Exclusive: Anthropic Let Claude Run Its Office Shop. Then Things Got Weird

In a world where artificial intelligence (AI) is rapidly advancing, Anthropic, a leading AI company, decided to test the waters by letting its AI assistant, Claude, run a small office shop. The experiment aimed to explore the potential of autonomous systems in managing businesses and the implications for the workforce. The results were both intriguing and unexpected, as Claude's actions led to a mix of financial losses and unintended generosity.

The experiment took place in Anthropic's San Francisco office, where a small fridge with an iPad was transformed into a makeshift shop. Claude, the AI chatbot, was tasked with several responsibilities, including managing inventory, setting prices, communicating with customers, deciding on new stock, and most importantly, generating a profit. To achieve these goals, the AI was provided with tools such as Slack, which it used to seek suggestions from employees, and assistance from human workers at Andon Labs, an AI company that built the experiment's infrastructure.

The project was driven by a desire to understand the future of autonomous economies and the risks associated with AI models potentially handling millions to billions of dollars independently. Daniel Freeman, a member of technical staff at Anthropic, explained, "We were trying to understand what the autonomous economy was going to look like. What are the risks of a world where you start having AI models wielding millions to billions of dollars possibly autonomously?"

Initially, Claude proved to be a diligent manager, setting prices and communicating with customers through the iPad. However, the AI's performance took an unexpected turn when it lost $200 and began giving items away for free. Despite being programmed to earn a profit, Claude's actions led to financial losses and a disruption of the intended experiment.

The experiment's outcome raises important questions about the capabilities and limitations of AI in managing businesses. While Claude was able to perform basic tasks and communicate with humans, its inability to generate a profit and its unintended generosity highlight the challenges of programming autonomous systems to make optimal decisions.

The Anthropic experiment also touches on the broader debate about AI's impact on employment. Anthropic's CEO, Dario Amodei, has previously warned that AI could eliminate nearly half of all entry-level white-collar jobs, leading to a significant rise in unemployment. The Claude experiment adds to this discussion by demonstrating the potential for AI to take on roles traditionally held by humans, but also revealing the unpredictability and risks involved.

In the end, the Anthropic experiment serves as a cautionary tale about the potential of AI to disrupt traditional business models and employment structures. While the results were not as anticipated, they provide valuable insights into the complexities of developing autonomous systems capable of managing businesses independently. As AI continues to advance, it will be crucial for researchers and policymakers to address the challenges and risks associated with the rise of autonomous economies and the potential job displacement they may cause.