OpenAI has launched a new AI agent in ChatGPT capable of performing a variety of complex digital tasks based on user instructions, including running code, creating presentations, and integrating with external services.
This is reported by Finway
Capabilities and Features of the New Agent
- The agent in ChatGPT performs actions on behalf of the user, such as running code or creating presentations.
- It is available to Pro, Plus, and Team subscribers and utilizes connections to third-party applications via API and terminal.
- OpenAI has limited the agent’s memory and implemented additional security measures to prevent biological threats due to the powerful capabilities of the model.
The new ChatGPT agent from OpenAI is focused on automating complex processes. It easily creates presentations, conducts competitor analysis, plans purchases, and utilizes information from connected applications like Gmail or GitHub. Interaction can be done through natural language prompts. To activate the tool, simply select the appropriate mode from the dropdown menu in the interface.
“ChatGPT can now do work for you using its own computer. Introducing ChatGPT agent—a unified agentic system combining Operator’s action-taking remote browser, deep research’s web synthesis, and ChatGPT’s conversational strengths.”
Safety and Usage Limitations
The agent combines the capabilities of previous solutions, including Deep Research, and the website navigation feature. The main distinction is its ability to perform complex tasks: from gathering and analyzing information to organizing actions in real-time. For example, the agent can automatically plan a Japanese breakfast or gather data on competitors and create a slide presentation. According to OpenAI, the new product significantly outperforms previous and competing solutions that struggled with multi-step tasks.
“Watching the ChatGPT agent use a computer to do complex tasks has been a real ‘feel the AGI’ moment for me; something about seeing the computer think, plan, and execute hits different.”
At the same time, the company highlights potential risks associated with the model’s high effectiveness in biological areas. To prevent undesirable scenarios, all requests undergo biological classification, and suspicious ones receive additional scrutiny. The agent’s memory is currently disabled to minimize the risk of unauthorized data access. The possibility of restoring memory will be considered later.
Despite the stated capabilities, the ultimate effectiveness of the agent can only be assessed after its widespread implementation. OpenAI reports that it is currently testing an even more advanced version; however, stability in real-world conditions requires further verification.
Meanwhile, experts from the Future of Life Institute (FLI) emphasize that AI model developers are “absolutely unprepared” for the potential threats posed by the development of artificial superintelligence.