Unlocking the Power of OpenAI’s New AI Agent: Discover How Helpful Operator Really Is!

— I got a new ice cream scoop delivered from Amazon.

— A fresh domain name was purchased and set up for me.

— Valentine’s Day plans were made for my wife and me.

— A haircut appointment was arranged.

I didn’t have to oversee everything. Operator, my AI assistant, managed most tasks on its own. But I had to step in sometimes to guide it back when it got stuck.

If you haven’t heard yet, Operator is a new AI agent from OpenAI, launched on January 23. It’s part of the ChatGPT Pro subscription, which costs $200 a month. This tool can browse the web, fill out forms, and perform tasks for you.

AI agents are gaining popularity in tech circles. They could be the next step in how we use AI. Instead of just providing information, these agents can take action, potentially filling roles in various industries. Companies like Google are also testing their own versions of these agents.

I upgraded to ChatGPT Pro to explore what Operator could do. On the surface, it seems like standard ChatGPT. But when I asked it to buy something, it opened a web browser and navigated to the site.

It asked questions for clarity, like preferred flavors or shipping options. Once it felt sure about its choice, it asked for my go-ahead before placing an order. While it didn’t handle passwords or payment info, Operator did handle most steps seamlessly.

While the dream is that it runs in the background, I found myself watching in awe. It was like watching a computer operate all by itself!

Operator completed a few simple requests quite well:

— It ordered lunch for a friend, choosing a Mexican restaurant and tipping $7.

— It managed to reply to many LinkedIn messages after I granted it access, though it did sign me up for a webinar without me knowing!

— It got me $1.20 by signing up for rewards programs that pay for taking surveys. However, I felt a bit guilty about the fake responses it submitted.

Yet, Operator had its hiccups:

— It couldn’t add my latest articles to my website, as it was blocked from The New York Times site and others.

— It wouldn’t assist with online poker, saying it couldn’t help with gambling tasks, which was a sensible refusal given the risks.

— Additionally, it struggled with CAPTCHAs when trying to log in to certain sites, a reminder that those tests are indeed effective against bots.

Overall, using Operator brought as many challenges as benefits. In many cases, I could have completed tasks faster myself. Operator needed so many approvals that I felt like I was managing a nervous intern rather than enjoying the assistance.

This is just the beginning for AI agents. They often improve with updates, so future versions of Operator could be more efficient. But for now, I can’t truly recommend it, especially at $200 a month.

However, I wouldn’t discount AI agents completely. As they evolve, they could take on more roles in the workforce. Some companies are already developing AI agents for technical tasks. There are valid concerns about potential risks if these agents gain too much power.

Letting many AI agents roam freely internet-wide could lead to conflicts with businesses dependent on human traffic. In the future, we might see sites taking action to prevent AI agents from accessing certain features.

As of now, AI agents are not a significant threat. But it’s easy to envision a day when the web is filled with bots handling tasks for other bots, executing transactions, and sending automated emails.

The age of the self-driving internet is just around the corner; enjoy the web while it’s still human-centered.

This article originally appeared in The New York Times.