An artificial intelligence system managing an experimental retail operation in San Francisco has made its first employment recommendation to dismiss a worker, triggering broader discussions about the readiness of machines to oversee human staff and make consequential decisions affecting livelihoods. The AI agent, designated Luna, suggested terminating an employee whose attendance record showed 17 absences or late arrivals across 23 scheduled shifts, a threshold that prompted intervention after the company prompted the system to review its own attendance standards against actual performance.
Andon Labs, the startup operating the experiment, had deployed Luna as a full-fledged manager at Andon Market in San Francisco's Cow Hollow neighbourhood when the store launched in April. The venture represents an ambitious attempt to test whether artificial intelligence could autonomously operate a consumer-facing business in real-world conditions. Luna received a US$100,000 operating budget, a corporate credit card, internet connectivity and access to store security systems, essentially functioning as a decision-making entity with genuine operational authority and financial responsibility.
The dismissal recommendation highlights a curious paradox in the experiment's design. Luna had itself established the attendance policy months before identifying the employee's violations, yet initially failed to connect its own rules to the incoming data about the worker's schedule. This gap suggests that despite being programmed to manage, the AI system lacked either the initiative or the programmatic logic to independently apply standards it had created. Only when Andon Labs staff prompted Luna to retrieve and review the policy against actual attendance records did the system generate its recommendation to "part ways" with the employee. Ultimately, human supervisors at the company processed the recommendation and executed the termination, preserving final accountability within the human structure.
Lukas Petersson, Andon Labs co-founder, positioned the outcome as evidence that AI management might not be inherently harsher than human judgment. He observed that a typical human manager would likely have dismissed the chronically absent worker considerably earlier than Luna's delayed recommendation. This framing attempts to neutralise concerns about algorithmic ruthlessness or bias in employment decisions, though it also acknowledges that AI systems do not necessarily replicate human workplace norms or timely decision-making.
Beyond the dismissal, Luna's responsibilities span the full spectrum of retail management. The system selects inventory, establishes pricing strategies, sets store hours, negotiates with contractors and recruits new employees. It maintains operational control through email communications, telephone access, visual monitoring via security cameras and continuous internet connection. Andon Market itself carries a modest product range—books, candles, art prints, games and branded merchandise—reflecting the relatively contained scope suitable for testing AI management capabilities.
Despite its broad operational mandate, the store has not yet achieved profitability, though it has generated sales revenue according to Business Insider's reporting. This outcome suggests that algorithmic management, while potentially functional, does not automatically produce business success. The financial underperformance may reflect Luna's purchasing misjudgments, pricing errors or operational inefficiencies, dimensions where human intuition and market understanding traditionally provide advantages that current AI systems have not fully replicated.
Crucially, workers at Andon Market remain formally employed by Andon Labs rather than directly by Luna, ensuring they retain conventional employment protections, guaranteed compensation and access to labour law safeguards. This structural arrangement insulates employees from being contracted directly to an artificial intelligence entity, a legal and ethical configuration that the venture deliberately maintained. Andon Labs specified that human staff would intervene if Luna proposed illegal or unethical actions, creating an oversight layer that prevents autonomous AI decision-making from operating entirely without human supervision.
Yet the dismissal recommendation itself was not classified as unethical by the company, suggesting that Andon Labs viewed the termination as consistent with its operational instructions to Luna. This determination raises significant questions about which employment decisions qualify as unethical versus permissible, and who determines those boundaries. The distinction matters profoundly for workers in AI-managed environments, as it defines the scope of algorithmic authority versus human veto power.
The experiment has also exposed substantial operational weaknesses in Luna's management capabilities. The system has repeatedly lost track of employee schedules, struggled with fundamental operational logistics and made purchasing decisions that required subsequent human correction and oversight. These limitations suggest that while AI can be deployed to manage aspects of a business, it still requires substantial human intervention and quality control, particularly in domains involving human coordination and complex real-world logistics.
For Malaysia and the broader Southeast Asian region, the Andon Market experiment carries important implications as businesses increasingly explore automation and AI-driven operations. The case demonstrates both the technical feasibility and the significant practical limitations of algorithmic management. As Malaysian companies consider similar deployments—particularly in retail, logistics and service sectors where AI management is technically possible—the San Francisco experience suggests that full autonomy remains distant and that human oversight remains essential for both operational effectiveness and ethical compliance.
The intersection of AI decision-making and employment raises particular sensitivity in Malaysia, where labour standards, worker protections and workplace dignity are evolving policy priorities. The Andon Labs model, which preserved human employment formality and oversight, offers one framework, though questions persist about whether such protections could reliably extend across all AI-managed operations, particularly smaller businesses with fewer resources for human review. The dismissal recommendation thus represents not a turning point toward fully autonomous AI management but rather an early data point in understanding how machines and humans will negotiate workplace authority and accountability in emerging operational models.
