At today's Google I/O 2025 conference, Google officially announced the launch of a new project called "Project Mariner," marking a new step for the company in the field of artificial intelligence. This project was originally named "Jarvis" and underwent testing at the end of last year; now it has reached a more mature stage.
Project Mariner is a powerful web AI agent that is expected to be fully launched this year, aiming to optimize users' browsing experience. It can interact with browsers and other software, supporting simultaneous processing of up to 10 tasks. Users only need to show Mariner one task, and it will learn and predict future scenarios, providing intelligent solutions.
Currently, Mariner has successfully integrated with the Gemini API and has been tested with partners such as UiPath. The Gemini SDK has also achieved compatibility with the MCP protocol, meaning that the agent will be able to communicate with APIs from different applications in multiple ways. Additionally, the agent mode will soon be available in Chrome, search, and Gemini apps, further enhancing user convenience.
At the conference, Google demonstrated practical use cases of the agent in the Gemini app. For example, when users ask Mariner to help find an apartment, the agent not only searches on Zillow but also manipulates the computer to adjust filters and schedule viewings through the MCP protocol. This experimental feature is expected to be available to Gemini app subscribers soon.
Google's exploration in the field of artificial intelligence continues to deepen. The launch of Project Mariner undoubtedly provides users with a more efficient and intelligent online experience. As technology advances, we may see more similar innovative tools in the future, changing the way we interact with the internet.