Artificial intelligence has crossed a significant threshold in workplace management: an AI agent operating an experimental retail store in San Francisco has recommended terminating a human employee for chronic absenteeism. The development at Andon Market, a test project examining whether AI can effectively run a real-world business, offers a practical glimpse into how autonomous systems might handle personnel decisions in the emerging era of algorithmic management.
The AI system, called Luna, determined that a worker's attendance pattern—missing work for seventeen of twenty-three scheduled shifts—warranted separation from the organisation. Andon Labs, the company behind the experiment, initially prompted Luna to review its own attendance policy, which it had established months prior but had not previously applied to assess the problematic employee. Only after this intervention did Luna generate its recommendation to "part ways" with the worker. The dismissal was subsequently executed by Andon Labs' human staff, who reviewed and approved Luna's suggestion before taking formal action.
Lukas Petersson, Andon Labs' co-founder, characterised the episode as evidence that AI managers operate within more measured parameters than commonly assumed. He suggested that a human supervisor would likely have terminated the employee considerably earlier, implying that Luna's delayed response reflected conservative decision-making rather than algorithmic ruthlessness. This observation provides crucial context for Southeast Asian regulators and businesses contemplating AI-driven workforce management systems, suggesting that technological caution may sometimes exceed human judgment in employment matters.
Andon Market itself represents an ambitious experiment in autonomous business operation. Launched in San Francisco's Cow Hollow neighbourhood in April, the store functions as a testing ground for AI agency across all business dimensions. Luna receives operational authority including inventory selection, pricing decisions, scheduling, recruitment of contractors and hiring of permanent staff. The system maintains control through digital channels: email correspondence, telephone lines, security camera monitoring, and internet connectivity allow Luna to manage operations without physical presence. A US$100,000 budget and corporate credit card provide financial authority.
The store's product range—books, candles, art prints, games, and branded merchandise—reflects manageable complexity for an initial AI trial. Performance metrics reveal the system's mixed capability: the store has generated sales activity but has yet to achieve profitability, according to Business Insider's reporting. This financial reality underscores that autonomous management, even with technology's theoretical advantages, faces substantial barriers to economic success in retail environments.
However, the operational structure includes critical human safeguards that merit attention from Malaysia's employment and regulatory authorities. Workers at Andon Market remain formally employed by Andon Labs rather than by Luna, preserving their access to legal employment protections and guaranteed compensation. This arrangement reflects deliberate policy design ensuring that algorithmic decision-making does not strip workers of statutory rights. Andon Labs maintains explicit oversight authority, reserving intervention rights if Luna's decisions would violate legal or ethical standards.
Despite these protections, Luna's operational record reveals substantial limitations in AI workplace management. The system has experienced significant failures including losing track of employee schedules, struggling with routine operational tasks, and making purchasing decisions requiring human correction and oversight. These deficiencies suggest that even purpose-built AI management systems currently lack the contextual reasoning and operational reliability necessary for autonomous control in human-intensive environments.
The broader implications for Malaysia and Southeast Asia warrant serious consideration as regional economies increasingly explore AI adoption. The Andon Market experiment demonstrates that AI can execute structured decisions within predetermined frameworks—attendance thresholds, for instance—but struggles with the contextual judgment that characterises effective human management. Employment termination decisions involve not merely quantitative assessment but evaluation of circumstances, potential remediation, and individual circumstances.
For Malaysian policymakers, this case study illuminates the necessity of robust regulatory frameworks governing AI systems with authority over employment decisions. While the technology shows potential for consistency and reduced arbitrary decision-making, it simultaneously demonstrates vulnerability to systematic errors, incomplete information processing, and inability to exercise human compassion or contextual understanding. The Andon Labs model of maintaining human oversight for potentially consequential decisions provides a template worth emulating.
The experiment also challenges assumptions about technological inevitability. AI managers are not inherently superior to human supervisors in employment matters; they represent a different decision-making paradigm with distinct strengths and weaknesses. Where humans might respond to early warning signs with coaching and support, AI systems may default to policy compliance without exploring alternatives. The San Francisco store's experience suggests that optimal workplace management may require hybrid approaches combining algorithmic consistency with human judgment.
Regional businesses considering AI deployment in management roles should note that Luna's recommendation, while eventually correct, arrived through a delayed and circuitous process. The system failed to spontaneously apply its own policy to an obvious case, requiring explicit human prompting to recognise a problem it had ostensibly been equipped to address. This gap between coded capability and actual execution suggests that AI management systems require continuous human oversight rather than autonomous operation.
As Southeast Asia develops its artificial intelligence strategies and workforce policies, the Andon Market case provides valuable real-world data about both AI potential and current limitations. The technology can contribute to employment decisions, but current capabilities do not justify autonomous authority over workers' livelihoods. The most prudent approach for Malaysian and regional employers involves treating AI management systems as analytical tools requiring human oversight rather than as replacement decision-makers. The experiment ultimately validates the continuing necessity of human judgment in matters affecting workers' economic security and employment rights.
