OpenAI Planning to Launch an AI Agent Called ‘Operator’ in January

While OpenAI has denied releasing its next big model codenamed ‘Orion‘ this year, it’s preparing to launch an AI agent called ‘Operator’ next year in January. Bloomberg reports that the Operator AI agent can autonomously perform actions on a computer or a web browser based on user input.

According to the report, OpenAI informed its staff on Wednesday that the company is planning to launch the tool in January. The Operator AI agent will be first available as a research preview and through OpenAI’s API to developers.

After the ChatGPT boom, AI companies are now moving towards action-driven agents. Recently, Anthropic released its Computer Use tool that can perform actions on a computer based on the user prompt.

I tested Anthropic’s Computer Use API through Docker and found that it’s promising but pretty slow. It finds the click target by analyzing what is on the screen. Often, it gets stuck in a loop and even the improved Claude 3.5 Sonnet (New) model can’t get out of it.

Nevertheless, 2025 is going to be the year of AI agents. Google is also preparing to release its Jarvis AI agent in December this year. It will be available in the form of a Chrome extension and can perform actions based on user input. Google is calling it “a helpful companion that surfs the web with you.”

Since LLM scaling has hit a wall, companies are looking to create new applications and use cases based on AI’s current strengths. In a recent Reddit AMA, OpenAI chief Sam Altman said, “But I think the thing that will feel like the next giant breakthrough will be agents.” Now, in the AI agent race, we will have to see which company outperforms the rest.