During Google I/O 2025, Google announced that it will launch Project Project Mariner, the company's experimental AI agent, which browses and uses the website to more users and developers. Google also said that this has greatly updated the way Project Mariner works, allowing agents to perform nearly ten tasks at a time.

New subscribers to the new US $249.99 AI Ultra program will visit the Mariner project, and the company says support for more countries is coming. Google also said it brings the functionality of the Mariner project to the Gemini API and Vertex AI, allowing developers to build applications powered by agents.

Project Mariner debuted in the second half of 2024, representing Google's greatest effort and has not yet revamped how users interact with the Internet through AI agents. Google Search head said at the release that they view Project Mariner as part of the basic user experience transfer, and people delegate more tasks to AI agents rather than visiting websites and completing them themselves.

For example, Project Mariner users can buy tickets to baseball games or buy groceries online without visiting a third-party website - they just chat with Google's AI agents and then visit the website and take action for them.

Project Mariner competes with other web browsing AI agents, such as OpenAI operators, Amazon's Nova Act and Anthropic's computer use. These tools are all in the experimental stage, and TechCrunch's experience proves that prototypes are slow and prone to errors.

However, Google said it has received feedback from early testers to improve the capabilities of project sailors. A Google spokesperson told TechCrunch that the company updated Mariner's Project Mariner to run on virtual machines in the cloud, just like the agents of OpenAI and Amazon. This means that users can work on other projects when Project Mariner completes tasks in the background - Google says the new project Mariner can handle 10 tasks at the same time.

This update makes Project Mariner more useful than its predecessor, which runs on the user's browser. As I pointed out in my initial comment, the early design of Project Mariner means that users cannot use other tags or applications on their desktops while the AI ​​agent works. This beats the purpose of AI agents - it will work for you, but there is nothing you can do when you work.

In the coming months, Google said users will be able to access Project Mariner in an AI model (the company's AI-powered Google search experience). When launched, the feature will be limited to Search Labs, which is an opt-in test ground for Google's search feature. Google said it is working with Ticketmaster, Stubhub, Resy and Vavaro for some agents to flow.

Today, Google has launched an early demonstration of another proxy experience called the "Proxy Mode". The company says this feature combines web browsing with research features and integrations, as well as other Google applications. Google says Ultra users will soon access proxy mode on the desktop.

In this year's I/O, Google seems to be finally willing to ship the proxy experience that has been talking about for years. Project Mariner, Agent Mode and AI models all seem ready to change the way users browse the network on the web and how vendors interact with customers online. Web browsing agents have a big impact on the internet economy, but Google seems ready to get all of them out.