Openai presents Operator, your AI agent to automate tasks such as buying or reserving a table

Jane Anderson
Jane Anderson
Interfaz de Operator

At the beginning of this year, in its reflections for 2025, Sam Altman, CEO of OpenAi, He pointed out that artificial intelligence agents will be added in the short term to the workforce and materially change the performance of the companies. Now, the company has presented Operator, A first agent development with the capacity to use the Internet browser and perform tasks for the user.

Operator at the moment is only available for PRO users in the United States And it is an experiment, or a previous view of research, as the company points out. This assumes that it presents limitations and will evolve according to the feedback that Openai collects from the users.

The new tool involves the first great attempt of the company in the development of Artificial Intelligence Agentscapable of performing tasks for users independently. As they point out from the firm, it will be enough to assign a task to Operator for him to execute it.

Operator: Assistance in daily tasks

The user will be able to ask the agent for a series of actions to perform in the browser, such as looking for the accommodation of a trip, booking table in a restaurant or making the purchase, which It will save time in daily tasks. In the operator interface, in addition, there are several categories of tasks that users can choose, such as purchases, deliveries, dinners and trips, all of which allow different types of automation.

Users must introduce the instruction of the task they want Operator to do and this will take it. However, users can take control of the action at any time and Operator is trained to request intervention of the user If necessary, as session or information to make payments.
In addition, if you find challenges or make errors, Operator can self -regree, and in case of blocking and needing help, it simply returns control to the user, which guarantees a fluid and collaborative experience.

As Openai explains, users can customize your workflows In Operator adding personalized instructions, such as configuring search preferences at Booking.com. It also allows you to save quick accesses on the home page, ideal for repeated tasks such as replenishing food in Instacart. Also, as would happen with multiple tabs in a browser, users can make Operator Execute various tasks simultaneously creating different conversations.

The company indicates that Operator works with a New Artificial Intelligence Model called Computer-Using Agent (CUA). It combines GPT-4O vision capabilities with advanced reasoning, so it is trained to interact with graphic user interfaces (GUI), buttons, menus and text fields of forms such as those that people usually see on a screen.
In this sense, Operator can “see” through screenshots, and “interact” with the browser using all the actions that allow a mouse and a keyboard.

Operator⁠ transforms AI of a passive tool to an active participant in the digital ecosystem”, Explain from OpenAi. “It will speed up user’s tasks and bring agents’ benefits to companies that wish innovative client experiences and greater conversion rates

Operator Security and Limitations

The company explains that it is collaborating with companies such as DORORASH, Instant, Openable, Priceline, Stubhub, Thumbtack and Uber, To integrate its platforms and services and ensure that the agent addresses the needs of the real world respecting the safety and privacy of the users, as well as the policies, terms of service and the experience of these brands.

In terms of security, Openai explains that Operator is designed for the user to take control at critical points, to request confirmations before finishing any important task and to reject confidential tasks. In addition, the user can voluntarily excluded from operator training and eliminating navigation data. He is also trained to pause a task if a site seems suspicious and for Reject harmful requests and block content not allowed.

From OpenAi they already warn that Operator Find limitations in carrying out certain tasks, especially with complex interfaces, such as the creation of slide presentations or the administration of calendars. Therefore, he hopes that the research test phase will allow you to solve these limitations with the help of user comments.

However, Openai is not the first company to present its artificial intelligence agents to the market. At the end of last year Google presented Project MarineR., his own agent, backed by his Gemini models suite, with the ability to perform actions on the web. And before that, Anthropic presented a new version of Claude, Also with the ability to interact with the computer.