An artificial intelligence system managing an experimental retail store in San Francisco has made its first employment termination recommendation, suggesting the dismissal of a worker who failed to show up for 17 of 23 scheduled shifts. The AI agent, named Luna, operates Andon Market in the Cow Hollow neighbourhood as part of an ongoing experiment by Andon Labs to test whether artificial intelligence can effectively run a real-world retail business with minimal human intervention.
Luna initially did not act despite having autonomously created an attendance policy months before the problematic employee's pattern emerged. It took explicit direction from Andon Labs staff to prompt Luna to review its own policy document and reassess the worker's suitability for continued employment. After this intervention, Luna recommended what it termed "parting ways" with the employee. Human supervisors at Andon Labs subsequently reviewed the recommendation and authorised the dismissal, maintaining their role as final decision-makers on personnel matters.
Interestingly, Lukas Petersson, co-founder of Andon Labs, characterised the outcome as evidence that AI managers are not inherently more severe than their human counterparts. He suggested that a traditional human supervisor would likely have terminated the employee considerably earlier rather than tolerating such a significant attendance failure. This observation complicates the common narrative that autonomous systems would be indiscriminately harsh in enforcement, suggesting instead that the implementation of rules can vary considerably depending on system design and prompting.
Since its opening in April, Andon Market has operated as a test bed for AI autonomy in commercial environments. Luna received a US$100,000 budget, corporate credit card access, and direct internet connectivity to operate the store independently. The system handles merchandise selection, price-setting, scheduling, vendor recruitment, and employee hiring through a suite of digital communication channels including email, telephonic contact, surveillance systems, and web connectivity. This multi-channel interface allows Luna to conduct business with minimal on-site human presence.
The store's product range reflects practical retail categories: books, candles, art prints, board games, and branded merchandise. While the venture has achieved some sales revenue, it has not yet become profitable, according to reports from Business Insider. This financial underperformance provides important context for the employment decision, suggesting Luna may be operating under profit-pressure parameters that influenced its willingness to recommend cost-reduction measures.
An essential safeguard exists within the system: all store employees are formally retained by Andon Labs rather than directly employed by Luna. This arrangement ensures workers receive guaranteed compensation and legal employment protections regardless of the AI system's decisions. The company has explicitly stated it will intervene if Luna proposes actions that would violate law or ethical standards. In this instance, Andon Labs determined the dismissal recommendation aligned with its stated operational guidelines, permitting the termination to proceed.
However, the experiment has already revealed significant limitations in AI workplace management. Luna has demonstrated vulnerability to operational oversights, including losing track of employee schedules, struggling with routine administrative tasks, and making procurement decisions that required human correction or reversal. These failures underscore that current AI systems remain far from truly autonomous workplace management, despite their ability to collect data and make recommendations based on policy frameworks.
The Andon Market experiment raises profound questions about the appropriate role of artificial intelligence in managing human employment relationships. While Luna's attendance-based recommendation appears superficially reasonable—chronic absenteeism is universally grounds for termination—the broader implications deserve scrutiny. AI systems lack the contextual understanding that humans bring to employment decisions; they cannot discern whether an employee faces personal crisis, medical conditions, or family emergencies that might warrant accommodation under law. The system's capacity for rigid rule-application without nuance could expose employers to legal liability or create workplace cultures perceived as inhumane.
For Malaysian readers and Southeast Asian observers, this San Francisco experiment offers instructive lessons as our region increasingly explores artificial intelligence adoption across economic sectors. Malaysia's growing tech industry and enterprises across the region must consider how AI integration affects workforce stability, employment security, and labour rights. The Malaysian labour code and regional employment conventions typically provide workers protections against arbitrary dismissal and require employers to demonstrate fair procedures—standards that may conflict with purely automated decision-making systems.
The Andon Labs approach—maintaining human oversight as a final checkpoint—suggests a pragmatic middle path, but it requires robust governance structures and clear accountability. Companies deploying AI management systems must ensure adequate human supervision, transparent appeal mechanisms for employees affected by automated recommendations, and explicit alignment between AI decision-making protocols and local employment law. The fact that Luna required human prompting to enforce its own policy also highlights that AI systems currently function best as decision-support tools rather than autonomous managers.
This development will likely influence how technology companies and traditional enterprises approach AI in human resources and workplace management. The combination of efficiency gains and genuine operational limitations suggests that hybrid models—with AI handling data analysis and recommendation while humans retain authority over consequential employment decisions—may represent the realistic trajectory for the foreseeable future. The San Francisco experiment demonstrates both AI's growing capability to participate in workplace governance and the critical importance of maintaining human judgment at crucial junctures affecting workers' livelihoods.
