CNN
—
OpenAI on Thursday announced a new feature for ChatGPT that allows the popular chatbot to execute actions on a user’s behalf. It’s part of an industry-wide push to change the way people get things done on the Internet: Tech giants hope that instead of bouncing between apps and manually searching the web, users might be able to one day rely on agents to do it all.
ChatGPT’s new agent mode, which begins rolling out immediately, is another sign that tech giants are doubling down on digital helpers that demonstrate significantly advanced capabilities. It also heightens the race between OpenAI and Google, which is pursuing similar ambitions with its Gemini helper.
OpenAI said on Thursday that ChatGPT’s new agent mode “thinks” and “acts” using its own virtual computer, enabling it to handle complex action-oriented requests. For example, users will be able to issue command such as “look at my calendar and brief me on upcoming client meetings based on recent news” or “plan and buy ingredients to make Japanese breakfast for four,” the company said in a blog post.
In a video demonstration, OpenAI employees wrote a long and detailed prompt asking the agent to help the user prepare for a wedding. It included a set of specific instructions such as “find an outfit that matches the dress code,” adding that it should propose five options, along with hotels that can accommodate a couple of buffer days around the event.
The new feature is available for those who subscribe to a Pro, Plus or Team plan. It builds on and combines capabilities from the ChatGPT Operator and Deep Research tools OpenAI already offers; Operator browses the web, while Deep Research analyzes online resources to do things like compile reports.
The update is another step in OpenAI’s efforts to turn ChatGPT in a more comprehensive universal assistant. At the same time, the broader AI industry is also grappling with how to address important shortcomings and privacy concerns around the technology. AI models are still prone to hallucinations and bias and can act in unpredictable ways, as xAI’s Grok chatbot demonstrated last week when it spewed antisemitic content after being prompted to do so.
In a blog post, OpenAI acknowledged that ChatGPT’s new functionality presents new risks. It said it has limited the data the model has access to, and certain tasks – like sending an email – require the user’s oversight. The model is also trained to refuse “high-risk tasks” like bank transfers, the company says.
“I would explain this to my own family as cutting edge and experimental; a chance to try the future, but not something I’d yet use for high-stakes uses or with a lot of personal information until we have a chance to study and improve it in the wild,” OpenAI CEO Sam Altman said in a post on X announcing the agent.
He advised users to be cautious when giving ChatGPT access to personal information. For example, granting access to a calendar to coordinate a group dinner might make sense, but the agent wouldn’t need calendar access to shop for clothes on a user’s behalf.
The announcement comes as tech giants are increasingly pushing to develop AI agents as they seek to win the AI race. Google made a flurry of AI-related announcements during its developer conference in May, including an agent that can make restaurant reservations and buy event tickets, among other tasks. Apple is working on a more advanced version of Siri that can use apps on a user’s behalf, although that update is delayed indefinitely.
