Google has shut down Project Mariner as a standalone product, according to reporting published on May 6, 2026, and the Project Mariner landing page now states that the experiment was shut down on May 4, 2026. The move matters because Mariner was Google’s most visible browser-based AI agent experiment, and its capabilities had already been spreading into Gemini, AI Mode, and other Google surfaces before the shutdown became public.
That makes this more than a product retirement. It is a signal that Google no longer wants browser agents to live as isolated demos. Instead, it appears to be folding the most useful parts of Mariner into a broader agent stack built around Gemini.
What Project Mariner was supposed to be
Google introduced Project Mariner in December 2024 as an early research prototype built with Gemini 2.0. The core idea was ambitious: let an AI agent understand what is on a browser page, reason across text, images, forms, and code, and then take actions through an experimental Chrome extension while keeping the user in control.
From the beginning, Google positioned Mariner as a browser-native agent for complex multi-step work rather than a normal chatbot. The company said it could navigate web pages, fill forms, and complete real-world tasks, while asking for confirmation before sensitive actions.
By May 20, 2025, Google had expanded the concept. Project Mariner had evolved into a system of agents that could complete up to 10 tasks at a time, including research, bookings, purchases, and other multitask workflows. That update made Mariner look less like a one-off lab demo and more like an early operating model for consumer-facing agents.
Why Google is folding Mariner into Gemini
The most important clue came before the shutdown. In December 2025, Google described Gemini Agent inside the Gemini app as being built on insights from Project Mariner, with support for multi-step task handling across tools such as Deep Research, Canvas, connected Workspace apps, and live web browsing.
In other words, Mariner had already started turning into infrastructure. Instead of asking users to learn a separate browser-agent brand, Google was moving the same underlying ideas into products with much bigger distribution.
That strategy makes sense. Standalone agent experiments are useful for proving what is technically possible, but they are harder to scale than agent features embedded inside products people already use every day. Gemini, Search, Chrome, and Workspace give Google far more reach than a separate labs-style destination ever could.
The shutdown also simplifies Google’s AI story ahead of Google I/O 2026, which starts on May 19. Rather than juggling too many overlapping agent brands, Google now has a cleaner narrative: Gemini is the front door, and browser action is one capability inside a wider assistant system.
What this changes for the browser-agent market
Project Mariner’s end does not mean Google is backing away from agentic browsing. It means the company seems to believe browser control is becoming a feature, not a product category on its own.
That is a meaningful shift for the wider market. For the last year, browser agents have often been framed as headline-grabbing demos: agents that shop, research, compare listings, or fill out workflows across the web. But as the category matures, the winners may be the companies that make those actions feel native inside larger software ecosystems.
Google’s own product history supports that reading. Project Mariner began as a research prototype, then its computer-use ideas were pushed into the Gemini API, Gemini Agent, and AI Mode. Shutting down the standalone surface suggests Google thinks the distribution advantage now matters more than preserving the experimental label.
For competitors, this raises the bar. Browser agents are no longer just about raw autonomy or flashy demos. They increasingly need identity, permissions, multi-tool orchestration, and a clear home inside a product people already trust enough to let an agent act on their behalf.
What businesses should watch next
The next question is not whether Google still believes in browser agents. It clearly does. The question is where those capabilities show up next, and under what controls.
Businesses should watch for four things at and around I/O 2026:
- Deeper Gemini integration: More browser and task execution features may appear inside Gemini rather than under separate experimental names.
- More explicit Chrome workflows: Chrome is still the most natural place for web automation, so Google may keep shipping browser action there without reviving the Mariner brand.
- Tighter confirmation and policy controls: The closer an agent gets to transactions, account actions, and workflow automation, the more Google will need stronger guardrails and user approval patterns.
- Clearer developer surfaces: If Google wants third parties to build on these capabilities, it will need to keep exposing them through Gemini and related APIs instead of isolated research programs.
The practical takeaway for AI teams is straightforward: do not mistake the death of a prototype brand for the death of the capability category. Google appears to be consolidating browser-agent behavior into a larger Gemini operating model, not abandoning it.
That matters for enterprise AI, too. The future of agent systems is looking less like a collection of disconnected bots and more like one coordinated work layer that can research, reason, browse, and act across tools with humans still in control. Project Mariner may be gone as a product name, but its underlying role in Google’s agent stack looks far from over.