ChatGPT can now 'think and act' on your behalf: Here’s what you need to know

ChatGPT can now 'think and act' on your behalf: Here’s what you need to know

Announced this week on Thursday, the update allows ChatGPT to plan, reason, and execute multi-step actions using a new suite of tools and short-term memory.

ChatGPT can now think through tasks and take actions on your behalf, turning it from a smart chatbot into a capable digital assistant.

Announced this week on Thursday, the update allows ChatGPT to plan, reason, and execute multi-step actions using a new suite of tools and short-term memory.

In simple terms, it does not just respond to what you say, it stops to “think” and figure out what needs to be done and starts doing it for you.

From a chatbot to an AI assistant

Previously, ChatGPT was reactive: you gave it a task, and it gave you a response.

In a briefing with the Verge, the project lead Yash Kumar and the research lead on ChatGPT Agent, Isa Fulford confirmed that the new model can perform tasks like using at a user's calendar to brief them on upcoming client meetings, planning and perform tasks like looking at a user’s calendar to brief them on upcoming client meetings researching online and more.

“This is powered by a new system of built-in tools, which includes a code interpreter, a file handler, and a smart browser called Operator, along with a web research feature known as Deep Research,” they said.

For example, you could ask ChatGPT to “plan a weekend trip to Nairobi with a budget of Sh50,000”. Instead of just giving travel tips, it could:

Search for flights and accommodations, compare prices, organise everything into a shared itinerary, and create a checklist or budget spreadsheet.

The system behind this is known as the “Agent”, and it combines decision-making (the "planner") with execution (the "controller").

It is ChatGPT’s first big step into autonomous digital help.

Privacy and safety are built-in

The OpenAI team, made up of between 20 and 35 people across product and research, says safety is a top priority.

While ChatGPT can undertake actions, it asks for confirmation before doing anything irreversible, like sending emails or making purchases.

“Performing online transactions has been restricted for now, and there is an additional protection called ‘Watch mode’ where a user navigates to a certain category of web pages, like financial sites, they must not navigate away from the tab Chat GPT Agent is operating in, or the tool will stop working,” Kumar said.

They also said that the new tool does not access your long-term memory by default, and the short-term memory helps it stay focused on a task, but it does not keep a permanent record of what it sees or does during tool use, avoiding privacy risks.

Who will access it?

The update is rolling out to ChatGPT Plus, Pro, and Team users, with limits based on subscription level. If you are a free-tier user, you will still get regular ChatGPT, but without the ability to take actions or use advanced tools.

“Think of it as a virtual assistant with initiative, one that understands your goals and can start moving toward them, safely, intelligently, and efficiently.”

Reader Comments

Trending

Popular Stories This Week

Stay ahead of the news! Click ‘Yes, Thanks’ to receive breaking stories and exclusive updates directly to your device. Be the first to know what’s happening.