Google shuts down Project Mariner: the browser AI agent is folded into Gemini

On May 4, 2026, Google shut down Project Mariner, a browser-based AI agent that could perform online tasks on a user's behalf. A short message appeared on the project's main page: "Project Mariner was shut down on May 4, 2026. Its technology continues to work in other Google products." This is not the end of Mariner's story; it is the move to its next stage inside the company's corporate structure.

A year of intensive development: from an idea to parallel tasks

Project Mariner debuted in December 2024 as one of Google's most ambitious AI experiments. The concept was simple but innovative: an AI agent sees the screen the way a person does, clicks buttons, fills in web forms, reads the results and reports back on the work done. The first months were modest in capability: the agent handled one task at a time, learning to interact with different interfaces.

Over the year, though, the Google team substantially expanded what Mariner could do. At Google I/O 2025, the company presented a significant update: the agent could now run up to 10 tasks in parallel. It was a leap in performance. Mariner could open several browser tabs, work in different web services at once and sync results between them. In public demos, Mariner ordered food for delivery, bought plane tickets and compared product prices, all in a single browser, the way a person would, but faster and more accurately.

Google released impressive videos. Investors saw it: this was not just a lab toy but a potentially useful tool. Here is the full list of capabilities Mariner accumulated:

• Ordering food and goods through marketplaces, delivery services and online stores
• Searching for and buying tickets for flights, trains and events
• Comparing prices and product specs across different stores
• Filling in tax forms and various documents
• Automating routine operations in web applications

The technology isn't going away; it's being absorbed into

Google's ecosystem

Google did not simply switch Mariner off as a failed experiment. The company integrated its components into the existing ecosystem. The main beneficiary is Gemini Agent, which has now gained Mariner's browser capabilities. Other Google AI tools will also receive pieces of the technology.

This is a typical pattern for Google Labs. Experimental projects rarely die. They either become full products, migrate to existing platforms, or have their technology distributed across the company. Mariner took the integration route, which makes financial and product sense. Browser agents are needed across Google's lineup, and maintaining a separate project is inefficient.

Browser agents move from the lab into the real world

Mariner was an experiment, but 2026 shows that browser-based AI agents are moving from curiosity to real functionality. OpenAI released Operator, a built-in feature for web tasks. Google is integrating Mariner's legacy into Gemini. Other companies are working on similar tools. This is no longer experimental territory; it is a new market.

For users, this means one thing: routine tasks such as booking tickets, filling in forms and searching for information may soon pass entirely to an AI assistant. A person gives a command, the agent carries it out in the browser and returns the result. That would save time and reduce errors.