It is reported that Google is developing an artificial intelligence system called 'computer-using agent'.
It is reported that it will initially only work in a web browser.
Google is expected to unveil its own interpretation of Rabbit's large action model in December, according to reports. This project, known as "Project Jarvis," aims to perform tasks for users, including gathering research, purchasing products, and booking flights. The information comes from three sources close to the project.
Based on a future version of Google Gemini, Jarvis would operate exclusively in a web browser, with particular optimization for Chrome. The tool is designed to help people "automate everyday web-based tasks," using the ability to capture and analyze screenshots to interact, whether by clicking buttons or entering text. In its current state, it reportedly takes "a few seconds" between each action.
Major artificial intelligence companies are developing models with features similar to those described by this tool. For instance, Microsoft is working on Copilot Vision, which will allow interaction with web pages being viewed. Meanwhile, Apple Intelligence is expected to recognize screen content and perform actions across multiple applications over the next year. Anthropic has launched a beta version of Claude that, while flawed, can perform actions using a computer, while OpenAI is also working on a similar version.
It is worth noting that Google's intention to present Jarvis in December may change, as the company is considering launching it to a limited number of testers to identify and address potential bugs.
