Artificial intelligence ran the store, the result was surprising! $100,000 melted in 5 months

The unconventional experiment conducted in San Francisco, USA, revealed striking results about AI's capacity to run a business. In the test where Anthropic's Claude was tasked with managing a retail store, the AI decided to fire an employee while the company's financial performance also failed to meet expectations.

In the experiment run by Andon Labs, the management of a store staffed by real employees was largely left to Claude. The goal was to observe whether AI agents could sustain a business's daily operations without human intervention.

Claude Fired an Employee

One of the experiment's most striking moments was Claude proposing the termination of an employee's contract.

The AI detected that an employee was late on 17 of 23 shifts and consequently decided to terminate them. However, it was noted that Claude took a long time to notice the employee's attendance issue.

It was stated that this situation was linked to limitations in the AI's memory capacity.

Things Changed When the Management Guide Was Deleted

The disappearance of the employee management guide prepared by Claude during the experiment also drew attention, as the system's limited working memory caused it to vanish.

After the guide vanished, it was observed that Claude's approach toward employees had changed. The AI began treating late‑arriving staff more leniently and, in some cases, reassuring employees not to worry.

Therefore, the fact that an employee was late for most of their shifts yet the issue went unnoticed for a long time was considered one of the experiment's major problems.

The Termination Decision Was Not Completely Independent

It also emerged that Claude's decision to terminate employment was not entirely independent of human intervention.

It was noted that the Andon Labs team conducting the experiment intervened with Claude from time to time during the process and asked guiding questions in some decision-making.

In the first stage, Claude suggested issuing an official warning instead of terminating the employee. After the company mentioned that verbal discussions had previously taken place with staff and following the questions directed at Claude, the AI this time recommended terminating the employment contract.

100,000 Dollar Capital Diminished

The financial results of the experiment, however, raised questions about the AI's commercial management skills.

The store managed by Claude began with a $100,000 capital, which fell to $61,000 after five months. Thus the business lost about $39,000.

Andon Labs CEO Lukas Petersson assessed that Claude's overly soft management style and shortcomings in commercial intuition had adversely affected the store's performance.

Petersson said that AI models could be developed to act more decisively toward their goals, but that this could create a new scenario with unwanted outcomes from a human perspective.

The Role of Artificial Intelligence in the Workplace Is Being Reexamined.

The Claude experiment has brought the debate back to the fore about AI being used not only in desk jobs but also in areas that directly affect human life, such as employee management and commercial decision-making.

In the United States, some companies' use of AI‑powered systems in employee evaluation and layoff processes is also sparking legal debate.

Meta's current and former 26 employees filed a lawsuit in California alleging that the company used AI algorithms in its mass‑layoff procedures. The plaintiffs argued that the automated systems unfairly targeted disabled workers and staff on legal health or family leave.

Can Jobs Lost to Artificial Intelligence Be Reclaimed?

Whether the workforce transformation driven by artificial intelligence will be permanent is a separate debate.

According to some analyses, companies may be forced to reverse those decisions if they encounter operational problems after replacing human employees with AI systems.

Forrester analysts predict that more than half of the layoffs caused by artificial intelligence can be reversed in the future. Gartner data also indicates that a significant portion of companies shrinking their customer support teams due to AI integration may be forced to rehire some employees by 2027.

The picture that emerged from Claude's store experiment, however, shows that artificial intelligence is neither entirely unsuccessful nor ready to replace human managers in business management. The experiment, as technology advances,"Can artificial intelligence manage a business?"alongside the question", "Who will be responsible for the outcomes if it is managed?"also brings the question to the agenda.