OpenAI has just introduced a powerful new feature inside ChatGPT called “Agents”, and it’s not just another upgrade. It’s a full-blown automation system that merges three of OpenAI’s key capabilities: ChatGPT, Deep Research, and Operator. This new blend allows users to delegate complex, multi-step tasks to ChatGPT, which can now execute them autonomously over the internet.
What Is the New ChatGPT Agent?
Think of it as an intelligent assistant on steroids. Inside the ChatGPT interface, users can now activate the “Agent” feature to plan and execute end-to-end tasks using real-time web access, third-party tools, and smart automation.
The Agent can:
• Browse the web for real-time information
• Write and execute code
• Use connected apps like Gmail, GitHub, and more
• Create documents, slides, and spreadsheets using tools integrated with LibreOffice
Interesting twist: While OpenAI has a strong partnership with Microsoft (the creator of Office), the choice to integrate LibreOffice, a competitor, raised some eyebrows. But it shows OpenAI’s commitment to open systems and flexible toolsets.
How It Works: Interactive and Controllable AI Execution
One of the standout features is that these agents operate through an interactive session. This means:
• You can pause, adjust, or stop the execution at any time
• You can provide additional inputs or completely redirect the task
• You receive live progress updates, and you can resume unfinished tasks later
Even though tasks may take time, the session remains transparent and editable, giving users full control over the process.
Built-In Security Against Prompt Injection Attacks
Since this is the first time OpenAI agents can autonomously perform actions on the internet, the company has added robust protections. Agents are trained to detect and avoid prompt injection attacks, a security threat where malicious content on websites tries to manipulate AI behavior, sometimes attempting to extract sensitive user data.
OpenAI’s Agents are built to ignore such traps and safeguard your privacy, even if they’re handling sensitive information like payment details for online shopping tasks (with prior approval, of course).
Can It Really Shop for You?
Yes. If granted prior authorization, the Agent can:
• Search for products
• Compare prices
• Make purchases using your saved payment info
It’s automation with a personal touch, and strict privacy guardrails.
Who Can Use It?
• ChatGPT Pro users: Get 400 Agent actions per month
• Other paid subscribers: Receive 40 Agent actions monthly
Rollout started today and will continue expanding to more users in the coming days.
A Glimpse Into the Future: Smart Browsers & AI-first Interfaces
In our opinion, this is OpenAI’s warm-up before launching its own AI-native web browser. Unlike Chrome or Safari, this browser could deliver a fully intelligent browsing experience, optimized for next-gen users (think Gen Z and beyond) who prefer automation and minimal manual effort.
Funny enough, during the live demo of Agents, the browser being used was Google Chrome, not a custom OpenAI browser. Also notable: three people on the stream were wearing glasses, one of which was a Ray-Ban (Meta is a major investor in Ray-Ban’s smart glasses). Coincidence or signal?
Final Thoughts
With tech giants like Meta aggressively recruiting AI talent, this launch feels like OpenAI’s bold response: “We’re still here. We’re still building. And we’re not done.”
ChatGPT Agents aren’t just smart. They’re autonomous. Controlled. Secure. And most importantly, they’re just the beginning.
