Opeli is now entering the most exhausted trend in artificial intelligence: AI agents or tools that take a step beyond chatbot to perform complicated, multi -stage tasks on behalf of the user. On Thursday, the company debuted with a chatgpt agent, which charges a tool that can finish working on your behalf with your own “virtual computer”.
In briefing and demo with The VergeYash Kumar and Isa Fulford – a leading cable and conducting research on the ChatGPT agent – found that it is powered by a up-to-date model, which OpenAI has developed especially for the product. The company announced that the up-to-date tool can perform tasks such as looking at the user calendar to inform him about upcoming clients’ meetings, planning and purchasing ingredients to make a family breakfast, and create slide deck based on the analysis of competitive companies.
The model behind the ChatgPT agent, which has no specific name, has been trained in the scope of complicated tasks that require many tools – such as a text browser, visual browser and a terminal in which users can import their own data – by learning to strengthen, the same technique used for all models of OpenAI reasoning. Opeli said that Agent Chatgpt combines the possibilities of both the operator and deep research, two existing AI tools.
To develop a up-to-date tool, the company connected teams behind the operator and deep research in one unified team. Kumar and Fulford said The Verge that the up-to-date team consists of 20 to 35 people in various products and research.
In the demo Kumar and Fulford, they showed potential cases of using a chatgpt agent, for example, asking him to plan the night, connecting to the Google calendar to see when the user has a free evening and then reference to reference to find opening in some types of restaurants. They also showed how the user can interrupt the process by adding, say, the next category of the restaurant to search. Another demonstration showed how Agent Chatgpt can generate a report on research on the creation of Labubus children compared to Beanie.
Fulford said that she liked to employ it for online shopping, because the combination of technology with deep research and the operator worked better and was more true than trying the process only using the operator. Kumar said that he began to employ a chatgpt agent to automate miniature parts of his life, for example a request for a up-to-date office car park at Openai every Thursday, instead of appearing on Monday, forgotten, forgotten to ask for no parking.
Kumar said that because Agent Chatgpt has access to “the whole computer” instead of just a browser, “they improved the set of tools” enough.
However, according to the demo, the tool may be a bit sluggish. When asked about delay, Kumar said that their team is more focused on “optimization in terms of difficult tasks” and that users do not want to sit and watch operating ChatgPT agents.
“Even if it takes 15 minutes, half an hour, it is quite a large acceleration compared to how long it will take,” said Fullord, adding that the OpenAI search team focuses more on cases of low delay employ. “This is one of those things in which you can start something in the background and then come back to it.”
Fullord said that before Agent ChatgPT does something “irreversible”, for example, sending E -Mail or making a reservation, he first asks for permission.
Since the model behind the tool increased the possibilities, Opeli said that it activated the security that it created for “high biological and chemical possibilities”, despite the fact that the company stated that there is no “direct evidence that the model could help the newcomer create serious biological or chemical damage” in the form of weapons. Anthropic activated in May Similar security For launching one of your Claude models, Opus 4.
Asked whether the tool can perform financial transactions, Kumar said that these activities have been circumscribed “for now”, and that there is an additional protection called Watch mode, in which the user navigates to a specific category of websites, like financial websites, he cannot leave the Tab Chatgpt agent agent acting or tool.
OpenAI will start implementing the tool for Pro, Plus and team users – select “Agent mode” in the menu or type “/agent” to access it – and the company said that it will provide them with chatgpt enterprise and Education Users this summer. There is no schedule for the European Economic Area and Switzerland yet.
The concept of AI agents has been a bastal trend in the industry for years. Ideal programmers are working on something like Jarvis Iron Man, a tool that can perform specific work functions, check people’s calendars for the best time to plan the event, buy a gift based on a friend’s preferences and more, but at the moment they are circumscribed to helping in coding and compilation of research reports.
The term “agent AI” became more common for investors and technical directors in 2023 and quickly raised the speed, especially after the announcement in February 2024, fintech clarna, that in just a month of activity, his own agent AI dealt with two-thirds from customer service-700 full-time employees. From there, the directors of Amazon, Meta, Google and others began to mention their goals of the AI agent Connection of earnings after combining with earnings. And since then, AI strategically employing these goals: for example, Google, last week, hired the general director of Windsurf, co-founder and some members of the research and development team to lend a hand in further agent AI projects.
The debut of the CHATGPT OPENAI agent takes place after the January issue of the operator, which the company Calculated as “An agent who can go to the network to perform tasks for you” because it has been trained to be able to operate internet buttons, text fields and many others. It is also part of a larger trend in artificial intelligence, because enormous and miniature agents of Chase AI, which will attract the attention of consumers and get the best habits. In October last year, Anthropic, supported by Amazon Ai Startup for Claude, released a similar tool called “computer”, which settled as a tool that could employ the computer in the same way as a person can perform tasks on behalf of the user. Many AI companies, including OpenAI, Google and embarrassment, also offer an AI tool, which all three called deep research, meaning an AI agent, which can write significant analyzes and research reports about everything that the user wants.