A look at the Openi operator, a new artificial intelligence agent

Last week, the Openai operator did the following things for me:

  • I ordered me a new ice cream palette on Amazon.

  • It bought me a new domain name and configured its settings.

  • I booked a Valentine’s Day appointment for me and my wife.

  • Has scheduled a haircut.

He has carried out these tasks mostly independently, even if I had to push him from time to time and occasionally save him from a cycle of failed attempts.

If you are just recovering-o if you have been distracted by Deepseek’s news this week, which has in shade all the other news of the operator is a new so-called agent at the released last week by Openai.

The tool, which has been invoiced as a “research preview”, is available only for people who pay $ 200 per month for the highest subscription level of the company, Chatgpt Pro. It offers users the opportunity to direct an agent of Artificial intelligence capable of using a web browser, fill in the modules and take other actions on their behalf.

Ai agents are highly fashionable in Silicon Valley right now. Some professionals think they are the next big step in artificial intelligence skills, because an artificial intelligence agent who can use a computer can actually carry out precious activities of the real world rather than providing assistance. Many of the main artificial intelligence companies, including Google and Anthropic, are testing autonomous agents who claim that companies will eventually be able to “hire” as full -bodied workers.

I updated my chatgpt subscription to put operators through its steps and see what an artificial intelligence agent could do for me.

On the surface, the operator resembles Gatgpt, except for the fact that when you give it a job-“buy me a 30 pounds of dog food on Amazon”, for example, the operator opens a browser window In miniature, “Amazon. COM” types in the address bar and start click, trying to follow the instructions.

It could ask some clarifying questions. (Do you want food to the chicken taste or beef taste? Shipping during the night or two days?) So, once you feel sure you have made the right choice, the operator asks you for a final confirmation, puts the food for Dogs in your cart puts the order. (The operator will not insert passwords or credit card numbers-deVI assume the mini-browser and type those things in yourself-but does the rest alone.)

The entire point of the operator is that you don’t have to supervise it: it can carry out tasks in the background while you do other things. But I found myself glued to the window, hypnotized by the sight of an autonomous web browser by clicking on the buttons, typing the words in boxes and selecting from the drop -down menu, all alone. Look, but, a computer that uses a computer!

The operator also made impressively on some relatively simple activities that I gave him:

  • He successfully ordered lunch on Doordash for my colleague Mike and sent him to his house. (I didn’t tell him what to order him, but the operator chose a Mexican restaurant, he chose a handful of dishes for him and even tilt the delivery person $ 7.)

  • He replied to hundreds of messages from LinkedIn not read for me, after I gave him control of my LinkedIn profile. (Although, with my horror, he also recorded me for a webinar.)

  • He earned $ 1.20 for me creating accounts on websites that offer small cash prizes to fill in polls. (It may have done more, but I started to feel guilty for spam for investigations with false and writing answers in robot.)

But the operator also failed in a lot of other tasks and revealed its limits:

  • He was unable to scan my recent columns and add them to my personal website, because the operator’s browser was blocked by the entrance to the Times website. (It is also blocked by a series of other sites, including Reddit and YouTube. The Times suits Openi and Microsoft for violation of copyright relating to the formation of the AI ​​models.)

  • It wouldn’t play online poker for me. (The operator replied: “I am not able to help with gambling or related activities”, which seemed to be a reasonable refusal, given the chaos that a game bot could create.)

  • And he was prevented from accessing a number of sites via captch test. (Which I found reassuring, since the entire captcha point is to discourage robots.)

Overall, I discovered that the use of the operator was usually more problems than it was worth. Most of what he did for me I could have done faster on my own, with less headache. Even when he worked, he asked so many confirmations and reassurances before reciting that I felt less as if I had a virtual assistant and more as myself supervising the most insecure trainee in the world.

This is, of course, the first days for artificial intelligence agents. The products to the tend to improve from the version to the other, and it is a good bet that the operator’s next iterations will be better. But in its current form, the operator is more an intriguing demo of a product that I would recommend using – and certainly not something that most people need to spend $ 200 per month.

Having said that, I think it is a mistake to delete artificial intelligence agents. When they become more capable, they could start replacing human workers in some professions. (Openai and Meta have already said that they are building agents of artificial intelligence engineers.) And some experts fear that more powerful and unbridled artificial intelligence agents can represent risks for safety, if they learn to perform commands such as “a bank account” or “Perform an attack.”

Tackling a lot of artificial intelligence agents on the internet could also cause a backlash from web publishers, e-commerce sites and other companies that are based on the traffic generated by man to pay bills. (If you are a company that buys ads on Amazon, you want those ads to be seen by humans, not bots who pretend to be human.) In the future, I can imagine more web sites that adopt measures to block artificial intelligence agents or guide them towards some pages or products.

At this moment, artificial intelligence agents are too incompetent to be a threat. But it doesn’t take much imagination to imagine a near future when most of the web is made up of robots that talk to robots, buy things from robots and write and -mail that only other robots read.

The independent internet is almost here, in other words: get your clicks enter while you can.

Leave a Reply

Your email address will not be published. Required fields are marked *