Project Mariner is an experimental AI agent from Google that can perform up to ten tasks simultaneously.
During Google I/O 2025, Google announced that it is making Project Mariner, its experimental AI agent capable of visiting and using websites, more widely available to users and developers. The tool is simultaneously receiving a significant upgrade: Mariner can now perform up to ten tasks concurrently in the background.
Surfing on your Behalf
Project Mariner is part of Google’s vision for agentive AI, where users have AI perform tasks instead of visiting websites themselves. Buying tickets, ordering groceries, or making restaurant reservations: soon all of this can be done through a conversation with the AI agent, without you having to take any action yourself.
The AI runs on virtual machines in the cloud and no longer locally on your PC. This allows users to continue working while Mariner gets to work.
Available for Ultra Subscribers
European users currently do not have access to the AI agent; it’s only available to American users of the new AI Ultra subscription, which costs $249.99 per month. Google promises that other countries will follow soon. Developers can also use the technology via the Gemini API and Vertex AI.
Soon, Mariner will also be integrated into AI Mode, the experimental AI version of Google Search. Agent Mode, another AI experiment that combines web navigation with research tools and Google apps, is also coming to desktop for Ultra users.
read also