The most valued companies in the world (to which global scholarships therefore grant a value in equity, temporary and fluctuating depending on market conditions) are technological companies. In the top 10 of July 2025, the first 6 places are occupied by Nvidia, Microsoft, Apple (this is the exception, AI is not its main business), Amazon, Alphabet/Google and Meta, American mastodons which provide the hardware or software allowing AI, and for some, develop the most powerful models.
Et openai? With a value of around $ 80 billion, it is for the moment a pure player (It only does AI, unlike others) which is very far from 3.800 billion Microsoft. And yet, the company of Sam Altman, as you know, is at the forefront of the AI with Chatgpt, which has just launched a new option this summer: agent. I tried it, and it’s quite crazy …
An agent who works for you (including by surfing the web)
The notion of agent in artificial intelligence is not new. This is the principle of configuring an AI to make him perform precise tasks, without having to explain what we want (“prompt”) to each new conversation. But with Chatgpt Agent, Openai goes a step further: he created the digital assistant of the future, who works on a virtual computer before your eyes, which clicks for you and fills data fields (among other things much more complex, but for the general public, it is the most telling).
Check my visa for the United States …
I’m going to the United States soon. In addition to a valid passport, you need a visa, which is now obtained online. He has a lifespan of 2 years, so I had to check if mine was still valid. I did it by myself, of course. A procedure not very complex but which requires finding the official site (there are many scams) of the ESTA, to encode personal data and those of my passport. The ideal for testing the capacities of this new Chatgpt agent!
On the chatgpt home page, in a new conversation, we can actually activate the agent function:
I just provided him with a photo of my passport (because I know he needs the data there on the main page) and I asked him to check if my esta was still valid. A very short instruction, in everyday French. The proof that you shouldn’t really know it to use AI …
After 3 seconds of reflection, he transformed the response zone into a browser screen, and he started his research. He combined what he knows (how to check an ESTA) and what he found by carrying out this task, in the first person, himself. He was therefore on the official page (he knows it, no need to go through Google, which avoids scam sites), and he read the page. He deduced that it was necessary to click on the “Check the Esta status” button, then he filled all the necessary fields, thanks to the information found on the photo of my passport that I joined to the initial request. He understood that certain data were optional and did not take it into account. We see him navigating and (trying to) fill the “date” fields. And at each stage, he briefly describes what he does (“I’m going to scroll the drop -down menu to find the number 22”).
After 9 minutes (He found it difficult to click on the 22nd of my day of birth, precisely) of encoding and reflection, he “closed” his surf window and provided me with the answer I already knew: my esta is still in order, I can go to the US … I was faster than him, but he succeeded – and we can let him work and close the window, we receive a notification when he finished. An unimaginable feat even 2 years ago. It’s a bit crazy, and a little scary if we think about everything he can now do “himself” (even if he reacts to a request).
Conclusions: What are we still going to do in 2040?
Chatgpt Agent, available for paid subscribers (around 20 €/month), is one more step towards active artificial intelligence, which exceeds the concept of tools providing answers to specific questions or requests. With a short sentence and a photo, the OpenAi agent checked if my visa was still valid. The example was probably not the best, because it is better to avoid entrusting data and a task also sensitive to an online platform. Chatgpt says he can buy cinema tickets, and even check your agenda to see when you are available in the evening to go. I tried and inevitably, he asks you to regain control to make the payment or fill out certain personal data, he asks you to connect to Google to read your agenda. This is not very intelligent, it is rather digital assistance, ultimately.
One thing is certain, as with Google 20 years ago, confidence will settle over the years, and you will give it all the necessary access because you prefer it to take care of certain tasks for you. In 2024, Chatgpt (and the other generatives of the market) summed up a long text in Spanish and transformed it into email into French. In 2030, he could take care of 90 % of the “Ideal vacation booking procedure for next summer”. But what are we still going to do in 2040, then? No one has the answer: the ethical and philosophical debate opens, and it is far too long for this article …
