At Google I/O 2025, the company announced major updates to Project Mariner, its experimental AI agent that can browse the internet and complete tasks across websites on users’ behalf. First introduced in late 2024, Mariner is now being rolled out more broadly—with a suite of performance upgrades, deeper integration into Google’s AI ecosystem, and access for paying subscribers and developers alike.
Subscribers to Google’s new $249.99/month AI Ultra plan in the U.S. are the first to gain access to Project Mariner. Global availability is expected to follow. Google is also integrating Mariner’s capabilities into the Gemini API and Vertex AI, opening the door for developers to build powerful agent-driven applications that interact directly with the live web.
The company says this marks a shift in how users interact with online content: rather than manually visiting sites and navigating interfaces, people will increasingly delegate multi-step tasks—like booking event tickets or shopping for groceries—to AI agents that browse and act autonomously on the web.
Mariner Gets Smarter, Faster, and Truly Autonomous
In response to early tester feedback, Google made substantial backend changes to Project Mariner. Most notably, it now runs on cloud-based virtual machines, similar to competitors like OpenAI’s Operator and Amazon’s Nova Act. This upgrade allows users to offload work to Mariner in the background while continuing with other activities on their devices—something the earlier browser-based version didn’t support.
This update also unlocks a key capability: Mariner can now juggle up to 10 concurrent tasks, making it significantly more practical for complex or time-consuming workflows. Whether it’s comparing hotel prices, placing online orders, or scheduling multiple appointments, the AI agent can now manage simultaneous actions across different websites.
The broader rollout includes integration with major platforms like Ticketmaster, StubHub, Resy, and Vagaro, enabling more reliable task automation for bookings, purchases, and reservations. These partnerships signal Google’s intention to make Mariner a seamless interface layer between users and the digital economy.
Agent Mode and AI Mode Signal Bigger Shifts Ahead
Alongside Mariner, Google previewed Agent Mode, a new experience that blends Mariner’s web navigation abilities with research tools and other Google app integrations. While still in demo stages, Agent Mode promises to unify productivity and automation in a single agent interface. Google plans to release Agent Mode to AI Ultra subscribers on desktop soon.
Meanwhile, AI Mode, a new AI-enhanced Google Search experience, is being introduced via Search Labs. It will offer select users the ability to trigger Mariner-style actions directly from the search page—another sign that Google sees agents not as add-ons, but as the future of how we use the web.
With Project Mariner, Agent Mode, and AI Mode all moving into public-facing products, Google is officially making the leap from showcasing agents to shipping them. While competitors still struggle with slow, mistake-prone prototypes, Google’s investment in infrastructure, developer tools, and partner ecosystems gives it a strong edge.
Whether users fully embrace AI agents that browse on their behalf remains to be seen, but Google appears ready to lead the charge—and redefine the next chapter of human-computer interaction.
