**OpenAI Introduces “Operator,” an Autonomous Multi-Step Task Agent**
OpenAI has launched *Operator*, an innovative agent crafted to autonomously manage multi-step tasks. The company, famed for developing ChatGPT, presented a preview of Operator on Thursday, highlighting its features and performance. This novel tool can undertake a range of activities, including web browsing, calculating refunds for canceled orders, pinpointing specific customer profiles in sales databases, grocery purchasing, and even emailing.
On desktop systems, Operator’s functionality broadens to file downloading, PDF merging, spreadsheet analysis, and image exporting, proving to be a robust assistant for both personal and business needs.
### OpenAI’s Commitment to Agentic AI
This launch signifies another advancement in OpenAI’s goal to declare 2025 the “year of agentic AI.” Just a week ago, the company introduced *Tasks* for ChatGPT, a capability that allows users to automate recurring prompts, such as obtaining daily tech news summaries or setting reminders. Although similar functionalities can be found in tools like Google Alerts and calendar applications, these developments underscore the increasing significance of AI bots in streamlining users’ workflows.
When paired with Operator’s capacity to autonomously tackle more intricate tasks, OpenAI’s vision of evolving ChatGPT into an essential, all-inclusive tool becomes ever more apparent.
### Understanding How Operator Functions
The underlying technology of Operator is a Computer-Using Agent (CUA) driven by GPT-4o’s vision mode. This allows Operator to “see” what appears on a user’s screen through screenshots and interact with graphical user interfaces (GUIs) by clicking buttons, entering text, scrolling, and more. For instance, Operator can look for campsites in Yosemite that feature picnic tables, demonstrating its knack for managing specific, user-defined inquiries.
### Emphasizing Safety with Operator
Due to the semi-autonomous characteristics of Operator, safety is paramount. OpenAI has put in place several safeguards to mitigate risks and curb misuse. For example, Operator is designed to prevent harmful or illegal tasks and is barred from accessing prohibited sites, such as gambling sites, adult content, and vendors of drugs or weapons.
Furthermore, OpenAI supervises user interactions with Operator in real-time via automated safety mechanisms. These systems aim to ensure adherence to the company’s Usage Policies and possess the ability to issue warnings or block prohibited actions. To improve oversight, OpenAI has established automated detection mechanisms and human review processes to identify and rectify misuse in critical domains like child safety and deceptive practices.
To minimize the risk of expensive errors, Operator necessitates user confirmation prior to executing actions such as placing orders or sending emails. This allows users to verify the agent’s tasks before any irreversible steps are taken. Currently, Operator is also prohibited from engaging in high-risk activities, such as banking transactions.
### Access and Pricing
At present, Operator is available in preview mode solely to subscribers of OpenAI’s premium *ChatGPT Pro* plan in the U.S., which costs $200 monthly. Over time, OpenAI intends to broaden access to additional subscription levels, including Plus, Team, and Enterprise users.
As OpenAI continues to enhance its agentic AI capabilities, offerings like Operator and Tasks signify a notable progression in integrating AI into the fabric of daily life.