OpenAI revolutionizes ChatGPT with an autonomous agent that performs complex tasks.

The race to provide artificial intelligence with greater capabilities has just taken a new leap with the presentation of the ChatGPT Standalone Agent, a feature that completely transforms the usual experience with OpenAI's famous chatbot. Now, ChatGPT is capable of performing tasks of a certain complexity autonomously using your own virtual computer, with access to different tools and the ability to interact in real time with websites, files, and external services.

This breakthrough means going beyond simple conversation: The agent makes decisions, navigates the web, generates documents and automates processes, without the user having to provide step-by-step instructions. Everything happens within the same ChatGPT environment, and always under the supervision and validation of the user, who can interrupt or adjust the task at any time.

A function based on the integration of previous technologies

The arrival of the agent to ChatGPT unifies several recent OpenAI developments, such as Operator (which already allowed the chatbot to interact with web pages) and Deep Research (focused on advanced reporting from multiple sources). This combination transforms the system from just being an intelligent chat into a a collaborator capable of executing tasks from start to finishThe user can delegate processes such as gathering information, planning events, creating spreadsheets, or generating presentations, and the agent handles the entire workflow, requesting confirmation when necessary.

What exactly can the ChatGPT agent do?

During the demonstrations, examples as varied as the comprehensive wedding planning (with hotel searches, weather forecasts, gift identification, and even reminders for pending purchases), the generation of official documents, or the automation of online purchases. No more switching between multiple tabs or apps.: The agent integrates everything into a single conversation.

Among the latests Moravia's compositions featured features are:

• Creation and manipulation of spreadsheets and presentations based on real data.
• Dinner planning, shopping lists, and shipping coordination.
• Search and purchase products in online stores from ChatGPT.
• Reservation of restaurants, hotels or services taking into account preferences and budget.
• personal management (agendas, reminders, appointments) and professional analysis of complex data.

A virtual computer at the disposal of artificial intelligence

The key to the new agent's operation is its permanent access to a virtual environment managed by OpenAIThere you can use a visual browser to interact with websites as a person would, access terminals to run code or manipulate files, and connect to external applications (such as Google Drive, GitHub, or Gmail) if the user authorizes it. Each task runs in an isolated temporary session, that It is deleted upon completion to strengthen security and privacy.

This design allows you to Analyze information in real time, edit documents, and make contextualized decisionsFor example, you can identify an appointment in your calendar and, if you wish, reschedule it or prepare a report linked to that date.

Benchmark: How does the agent perform against humans and other models?

The first comparative tests (SpreadsheetBench, DSBench y BrowseComp) have shown that the ChatGPT agent outperforms previous AI models and many humans in analysis, editing, and navigation tasksIn spreadsheet and data science benchmarks, the agent has achieved record-breaking accuracy and efficiency, indicating a significant technological maturity.

In real environments, It has proven capable of analyzing large volumes of emails or customer messages, synthesizing user profiles, automating financial reports and managing marketing tasks., all in minutes and without constant user intervention.

Privacy and security: absolute user control

OpenAI has put special emphasis on security and giving the user complete control over the agent's actions. Any sensitive operations (such as submitting forms, making reservations, or executing code) requires explicit confirmation before being performed. The agent is programmed to reject high-risk actions, such as financial transactions or potentially dangerous tasks. Additionally, all activity can be erased with a single click, and user data is not stored after the session.

The system monitors the entire process and allows real-time observation of the steps taken by the AI. If the user wishes, they can pause or interrupt any action before its final execution, reinforcing transparency and trust in the tool.