OpenAI Launched Operator: AI Agent to Automate Your Tasks

As reported in November last year about OpenAI’s plan to announce an AI agent called Operator in January this year, OpenAI launched it on Thursday. It can browse the web for you, can book tickets, order grocery items, download lectures, fill in forms, and perform many such tasks of similar nature.

The Operator is currently available to ChatGPT Pro users, with a $200 per month subscription plan, and to users only in the USA.

What is Operator and How Does it Work?

OpenAI explains Operator as “an agent that can go to the web to perform tasks for you,” and it is designed to interact with “the buttons, menus, and text fields that people see on a (web) screen.” It can type, click, and scroll a webpage to perform your task autonomously.

OpenAI’s AI agent is backed by a model called Computer-Using-Agent, similar to what Anthropic, the owner company of Claude, launched in October last year. This AI agent also has the capabilities of ChatGPT-4o’s vision that allows it to interact with the visual elements on the web.

It’s also capable of advanced reasoning and works by reinforcement learning – machine learning where an agent learns to make decisions by interacting with an environment, maximizing long-term rewards through trial and error.  

OpenAI describes it as:

“Operator is one of our first agents, which are AIs capable of doing work for you independently – You give it a task and it will execute it.”

Is Operator Fully Autonomous?

This agent tool is not completely independent yet. Sometimes, it can perform wrongly. You can manually stop it from working to correct its mistake and instruct it to work in the right way.

Operator has been released as a research preview, so it’s not perfect right now. It can make mistakes. It has been designed to ask for confirmation before completing sensitive tasks like order completion or sending an email.

Additionally, it can’t browse gambling and adult sites for safety reasons.

Try ChatGPT free, an AI chat tool powered by latest model of ChatGPT API.

How to Use This AI Agent?

Just describe the task to the AI agent you want to be done, and it’ll take care of it. You can take control of the browser anytime, and it’ll prompt you to step in for things like logging in, entering payment details, or solving CAPTCHAs.

You can customize your workflow in Operator by giving specific instructions for all websites or just certain ones, like setting default shipping options on Amazon. You can save prompts for quick access on the homepage, perfect for tasks like scheduling weekly fitness classes.

Just like using multiple browser tabs, OpenAI’s AI agent can handle multiple tasks at once, such as ordering a custom phone case on Redbubble while reserving tickets for a local event.

Who Can Access Operator?

If you’re a ChatGPT Pro plan subscriber and living in the United States of America, you’ve every good reason to access it. You can access it at Operator.ChatGPT.com.

OpenAI intends to extend the Operator’s availability to ChatGPT Plus, Team, and Enterprise users in the future.

Can OpenAI’s Computer-Using Agent Collect Your Data?

Just like ChatGPT, OpenAI can use your conversation in this tool to train its AI systems. However, OpenAI claims that this AI agent doesn’t collect sensitive information like usernames and passwords that it can use to log in to websites and carry out tasks autonomously.

You can also opt out of this data collection from ChatGPT’s setting by turning off “improve the model for everyone” and thus can disallow the use of your data for training AI models.

Final Thoughts

The launch of OpenAI’s Operator is a step further in the race of AI tech companies like Google and Anthropic to achieve AGI. In OpenAI’s 5-level roadmap to AGI, the announcement of this agent shows the achievement of the third level, only two levels behind.