Project Mariner → Agent Mode

In our view, agents are systems that integrate the intelligence of advanced AI models with access to various tools, enabling them to perform actions on your behalf while under your control.

Our initial research prototype, Project Mariner, represents a significant step forward in the development of agents with computer-use capabilities, allowing them to interact with the web and accomplish tasks for you. Since its release as an early research prototype in December, we have made considerable progress, incorporating new multitasking capabilities and a “teach and repeat” method. This method enables you to demonstrate a task once, and the agent learns plans for similar tasks in the future. We are now making Project Mariner’s computer-use capabilities available to developers through the Gemini API. Trusted testers, such as Automation Anywhere and UiPath, are already utilizing it to build new applications, and it will be more widely available this summer.

Computer use is one component of a broader set of tools necessary for the development of a thriving agent ecosystem.

Other essential tools include our open Agent2Agent Protocol, which allows agents to communicate with each other, and the Model Context Protocol introduced by Anthropic, which enables agents to access other services. We are excited to announce that our Gemini API and SDK are now compatible with MCP tools, effective immediately.

Additionally, we are starting to integrate agentic capabilities into Chrome, Search, and the Gemini app. For instance, a new Agent Mode in the Gemini app will assist you in accomplishing more tasks efficiently. If you are searching for an apartment, it will help find listings that match your criteria on websites like Zillow, adjust filters, and use MCP to access the listings and even schedule a tour for you. An experimental version of Agent Mode in the Gemini app will soon be available to subscribers. This feature is also beneficial for companies like Zillow, as it attracts new customers and improves conversion rates.

As this is a new and emerging area, we are excited to explore the best ways to bring the benefits of agents to users and the broader ecosystem.