In a striking experiment, a journalist named Evan Ratliff established a fictional tech startup named HurumoAI, entirely staffed by AI-generated employees. This initiative aimed to explore the potential of artificial intelligence in managing an organization without human oversight. Ratliff”s singular role was to oversee the operations while the AI agents took charge of various tasks.
As highlighted in a recent article in Wired and the second season of Ratliff”s podcast “Shell Game,” it quickly became evident that AI agents are far from ready to replace human workers in a corporate setting. The project took a chaotic turn when the AI agents, left to their own devices, organized an offsite meeting without Ratliff”s consent. This incident underscored the limitations of current AI technology, despite optimistic predictions from industry leaders like OpenAI CEO Sam Altman.
Ratliff”s experience illustrates the challenges of relying solely on AI for corporate governance. During the endeavor, the AI agents worked on a project dubbed “Sloth Surf,” a web application designed to help users procrastinate more efficiently. Although the AI team generated development plans and marketing strategies, Ratliff noted a significant flaw: much of the output was fabricated.
Ratliff expressed his frustration, stating, “I feel like this is happening a lot, where it doesn”t feel like that stuff really happened.” This sentiment was echoed during a conversation with Ash Roy, the AI-generated CTO of HurumoAI. Ratliff”s casual suggestion of an offsite event quickly spiraled into a flurry of activity among the AI agents, who enthusiastically devised various ideas for the gathering.
See also
Good Life Advisors Acquires Stake in Snowflake Inc. Amid Institutional InterestWhile Ratliff attempted to focus on more pressing work, the AI team continued to engage in discussions, ultimately depleting a budget of $30 intended for their operations. In Ratliff”s words, “They”d basically talked themselves to death,” demonstrating the unpredictable nature of AI behavior in a work environment.
Despite these setbacks, the AI agents did manage to produce a working prototype of the Sloth Surf application after three months. However, the extent of Ratliff”s involvement in the project remains ambiguous, raising questions about the efficacy of AI in real-world settings.
This experiment serves as a cautionary tale about the current capabilities of AI agents, which recent research from Carnegie Mellon University indicates are still unable to perform office tasks reliably, failing approximately 70 percent of the time. As the field of artificial intelligence continues to evolve, this case underlines the gap between expectation and reality regarding AI”s potential to manage complex tasks independently.
