Want to use the ChatGPT Agent more efficiently? Here's a rundown

OpenAI launches ChatGPT Agent, a new AI tool designed for complex tasks, functioning as a digital assistant capable of browsing, analyzing, and scheduling.

Meet OpenAI's ChatGPT Agent, the company’s newest tool designed to take on complex, multi-step tasks for you that don’t just chat, they act. Your “virtual computer” or Iron Man’s J.A.R.V.I.S. that can browse, analyze, schedule, plan, and even shop .

Rolling out to Pro users, ChatGPT Agent offers significant capabilities while incorporating safeguards to prevent misuse, marking a notable advancement in AI agents. (Pexels)

While sitting with The Verge, Yash Kumar and Isa Fulford, the product and research leads behind ChatGPT Agent, said the tool runs on a brand-new model that OpenAI built specifically for this project.

ChatGPT Agent is powered by reinforcement learning, the same technique used in other OpenAI reasoning models, and brings together tools like a text browser, visual interface, and terminal access so users can import and work with their own data. Essentially, it’s combining the best of OpenAI’s earlier tools: Operator and Deep Research.

Notably, in the demo, ChatGPT Agent planned a date night by syncing with Google Calendar and checking OpenTable for restaurant bookings.

ChatGPT Agent doesn’t just browse, it runs the show

Fulford mentioned she uses it for online shopping, saying it works “better and was more thorough” than just using Operator alone. Yash uses it to handle mundane tasks like requesting OpenAI office parking every Thursday. “Since ChatGPT Agent has access to ‘an entire computer’ instead of just a browser, they’ve ‘enhanced the toolset quite a bit,’” he said.

When asked about latency, Yash admitted the team is focused more on “optimizing for hard tasks.”

Fulford echoed the same, saying, “Even if it takes 15 minutes, half an hour, it’s quite a big speed-up compared to how long it would take you to do it.”

“Before ChatGPT Agent does anything ‘irreversible,’ like sending an email or making a booking, it asks for permission first.”

OpenAI adds safeguards to ChatGPT

Now, because this model is more capable than past versions, OpenAI has turned on specific safeguards meant to block potential misuse, especially when it comes to “high biological and chemical capabilities.” The company clarified it does not have “direct evidence that the model could meaningfully help a novice create severe biological or chemical harm” in the form of weapons.

Want ChatGPT Agent to handle financial tasks? Not yet. “For now,” Yash said, those actions are restricted. There’s also a feature called Watch Mode if you’re on a financial site, and you navigate away from the tab where Agent is running, the tool immediately stops.

OpenAI began rolling out ChatGPT Agent to Pro, Plus, and Team users just type “/agent” or select agent mode in the tools menu to try it. Enterprise and Education users can expect access later this summer, though there’s no launch date yet for the European Economic Area or Switzerland.