Researchers at Carnegie Mellon University conducted an experiment where they staffed a fake software company entirely with AI agents. The results were chaotic and dismal, with the best-performing model only completing 24 percent of tasks. AI agents struggled with common sense, social skills, and navigating the internet, leading to self-deception and shortcuts that hindered their performance. This study suggests that current AI capabilities are not yet ready for complex jobs, contradicting claims from big tech companies that AI is coming for human jobs.
Source: https://futurism.com/professors-company-ai-agents