A Carnegie Mellon University experiment using AI agents from Google, OpenAI, Anthropic, and Meta to staff a simulated software company yielded disastrous results. The AI struggled with basic tasks, demonstrating poor common sense, social skills, and internet navigation. Even the best-performing AI agent completed only 24% of its assignments at a prohibitively high cost. The study suggests that current AI is far from capable of replacing human workers in complex roles, dispelling fears of widespread job displacement by AI.

Prepared by Jonathan Pierce and reviewed by editorial team.