Scientists have created a company where all employees are artificial intelligence. Guess what went wrong (2 photos)

Researchers from Carnegie Mellon University conducted an experiment in which they created a virtual IT company with robot employees. Even the best AI worker could only complete a quarter of the tasks, and the rest showed even more dismal results, writes Futurism.





The researchers created a virtual company called TheAgentCompany, where all positions — from financial analysts and programmers to project managers — were not occupied by people, but by AI agents from Google, OpenAI, Anthropic, and Meta. An agent is an artificial intelligence model that must independently perform assigned tasks, like a person.

Virtual employees were given routine office tasks: working with files, taking virtual tours of new office spaces, and writing reviews of their colleagues.

The results were disastrous. The best model was Anthropic's Claude 3.5 Sonnet, but even it could only complete 24% of the tasks. It was also extremely inefficient, with each task taking an average of 30 steps and costing over six dollars.

Google's Gemini 2.0 Flash took about 40 steps to complete each task, but only completed 11.4% of the tasks. Amazon's Nova Pro v1 was the worst "employee," completing only 1.7% of the tasks.



The researchers noted that digital workers lack common sense, social skills, and an understanding of how to work on the Internet. AI workers often came up with ridiculous solutions to problems - for example, when one robot couldn’t find the right person in a company chat, it simply renamed another user.

While current AI models can handle some simple tasks well, the results of this and other studies show that they are not yet ready for more complex work that humans can easily handle. This is largely because today's "artificial intelligence" is essentially an advanced version of the predictive text input on your phone, rather than a sentient being capable of solving problems, learning from experience, and applying it to new situations, the authors note.