OpenAI surprised the world on Thursday when it announced Operator AI agents for ChatGPT. Given the latest leaks, I knew the Operator launch was imminent, but I thought OpenAI would want to make a bigger deal about releasing its first AI agent that can do things for the user.
It turns out Operator doesn’t need much fuss, and a short livestream was enough to demonstrate this evolution of ChatGPT that I’ve been dying to see. After watching the demos on OpenAI’s website and the full live stream, I’m absolutely convinced OpenAI is onto something here.
Operator appears to be a brilliant tool that could change how people use computers. But I also know for certain that I wouldn’t pay the $200/month ChatGPT Pro subscription needed to access it.
The best part about Operator is that it’s not what I expected it to be. But that’s because I’ve only seen AI agent demos from Anthropic and Google so far. That includes Gemini’s agentic capabilities on the Galaxy S25 series. All these examples involve having the AI control a computer or mobile device. Even ChatGPT’s ability to interact with coding and writing apps falls into the same category.
Operator is different. It’s a browser in the cloud that relies on mouse and keyboard input, just like a human would. It can multitask, which means executing different complex commands at the same time, and then notify you when it’s done or when you have to take over.
Operator does the web browsing securely and privately and doesn’t screenshot sensitive data. I’m assuming here that you have set your privacy settings correctly and instructed ChatGPT never to send your chats to OpenAI for training the AI.
Given all that, I can see myself moving to Operator whenever I have an idea that requires an internet search.
Maybe it’s about buying products or researching work-related topics. I’d hand over the task to Operator inside a ChatGPT-like window while I work on something else. It’s similar to what I would do now, but the difference is I’d spend less time inside the ChatGPT window, as Operator would do the computing for me.
The livestream, available below, will give you an idea of how Operator works and how it can save you minutes each day. It could be an incredible assistant for getting things done, like ordering groceries online, making reservations, and even buying products.
Operator might not work correctly all the time, and it could produce errors. But it would handle much of the repetitive web browsing work you do, freeing up time for other activities.
The more I describe Operator, the more it feels like a service I’d want to pay for. I want Operator to work in the cloud privately and securely. I want multitasking. I want it to ask for permission to do certain things, and I don’t want it to handle payments and sensitive activities.
But I’m not ready to pay $200/month for ChatGPT Pro for the sole benefit of testing Operator. This isn’t a finished product, and OpenAI needs help to see what people do with the AI agent. That’s perfectly fine; OpenAI should rely on external input to deliver the kind of products customers want to use. I can’t justify paying $200/month solely for that. A limited, free beta test for Operator would make more sense.
I say that as a longtime ChatGPT Plus user who doesn’t need all the other perks that come with the Plus subscription. If you’re already on ChatGPT Pro, there’s no reason not to test Operator right away.
There is another issue. Operator is currently a US-only product. Even if I were a ChatGPT Pro subscriber, I couldn’t use it in the EU.
Separately, I will mention the biggest downside with Operator, at least for now. This AI agent only solves one of my problems: internet browsing for specific chores. Operator can’t control on-device actions. I can’t tell it to find a document or image online and save it locally on my computer. Or, at least, that’s what the live stream presentation implies.
Operator is an AI working in a predefined “box,” the virtual browser that runs in the ChatGPT/Operator web interface.
I can’t tell Operator to change computer settings, open or close apps, or perform tasks involving multitasking in the background. Then again, I will need time to trust an AI to control my computer. And I won’t want that AI to talk with outside servers, but that’s a topic for a different day.
Back to the cost of trying Operator, I will say that I don’t expect it to be free of charge. Nor should it be. I’d trust a free AI agent even less.
I would buy the ChatGPT Plus subscription for Operator access. I’d actually pay a reasonable price hike to have such AI agents added to the ChatGPT Plus experience. Hopefully, we don’t have to wait too long before Operator is available to the more affordable ChatGPT tiers.
