Google is expanding its AI capabilities by integrating Project Mariner's computer use features into its Gemini API and Vertex AI platforms, representing a major step forward in the development of agentic AI systems.
Project Mariner, first unveiled in late 2024, is Google DeepMind's research prototype that explores human-agent interaction through web browsers. The system can observe what's displayed in browsers, interpret complex goals, plan actionable steps, and navigate websites to complete tasks autonomously. It can handle multiple operations simultaneously, with the latest version capable of completing up to ten different tasks at once.
Several technology companies are already exploring Project Mariner's potential. Automation Anywhere, a leader in agentic process automation, and UiPath, known for its automation platform, are among the early adopters. Other partners include Browserbase, which develops AI browser automation frameworks, Autotab, The Interaction Company, and Cartwheel, a text-to-animation platform founded in 2023.
The integration with Gemini API and Vertex AI will allow developers to build applications powered by these agent capabilities. Google has also significantly enhanced security protections against threats like indirect prompt injections, making Gemini 2.5 its most secure model family to date.
Google AI Ultra subscribers in the US already have access to Project Mariner, with broader developer access planned for this summer. The company is also bringing some of Mariner's capabilities to other Google products, including AI Mode in Search Labs, where it will enable tasks like purchasing event tickets and making restaurant reservations.
This development represents a fundamental shift in how users interact with the internet, potentially moving from direct website interaction to delegating tasks to AI agents. As these capabilities mature, they could revolutionize automation across industries and enable entirely new applications for AI assistants.