OpenAI continues to release new products. And Operator was eagerly awaited.
OpenAI has just unveiled Operator, its first AI agent capable of performing concrete tasks on a computer.
To go further
“Operator” or how ChatGPT is preparing to take control of your computer very soon
This is a step in the evolution of artificial intelligence. To explain: until now, OpenAI offered ChatGPT, a rather passive conversational chat, since it could not perform specific tasks.
With Operator, we go from simple conversational assistants to true autonomous agents. Based on the GPT-4o model, Operator can browse the web, fill out forms and interact with different interfaces as a human user would.
The particularity of Operator is its ability to break down complex tasks into simple actions, thanks to its model Computer-Using Agent (CUA). Unlike traditional solutions that require specific APIs, Operator directly analyzes pixels on the screen to understand and interact with any GUI. We therefore see the mouse move and perform actions on its web browser.
Some examples? You can combine PDF files, compress images, take screenshots, send an email… you can combine everything to perform complex tasks.
-This is not the first AI agent. But Operator already outperforms its competitors like Computer Use from Anthropic or Mariner from Google DeepMind on several benchmarks, but it remains limited to browser use et requires a premium subscription at $200 per month.
The security implications have been studied by OpenAI, as the American firm explains on its blog. The company has implemented safeguards to prevent malicious use, including training the model to ask for confirmation before performing actions with external consequences.
AI agents will change our PCs and smartphones
The arrival of such AI agents is a very important step. Daily tasks such as restaurant reservations or shopping management can now be delegated to an AI, in order to free up time for higher value-added activities.
The example of Yash Kumar, a researcher at OpenAI, perfectly illustrates this potential: he uses Operator to automatically manage his restaurant reservations, a simple but time-consuming task that can now be fully automated.
However, it is important to note that this technology is still in its infancy. As Sam Altman himself points out, we must moderate expectations and not give in to the media hype. Errors remain possible and the tool still requires improvement.
This week, Samsung announced its Galaxy S25 which also integrates an AI agent based on Google Gemini, you can also let your smartphone perform actions on apps, without touching anything.
