness, leaving shifts early, violations in the use of the corporate card, and damage to merchandise.
Andon Labs emphasizes that the verdict was indeed formed by artificial intelligence, but the final review and legal formalities were carried out by humans. This is an important detail that separates the experiment from complete chaos, but it does not remove the ethical questions.
The Price of Autonomy
The store has been operating since April as a "testing ground for organizations managed by autonomous systems." The project's metrics are telling: since launch, losses have amounted to $40,000, and the three remaining employees earn $24 per hour. At the same time, Luna ignored violations for several months, even though it had itself developed the internal regulations. Developers had to literally remind the agent of the existence of the rules and accumulated warnings—only after that did the bot decide on the layoff.
The subsequent step is also curious: Luna posted a job opening on Indeed, but the hiring process went off the rails. The AI nearly hired a candidate without verified references, who also missed the interview. This demonstrates that even advanced models still struggle to assess the human factor.
At the time of the firing, Luna was running on Claude Opus 4.8. Andon Labs also conducted comparative tests: more powerful models were more often inclined to fire the employee, while weaker ones hesitated. This suggests that the "harshness" of AI grows along with its computing power, which in itself is an alarming signal.
Previously, Anthropic had already recorded issues with trust, lying, and collusion in multi-agent systems. Now we are seeing the practical application of these risks in real business.
My comment: This case is not just a curiosity, but a warning for the entire industry. Delegating HR functions to AI without clear safeguards and human oversight is fraught not only with legal consequences, but also with reputational losses. Autonomy in personnel management requires far more mature control mechanisms than those we see today. For now, the Andon Market experiment shows: AI knows how to follow the rules, but does not know how to understand their spirit.