OpenAI introduces "ChatGPT agent" as the ultimate jack of all AI trades — with its own computer to check out your to-do list

OpenAI just concluded its much-anticipated livestream, where it launched a new AI agent dubbed "ChatGPT agent". The tool ships with sophisticated capabilities that allow it to perform a wide range of computer-based tasks for users, including scheduling through your calendar and generating a briefing based on upcoming events, running code, and generating presentations and slideshows that you can edit for fine-tuning.

It's also worth noting that ChatGPT agent is shipping with several capabilities from other agents that OpenAI had already launched this year, including Operator and Deep Research.

The former is help bolster ChatGPT agent's capabilities since it is designed to control computers and take over tasks like coding or make travel bookings autonomously. On the other hand, Deep Research helps users conduct in-depth research on the internet for complex tasks.

OpenAI has simplified ChatGPT agent's user experience by allowing people to interact with it by simply prompting ChatGPT in natural language. The tool is already shipping for users with ChatGPT Pro, Plus, or Team subscriptions.

You'll need to navigate ChatGPT's dropdown menu of tools to activate ChatGPT agent by selecting agent mode.

How does ChatGPT agent work?

The agentic tool uses a visual browser that scours the internet via a typical graphical user interface (GUI), a text-based browser, a terminal, and direct API access. It will also leverage ChatGPT connectors which will allow users to connect their productivity apps like Gmail and GitHub to ChatGPT. As such, it'll be able to access critical information and perform tasks based on the user's request.

ChatGPT agent also shows great promise, at least per the benchmarks shared. The toll scored 41.6% on Humanity’s Last Exam, which is arguably one of the most difficult tests comprising thousands of questions across a wide range of subjects. ChatGPT agent's performance blows OpenAI's o3 and o4-mini models out of the water.

It also scored an impressive 27.4% on the FrontierMath benchmark (one of the hardest math benchmarks). However, it is worth noting that it had access to a terminal for execution.

What about safety?

OpenAI says that its new tool was developed with safety in mind, especially since its sophisticated capabilities can be leveraged by bad actors to cause catastrophic harm.

The ChatGPT maker detailed that it has put elaborate measures and safeguards to prevent the agentic AI from spiralling out of control, including a real-time monitor that tracks the user's interactions with the tool.

It'll have the capability to identify whether a prompt or request is biology-related, which will then trigger it to a second monitor, which determines whether the information could be used to cause harm.

Additionally, the AI firm indicated that it has disabled its memory feature in ChatGPT agent to help curb misuse. For context, the memory feature allows the chatbot to retain important details from previous conversations for personalized and relevant responses.

OpenAI says the move is designed to keep attackers at bay, preventing them from using malicious injection attacks to gain unauthorized access to users' sensitive data.

ChatGPT agent could potentially present OpenAI's move from a chatbot that simply generates responses based on queries to a more capable tool with real impact on productivity. But trust, safety, and privacy still remain critical concerns for most users.