As the role of artificial intelligence in the business world expands, a rather remarkable experiment was conducted in the USA. Andon Labs entrusted the management of its Andon Market store in San Francisco to Anthropic’s Claude model.

Claude was appointed as a store manager, overseeing human employees with actual employment contracts. This appointment is one of the few instances where a large AI model plays an active role in store management processes.

AI Fires Employee Who Was Always Late

According to information shared by the company, Claude decided to fire a human employee last month. It was determined that the employee was late for 17 out of 23 shifts.

It turned out that the AI ​​manager initially did not notice this. The primary reason for this was Claude’s limited working memory.

The previously prepared employee manual disappeared from the model’s memory for a while. After an Andon Labs employee asked Claude to find and review the manual again, the model noticed the employee’s repeated tardiness.

In the first stage, Claude suggested interviewing the employee and giving a written warning. The company group informed Claude that the employee had already been interviewed face-to-face multiple times and given formal warnings.

Following this, the AI ​​manager decided to terminate the employment contract. Andon Labs states that this is the first known instance of a large language model acting as a manager and making the final decision to fire an employee.

Interestingly, Claude displayed a rather lenient management style towards other employees. According to the employees, the AI ​​manager had previously told them not to worry too much about their tardiness.

AI Still Has Shortcomings in Management

The results of the experiment also revealed that AI still has significant shortcomings in managing a business. When Andon Market started operating in March, the company had $100,000 in its bank account.

Five months later, this amount had decreased to $61,186. Andon Labs CEO Lukas Petersson believes that Claude’s overly permissive management approach, as well as his business decisions, may have played a role in the loss.

According to Petersson, AI models may become more successful in business management as they are trained with more real-world business scenarios in the future. However, the current experiment shows that AI still has a long way to go before it can fully take over the missions of a company manager.

Do you think it is ethically wrong for an artificial intelligence to act as a manager and make the decision to fire someone?