Openai launches his agent, capable of performing complex tasks

Assign him a task, the chatgpt agent ” takes care of work for you ». Meanwhile, This is the promise. Moreover, undoubtedly a little occurred, of the tool launched Thursday, July 17 by Openai, accessible only to paying subscribers, and unavailable for the time being in the European Union. Furthermore, According to the American company. Similarly, the Chatgpt agent can ” Watch my agenda then brief me on my next customer meetings according to the news »Or« rearrange my appointments Automatically, we read on his presentation page. Meanwhile, This involves granting Chatgpt an access to part of the digital services that we use – mail. However, agenda box for example – by creating what the company calls a “connector”.

According to Openai, the Chatgpt agent can also ” fill in online forms, or create neat presentations ». In addition, To do this. Consequently, he can harvest information on the internet, possibly putting us to use if we must authenticate on a site to which we are subscribed. The agent will then sail by himself, by clicking on the buttons that seem to him. Overall. a task can take him up to several tens of minutes, a notification appearing on the user’s smartphone once the operation is completed.

Lively competition. dubious efficiency – Openai launches his agent, capable

OPENAI is far from the only company to have engaged in the quest for openai launches his agent, capable agents capable of performing complex tasks: many other large American technological companies have pursued the same goal since 2023. And the first general public tools have emerged in 2024. The artificial intelligence startup Anthropic announced its Computer System USE last October, followed by Microsoft Copilot Actions in April 2025, then by the Gemini in Google Chrome in May, clearly criticized by the American press.

In fact, this type of tool seems particularly difficult to develop. This is evidenced by Apple’s misadventure which. in September 2024, had praised the arrival within a few months of an “agent” moved by artificial intelligence, then had completely passed it over in silence in June 2025, during a new presentation of its future features of AI. The release date of the future Apple agent is today unknown.

Openai himself recognizes that his agent chatgpt. is not mature: “He can make mistakes”we read on his presentation page. But the American company certifies that its performances are clearly superior to those of the experimental agent Operator which it had launched in January 2025. also very criticized by the American media. To improve it, the company merged it with Deep Research, its in -depth research tool.

Interrogation on security

To dissipate the fears of its customers. the company maintains having particularly treated the security of its new agent. But she recognizes that her job increases the risks associated with the use of chatgpt. in particular due to the very rich personal openai launches his agent, capable information that he can consult. The company also recommends thinking that it will grant him permission to access his agenda or mailbox.

Openai also warns that, “ Since the agent Chatgpt can perform direct actions, successful IT attacks can have a greater impact ». So she trained her agent ” to refuse high -risk tasks such as banking transfers ». On its website. the company demonstrates an automated task: the Chatgpt agent creates a sticker with the image of a company, finds an online printer, then commands a few hundred copies. But the robot stops at the time of finalizing the purchase: it runs out to humans. the only one authorized to grasp its bank details.