OpenAI is giving ChatGPT a major upgrade, and this time, it is not just answering questions. It’s taking action. Meet the ChatGPT agent: the AI assistant that does not just talk back, it works. This upgraded tool can now navigate your calendar, build presentations, and run code.
This is not OpenAI’s first rodeo with agents. But previous versions often felt more like overachieving interns, full of potential, yet prone to fumbling the basics. This time, OpenAI says it has built something smarter, faster, and much more useful. The ChatGPT agent rolls out Thursday for Pro, Plus, and Team subscribers. To use it, just switch on “agent mode” in ChatGPT’s tool menu and start typing like you normally would. No secret handshakes required.
It combines powers from past tools like operator, which lets the agent click around the internet like a human, and Deep Research, which reads through the chaos of the web and distills it into clean, bite-sized knowledge bombs.
Related: India: Perplexity Secret Weapon in the AI Race Against Open AI
As of right now, ChatGPT agent can do things like:
- Analyze your competitors and create a slide deck (without breaking a digital sweat)
- Connect to Gmail, GitHub, and more via connectors
- Execute code in a terminal
- Call APIs like a junior dev with 5 espresso shots
The company says the ChatGPT agent model scores 41.6% on Humanity’s Last Exam (roughly double previous models,27.4% on FrontierMath, previously considered a math Everest for machines. Obviously, with great power comes great… risk management.
OpenAI is playing it safe. The agent has been labeled “high capability” in potentially dangerous domains like bio/chemical weapons, not because it has done anything wrong, but because it could. So, the company installed safeguards:
- Real-time monitoring of user prompts (especially for bio-related topics)
- Double-checks on outputs that could pose a threat
- Memory is disabled by default to avoid any sneaky data-leaking hacks
It is a cautious and necessary move. After all, a tool this powerful in the wrong hands is the plot of half of Silicon Valley’s favorite dystopian dramas.The bigger question: Can it finally deliver on the AI agent hype? For years, tech companies have promised “AI that works for you”, not just chatbot therapy sessions or reworded Wikipedia entries. The reality? Most AI agents have crumbled when asked to do anything too real.
OpenAI thinks it’s cracked that ceiling. The ChatGPT agent isn’t just a shiny upgrade; it is OpenAI’s boldest shot yet at building an AI that doesn’t just respond… it acts. Now it’s your turn: Will you let an AI plan your next pitch deck… or your next dinner?
