OpenAI Launches Agent that Performs Tasks Independently

openai

ChatGPT agent can browse websites, execute code, and create presentations.

OpenAI adds a new AI agent to ChatGPT that independently performs computer tasks. With a simple prompt in natural language, you can ask the agent to manage your schedule, generate a slide deck, or conduct a competitive analysis, for example.

More than a Smart Chatbot

The agent combines tools like Operator and Deep Research. It has access to a terminal, APIs, and apps like Gmail or GitHub via connectors. This positions OpenAI more strongly in the field of agentic AI, which focuses on automating tasks rather than just providing answers. AI agents are meant to automate repetitive tasks and can also be used in your personal life to plan travel routes or come up with dinner ideas.

benchmark
Source: OpenAI

The underlying AI models score better than previous versions on benchmarks like FrontierMath and Humanity’s Last Exam, according to OpenAI.

Safety Measures Implemented

Thanks to the vast amount of knowledge the agent consults, additional safety measures have been implemented. A monitor screens all prompts in real-time for sensitive topics. The memory function of ChatGPT is disabled to prevent data leaks. If you let the agent use the web via the ChatGPT browser, all your actions and entered text remain private.

The agent is available for paying users of the Pro, Plus, and Team plans. In July, Enterprise and Education users will also gain access. Pro users have an unlimited number of tasks per month, while others get 50 tasks per month. Whether the agent can reliably complete tasks in practice remains to be seen.