ChatGPT Agent is now able to interact with websites and take actions like clicking on buttons, sorting through content, and extracting only the needed information. As the agent is within the chatbot itself, users are able to transition from conversation to requesting specific actions without exiting the chat. This also enables the agent to have more context.
The agent includes a set of tools that it utilizes to execute actions autonomously, the firm noted in a blog post. It includes a visual browser that the agent leverages in order to interact with the web, a text-based browser for accelerating the process of gathering information, a terminal, and direct application programming interface (API) access.
OpenAI unveiled a fresh artificial intelligence (AI) agent, to be baked into the company's chatbot, on Thursday. Named ChatGPT Agent, it's an all-purpose agent that has its very own virtual computer to use to surf the web and search for facts, plus a built-in development environment (IDE) to write code. The San Francisco AI company stated that ChatGPT Agent is really the combination of the Operator agent and the Deep Research feature. The new agentic feature can be accessed by the paying subscribers on the chatbot's web and desktop application.
The AI company introduced the new AI agent through a live stream. This is the third agentic product of the company (following Deep Research and Operator) and second standalone agent. OpenAI stated that ChatGPT Agent includes "Operator's ability to engage with websites, deep research's ability to synthesise information, and ChatGPT's intelligence and conversational fluency."
OpenAI explained that ChatGPT Agent also has the ability to use connectors to integrate with third-party applications like Gmail and GitHub. If a site prompts the agent to log in, the user can intervene and provide credentials on behalf of the agent. Of note, all of these tools are made available to the agent through a virtual computer. When it is working on a task, its users can observe a small window in the chat window to know what it is doing.
All this ability is directed towards performing much more sophisticated tasks than earlier released agents. The AI company emphasized the fact that the users can request the ChatGPT Agent to translate screenshots or dashboards into presentations comprising editable vector elements, reschedule meetings, organize and reserve event venues, insert new information into existing spreadsheets without losing formatting, and so on. The agent can also design an early retirement scheme by seeking out local tax rules and customised investment plans, plan and arrange travel schedules, parties, and even arrange and book visits with experts.
ChatGPT agent is already accessible to the Plus, Team, and Pro users. Although Pro users alre ady have access to it, other users are set to receive it within the next few days. Also, the agent will be rolled out to the Enterprise and Education levels soon. What is interesting is that Pro users get a 400-message monthly limit, while other plans get a 40-message monthly limit.
ChatGPT Agent is only accessible to paid subscribers
Pro subscribers receive a monthly allowance of 400 agentic messages
OpenAI is set to close down the Operator agent
The feature will not be accessible in the European Economic Area (EEA) and Switzerland, although OpenAI said it is trying to make it work in that area as well. With the ChatGPT Agent launch, the AI company is also going to shut down Operator in weeks to come.