OpenAI has just released one of the biggest updates on ChatGPT called Agents. It’s a fully functional AI agent that doesn’t just answer your questions; it works for you. The feature allows the AI to act on your behalf to complete multi-step tasks, like browsing the web, logging into accounts, running code, generating editable presentations, and updating spreadsheets, all with the help of its own virtual computer. So in this article, I have compiled the seven key things you need to know about this powerful new rollout:
It’s More Than Just a Chatbot
You can now ask ChatGPT to do things like: Brief you on client meetings based on your calendar and relevant news, or buy ingredients for a Japanese breakfast. And the agent will open websites, filter results, run code, and output formatted documents like slideshows or spreadsheets, all within one session.
It Combines Operator and Deep Research in One System
Earlier, OpenAI had an Operator for web interaction and Deep Research for analysis. However, the new ChatGPT agent merges both. It can now interact with websites and think critically about what it finds. Besides, you don’t have to choose between browsing and analysis, as it does both seamlessly.
You Stay in Control
While ChatGPT has some new power, it still requires permission before doing anything consequential. It will also allow you to take over, pause, or stop a task anytime. Apart from that, it will also notify you once the task is completed and will send a notification to your mobile as well.
It Comes with a Toolbox
The agent operates using a virtual computer and has access to a visual web browser, which is used for clicking, logging in, and navigating websites. It also comes with a text browser, which is for fast, reasoning-based queries, and includes a terminal for running code. Apart from that, to link Gmail, Google Calendar, GitHub, and more, it includes API connectors. This toolkit allows it to move between web scraping, data manipulation, and user-specific app integrations fluidly.
Outperforming Humans in Real-World Benchmarks
According to OpenAI, it beats humans on realistic data science tasks on DSBench. It also outperforms previous models on spreadsheet editing on SpreadsheetBench. The company also said that the agent matches or exceeds human quality in 50% of complex professional tasks.
It’s Already Available, But Only for Paid Users
The rollout is already live for Pro users, where they will get 400 messages/month. On the other hand, Plus and Team users will have to wait for a few days to receive access, and they will get only 40 messages/month. Apart from that, Enterprise and Education users will also get access soon.
Still a Work in Progress
However, it’s important to keep in mind that the feature is not complete yet. There are some limitations, like slide decks can feel unpolished and basic in layout, and discrepancies can occur between slide previews and exports. Apart from that, you can’t yet upload existing slide decks for editing.
But OpenAI says this is just the start, and future updates will bring cleaner outputs, more integrations, and smoother user control.
That’s all from OpenAI for now. But there are reports floating on the internet that suggest that the AI giant is also working on GPT-5 which will be the biggest model update in recent times. It is said to combine the advanced reasoning of its “O” series with the multi-modal capabilities found in the GPT-4 series. However, there’s no clarity on when this model is going to arrive. But I believe that we will not have to wait longer, and the launch is just around the corner. But until we get some official confirmation from the brand, it’s better to take all the information with caution.
If you liked this story and want to stay updated with all the OpenAI, ChatGPT, and other AI news, join us on WhatsApp. We share the latest tech news, in-depth reviews, analysis, and more.
