The personal assistants of AI have just taken a new step. In a live broadcast on July 17, 2025, Sam Altman and a team of Operai engineers presented the novelty that will integrate Chatgpt, the “agent” function.
Versatile, this AI is able to perform several tasks at the same time. It resembles a fusion between two existing characteristics of ChatgPT: Deep Research, which allows long tasks that require in -depth investigation and an operator are performed, which allows in particular to carry out reserves interacting with the web pages.
Multiple
Agent finds all its use in long and complex consultations, which requires several actions. However, it will take a certain time before generating a complete response (usually between 5 and 15 minutes) depending on the complexity of the task that will be requested.
In an example presented by Operai employees, a user asks AI to help him prepare for the wedding of his friends. Dress and accessories according to the “clothing code”, the hotels available nearby, the gift for the bride and the boyfriend, the agent is responsible for informing with the best results according to the needs of his human interlocutor.
Agent can even interact with other applications such as Google Drive, Gmail or Google Calendar. The team demonstrates it with a submission of slides generated from data recovered in a DOC of Google.
Finally, Agent incorporates the generation of images of images in his competition palette, which may include in a broader response. In the presentation of slides mentioned above, the AI has created, for example, the images used to decorate the slides.
Unpublished additions
Concerned about the transparency of its AI, the Operai team has thought about including an interesting feature. Through a small instructor, ChatgPT shows all the actions carried out live or the sites consulted. As if the chatbot used an internet browser for us.
The user can follow the progress of AI or review certain reflection passages (a bit like a live YouTube). “As these tasks are a little longer, you can also see that what Ai does,” said Yash Kumar, a team engineer behind the agent.
Another practical addition, the user can bring elements to AI while in the middle of their reflection. You can request an additional task, bring details or request to completely change certain instructions. In his final response, the agent will have taken these requests into account.
The team behind this feature also wanted to guarantee a safety network in case the ICA errors. The engineer Casey Chu explains that the agent will systematically request confirmation from the user before carrying out an important action, such as sending an email, for example.
As for most Chatgpt functionality modes, the agent will be available in the “Tools” tab of the Chatbot, both on the computer and in the mobile version. Its launch date in France has not yet been formalized, but it will only be accessible to paying professional versions and more chatgpt.
Source: BFM TV
