AI Employees Fail Miserably in Fake Company Experiment
Will artificial intelligence replace your job soon? A new experiment suggests not. Researchers built a virtual company run entirely by AI. The results were surprisingly bad.
How Did The AI Company Work?
Scientists from Carnegie Mellon University created this test. They staffed the fake company with AI agents from top models. For example, they used Claude, GPT-4o, Gemini, and others. Each AI received a specific job role. Some acted as project managers or software engineers. Another platform even simulated human coworkers for the AIs to contact.
Shockingly High Failure Rates
The AI “employees” received real workplace tasks. Their assignments ranged from data analysis to giving virtual office tours. So, how did they perform? Frankly, they failed. The agents could not complete over three-quarters of their assignments. Claude 3.5 Sonnet was the best performer. However, it only fully finished 24% of its tasks. Most other models completed less than 10%. Therefore, total autonomy is still a distant goal.
Why Did The AI Struggle So Much?
The agents failed for several clear reasons. Often, they misunderstood simple instructions. For instance, some did not recognize a “.docx” file as a Word document. They also struggled with basic social reasoning and web browsing. Pop-up windows frequently confused them. As a result, they would skip steps and wrongly assume success.
What Does This Mean For Our Jobs?
This experiment offers genuine reassurance. It shows that AI cannot replicate human teamwork and judgment yet. Human creativity and adaptability remain irreplaceable. For now, AI works best as a helper, not a replacement.The future of work likely involves collaboration. People and AI will work together. This study proves that human skills are still vital for success.
