AI agents seem to be the next big trend in the artificial intelligence scene, and OpenAI did not hesitate to hop onto the bandwagon. The company has recently unveiled ChatGPT Agent, a tool that can perform actions on the user’s behalf. This new agent mode is equipped with its own virtual computer, allowing it to execute complex tasks.
The user can access this new feature at any point in a conversation by selecting “agent mode” in the tools dropdown. Then, the user can simply describe the task they want the agent to perform. ChatGPT Agent will explain what it is doing as it performs the task, and the user can interrupt the agent and take control when necessary. Additionally, the AI itself may ask for additional information to ensure that it is acting to serve the intended goal.
According to OpenAI, ChatGPT Agent can perform actions like checking the user’s calendar, downloading and manipulating files, and connecting to apps like Gmail and Github to gather more information. Certain tasks with real-world consequences like making purchases and sending emails still require user input, however. The AI will also refuse “high-risk” tasks as part of the safeguards against model mistakes.
To create ChatGPT Agent, the company combined the capabilities of its existing tools, namely Operator and Deep Research. The former can click around in websites, while the latter can analyse information gathered from multiple sources and condense it into a concise report. According to OpenAI, the end result is a natural evolution of the two tools.
ChatGPT Agent started rolling out to subscribers to OpenAI’s Pro, Plus, and Team plans on Thursday. The company also intends to bring this mode to Enterprise and Education users “in the coming weeks”.
(Source: OpenAI)
