OpenAI is trying hard to break through the walls of a chatbot and expand to every corner of your device. On Thursday, the company ChatGPT Agent is its entry into the “agenttic” space, which allows artificial intelligence models perform specific tasks.
The company claims that the ChatGPT agent uses a virtual machine to complete specific requests. The company gave examples such as “look at my schedule and inform me about upcoming client meetings using recent news,” “plan ingredients for a Japanese breakfast for four” and “analyze and create a PowerPoint deck.” Once given a task, ChatGPT agent acts independently, without additional input, to navigate websites, to open apps, and to perform the tasks that were given. The Verge reports that ChatGPT Agent is built on an unnamed, new model that OpenAI created specifically for this project. It combines the functionality from its Operator Tool, which can navigate through web browsers, with Deep Researchwhich can complete research and analysis tasks in multiple steps. Both features were introduced earlier this summer. The Agent tool has also been trained to perform complex tasks, such as switching between a terminal and a browser.
Initial impressions of the Agent are mixed. The Verge reported while it seems capable of handling multi-step operations it can be slow in completing its tasks and it is not designed for people to sit and watch it work. Wiredalso said that it isn’t exactly blazingly fast. Isa Fulford of OpenAI, the research lead for the company, took an hour to order a dozen cupcakes. She said it was still a time-saver because she “didn’t want to” do it. Wired highlighted the “replay feature” which allows you to replay the steps the agent takes to complete its tasks.
In the world of AI agents are the rage, and they’re often used to replace humans for certain tasks. Klarna, as an example, used AI agents to replace its customer service operation, but later decided to bring back human operators because of the quality of the work they provided. ChatGPT Agent is available to those who subscribe to a ChatGPT Plus, Team, or Pro plan. They can access it by selecting “agent modes”. The company will introduce the agent to ChatGPT Enterprise and Educational users later this summer.
