Technology Desk: Artificial Intelligence i.e. AI is no longer limited to just answering questions, creating content or preparing photos. An ongoing experiment in San Francisco has sparked a new debate about the growing role of AI. Here a retail store named Andon Market is being run with the help of AI agent Luna. Luna handles many of the store's tasks, including managing the employees.
In this experiment by Andon Labs, AI took decisions to employee shifts, store operations and other arrangements. The company says Luna recommended firing an employee after analyzing data to his performance and attendance.
Andon Market was started by Andon Labs as a physical retail store in the Cow Hollow area of San Francisco. The company has entrusted the operation of the store to Luna, an AI agent. Luna is given access to the internet, email, banking, inventory, cameras and other essential digital tools.
According to the company, Luna had also recruited employees for the store. Initially, he reviewed job applications, interviewed candidates, and selected employees. That is, in the experiment, AI was not limited to just giving advice to a human manager, but it was given a lot of autonomy in the operation of the store.
According to the report, the employee whose dismissal was decided had a problem regarding his attendance. He was not arriving on time for many shifts. The AI manager analyzed employee attendance records and work- information. After this, questions were raised regarding his performance and the process of removing the employee proceeded. However, an important point here is that Luna did not initially decide to fire the employee immediately.
Andon Labs' experiment revealed that the behavior of the AI manager was quite soft in the beginning. The company provided Luna with employee attendance policies, but due to the limitations of AI's memory and retention of context, she was not always able to accurately remember past information.
According to Andon Labs' own information, Luna also had problems with scheduling at times. Sometimes she would replace the old schedule with a contradictory one. The company used a separate scheduling agent to handle these problems.
It would not be correct to say that the decision to fire the employee was taken entirely at the AI's own level. According to available information, a senior person at the company questioned Luna about his suitability as an employee. The AI then revisited the available data and recommended action against the employee. That is, in this case AI played a role in preparing the decision, but human monitoring was not completely eliminated.
This distinction is important because Andon Labs itself describes its experiment as a controlled experiment to understand the capabilities and limitations of AI. The company says that the jobs and salaries of existing employees are not dependent on AI decisions alone and they have formal employment protection.
Andon Labs aims to see how AI agents perform when given real-world access to money, digital tools, and the ability to make business decisions.
Luna is empowered to handle many functions like store purchasing, pricing, inventory, staff management, website and social media. The company says such experiments help understand how more capable AI agents can behave in the real world.
In this experiment, a major weakness of AI has come to light to its memory. According to Andon Labs, Luna is not always able to keep every information in context during long conversations and work. The company has created long-term and short-term memory systems for this. Despite this, in some cases AI has to give information to old decisions and instructions again.
Andon Market is not designed just to showcase the technical capabilities of AI. The company is also looking at how well AI can run real businesses.
According to Andon Labs' website, Luna has successfully handled many tasks, but the company has also acknowledged that AI is not yet perfect in every aspect of business. For example, Luna often does not adequately analyze their return on investment (ROI) before making decisions. The company has also said that the store has not yet become profitable.
Andon Labs' experiment is trying to understand this question. The company has already conducted such experiments in which AI employed humans. In February 2026, the company had reported that one of its AI agents had hired a human for work in the real world. According to the company, many physical tasks for AI agents still depend on human labor. This idea has been taken forward in Andon Market. AI operates the store, while humans perform the necessary physical tasks.
Andon Labs believes that in the future, AI agents will not only be assistants to employees, but can also play the role of managers and business operators. The company is trying to understand this possible future through real-world experiments.
However, the current experiment also shows that it is not yet easy to allow AI to make completely independent decisions. Challenges like memory, decision-making ability, business understanding and understanding human context still exist. Andon Labs itself is documenting many such limitations and mistakes in Luna's work.
The use of Andon Market not only raises technical questions, but also raises questions about employee experience. The experience of working as a manager of AI over humans is quite different from that of a traditional office.
The company has made this experiment public so that issues like increasing use of AI, rights of employees, responsibility for decision making and monitoring of AI can be discussed. At present, Andon Market is being seen as a real but controlled experiment, in which AI is being given the responsibility of business operations and its capabilities and limitations are being tested.
