Andon Labs' artificial intelligence manager, Luna, has made its first employment recommendation to dismiss a worker at the experimental Andon Market in San Francisco's Cow Hollow neighbourhood, marking a significant milestone in the ongoing experiment to test whether machines can operate real-world businesses independently. The recommendation came after the employee failed to show up on time for 17 of 23 scheduled shifts, demonstrating what would normally constitute grounds for immediate termination under most workplace standards.
The circumstances surrounding the dismissal reveal an intriguing paradox about artificial intelligence decision-making. Luna had itself created the attendance policy months earlier but initially did not connect the dots between its own guidelines and the employee's poor record. Only when Andon Labs staff prompted the system to review its own policy did Luna conduct the reassessment that led to the "parting ways" recommendation. This intervention by human operators underscores that even when AI systems have clear criteria and authority, they do not automatically apply rules with the consistency one might expect from a purely algorithmic approach.
Lukas Petersson, co-founder of Andon Labs, characterised the outcome as evidence that AI managers are not inherently more harsh than their human counterparts. He argued that a human supervisor would likely have terminated the employee sooner, suggesting that the delay in Luna's response actually demonstrates greater restraint rather than ruthlessness. This observation challenges assumptions about artificial intelligence replacing humans with cold, emotionless efficiency in workplace management. The implication for Southeast Asian businesses watching AI adoption trends is that implementation will require careful calibration; automated systems do not automatically solve management problems but rather introduce different operational patterns.
Andon Market itself represents an ambitious test of artificial intelligence capability in a real commercial environment. Launched in April with a US$100,000 budget, a corporate credit card and internet connectivity, Luna operates through email communication, phone lines, security cameras and digital access to manage all aspects of the business. The system independently selects merchandise inventory, determines pricing strategies, sets store operating hours, hires contract workers and recruits permanent staff. This breadth of responsibility distinguishes the experiment from most existing AI applications, which typically automate specific tasks within human-supervised structures.
The store's product range—books, candles, art prints, games and branded merchandise—was chosen to test Luna's ability to curate inventory and make commercial decisions in a competitive market. Results have been mixed. According to Business Insider, Andon Market has generated sales revenue but has not yet achieved profitability. This underperformance is significant because it demonstrates that even with AI management overhead removed, achieving sustainable business operations involves complexities that pure algorithmic optimisation has difficulty navigating. For Malaysian retail businesses considering automation investments, the lesson is that AI excels at specific, well-defined tasks but struggles with the holistic judgment required in retail curation.
Critically, Andon Labs maintains human oversight mechanisms. All workers employed at the store are formally hired by Andon Labs rather than by Luna, ensuring they retain standard employment protections, guaranteed compensation and legal rights. The company has declared that human staff will intervene if Luna's decisions breach legal or ethical boundaries, though the company determined the dismissal fell within appropriate parameters. This hybrid governance model reflects growing recognition in the technology sector that AI autonomy requires guardrails, particularly when decisions affect human employment and welfare.
The experiment has already revealed significant operational limitations in Luna's management approach. The system has previously lost track of employee scheduling information, struggled to execute routine operational tasks and made purchasing decisions requiring human correction. These gaps suggest that while AI can process data systematically, translating that capability into reliable business management remains a developing frontier. The scheduling failures are particularly notable because workforce coordination is arguably a core function of any manager, yet Luna has demonstrated vulnerability in this area despite having digital access to relevant information.
For the broader Southeast Asian business context, the Andon Market experiment carries important implications. As companies across Malaysia, Singapore, Indonesia and Thailand increasingly adopt automation technologies, they face fundamental questions about whether AI can genuinely replace human judgment in strategic roles. The San Francisco case demonstrates that even with significant investment and sophisticated systems, AI management generates unpredictable outcomes and requires substantial human oversight. This reality should inform how regional companies approach automation roadmaps, particularly in roles involving employment decisions that carry legal and ethical weight.
The dismissal itself, while technically a machine's first employment recommendation, was ultimately a human decision. Andon Labs reviewed Luna's analysis and chose to implement the termination. This final layer of human authority suggests that AI deployment in management roles may evolve into a more distributed model where machines provide analysis and recommendations but humans retain decision-making responsibility, particularly for sensitive personnel matters. Such an approach could appeal to Malaysian employers concerned about accountability and regulatory compliance.
The intersection of artificial intelligence capability and human employment raises emerging questions that regulators and businesses throughout Southeast Asia will eventually need to address. If AI systems are recommending dismissals, who bears responsibility for wrongful termination claims? How should employment law adapt to accommodate AI-assisted decision-making? Are there sectors or company sizes where such arrangements are more appropriate than others? These questions lack clear answers currently, but they will become increasingly pressing as AI adoption accelerates.
Andon Labs' willingness to conduct this experiment publicly and share findings about both successes and failures provides valuable data for the technology and business sectors. The result—showing that AI can identify violations of policies it created but requires human prompting to do so—suggests that artificial intelligence in management works best not as autonomous decision-maker but as analytical tool supporting human supervisors. For Malaysian companies piloting AI solutions, this framing may prove more realistic and implementable than visions of fully autonomous management systems.
