OpenAI's new AI agent struggled with basic tasks

OpenAI's new ChatGPT Agent, designed to automate tasks, took an hour to order food and recommended a baseball stadium in the ocean, highlighting its limitations. The AI agent, which uses a "virtual computer," requires human approval for significant actions, slowing down its performance. It struggled with simple tasks and made factual errors, such as suggesting a stadium location in the Gulf of Mexico. The agent is initially available to Pro users with a prompt limit, and will be rolled out to other subscribers later. The release underscores ongoing challenges in AI reliability and the need for human oversight.