In the past week, the Operator of OpenAi did the following things for me:
-
Ordered me a new ice cream scoop on Amazon.
-
Bought me a new domain name and configured the settings.
-
Booked a Valentine's day date for me and my wife.
-
Planned a haircut.
It usually did these tasks autonomously, although I had to push it from time to time and occasionally had to make a loop of failed attempts.
If you just catch up or if you are distracted by the Deepseek News this week, that all other AI news has overshadowed a new so-called AI agent released by OpenAI last week.
The tool, which was invoiced as a “research review”, is only available for people who pay $ 200 a month for the highest subscription of the company, Chatgpt Pro. It gives users the option to send an AI agent who can use a web browser, fill in forms and take other actions on behalf of a user.
AI agents are now completely in the anger in Silicon Valley. Some insiders from the industry think they are the next big step in AI options, because an AI agent who can use a computer can actually perform valuable real-world tasks instead of only offering help. Many of the leading AI companies, including Google and Anthropic, test autonomous agents whom they claim will eventually be able to 'hire' as fully-fledged employees.
I upgraded my chatgpt subscription to put the operator to the test and see what an AI agent could do for me.
At first glance, operator looks a bit like regular chatgpt, except that when you give it a job “buy me a bag of dog food on Amazon, for example operator opens a miniature browser window, types” Amazon. com ”in the address bar and starts clicking around and tries to follow your instructions.
It can ask a few clarifying questions. (Do you want to eat with chicken flavor or beef taste? Overnight shipping or two -day?) Then, as soon as it is convinced that it has made the right choice, the operator asks you for a definitive confirmation, the dog food will stop in your shopping cart and stops and puts the dog food and places the order. (Operator does not carry passwords or credit card numbers in-U must take over the mini-browser and type those things in yourself, but the rest does it alone.)
The entire operator is that you do not have to supervise it – it can perform tasks in the background while you do other things. But I noticed that I was glued by the window, mesmerized by the sight of a self -driving web browser who clicked on buttons, type words in boxes and select from subsequent choice menus, everything in itself. Look, ma, a computer with a computer!
Operator also impressively did well on a few relatively simple tasks that I gave it:
-
It successfully ordered lunch on Dordash for my colleague Mike and sent it to his house. (I didn't tell what to order it, but Operator chose a Mexican restaurant, chose a handful of dishes for him and the delivery person even tipped $ 7.)
-
It responded to hundreds of unread LinkedIn reports for me, after I had checked it over my LinkedIn profile. (Although to my horror it also registered for a webinar.)
-
It earned $ 1.20 for me by setting accounts on websites that offer small money rewards for completing surveys. (It may have made more, but I started to feel guilty for spamming the surveys with fake, robot written answers.)
But Operator also failed with a number of other tasks and revealed his limitations:
-
It could not scan my recent columns and add them to my personal website, because the operator's browser was blocked to enter the Times website. (It is also blocked for a number of other sites, including Reddit and YouTube. The Times sues OpenAi and Microsoft for copyright infringement with regard to the training of AI models.)
-
It would not play online poker for me. (Operator replied: “I am unable to help with gambling or related activities”, which seemed to be a reasonable rejection, given the chaos that a gambling bone could create.)
-
And it was prevented from logging in to a number of sites by Captcha tests. (Which I found reassuring, given that the entire point of captchas is to scare robots.)
All in all, I discovered that the use of operator was usually more problems than it was worth. I could have done most of what it did for me, with less headache. Even when it worked, the OM asked so many confirmations and reassuring before I acted that I felt less like I had a virtual assistant and more as if I supervised the most uncertain trainee in the world.
These are of course early days for AI agents. AI products improve version to version, and it is a good gamble that the following iterations of the operator will be better. But in its current form, the operator is more an intriguing demo than a product that I would use – and certainly not something that most people need to indicate $ 200 a month.
That said, I think it's a mistake to write off AI agents. When they become more capable, they can start replacing human employees in some professions. (OpenAi and Meta have already said that they are building AI engineer agents.) And some experts are concerned that more powerful, unrestrained AI agents can provide the safety risks, if they learn to carry out assignments such as “remove a bank account” or “a Perform cyber attack. “
Setting up a number of AI agents on the internet can also cause a recoil from web results, e-commerce sites and other companies that depend on traffic generated by people to pay their bills. (If you are a company that buys advertisements on Amazon, you want those advertisements to be seen by people, not bots that occur as people.) In the future I can imagine that more websites take steps to block AI agents or to be sure to send them pages or products.
AI agents are currently too incompetent to be a big threat. But not much imagination is needed to imagine a nearby future, where most of the web will consist of robots that talk to robots, buy things from robots and write e -mails that will only read other robots.
In other words, the self-driving internet is almost here-hat you click inside while you can.
