OpenAI has released Operator, an AI agent that can carry out tasks on your behalf using a web browser. The tool is available only to users of the company’s ChatGPT Pro subscription tier for $200 a month. While Operator showed impressive capabilities in some tasks, such as ordering lunch and responding to LinkedIn messages, it also struggled with others.
In testing, Operator performed well when given specific instructions, like buying dog food on Amazon or setting up accounts to earn rewards. However, it often required human intervention and failed at more complex tasks, like scanning articles for my personal website.
Operator’s limitations include being blocked from accessing certain websites, such as The Times and Reddit, due to copyright infringement concerns. It also couldn’t play online poker or log in to sites using CAPTCHA tests.
Overall, I found that using Operator was usually more trouble than it was worth, with too many confirmations and reassurances required before actions were taken. While the technology holds promise, its current form is more of a demo than a practical tool for everyday use. Nevertheless, as AI capabilities improve, operators like Operator could potentially substitute for human workers in certain occupations, raising concerns about safety risks and potential backlash from businesses that rely on human-generated traffic.
Note: The article has been simplified to retain the essential information while reducing the complexity of language and sentence structure.
Source: https://www.nytimes.com/2025/02/01/technology/openai-operator-agent.html