Google is shaking WIRED has learned that it will strengthen the team behind Project Mariner, an AI agent that can navigate the Chrome browser and perform tasks on the user’s behalf. In recent months, some Google Labs employees who worked on the research prototype have moved on to higher-priority projects, according to two people familiar with the matter.
A Google spokesman confirmed the changes, but said the computing capabilities developed with Project Mariner would be incorporated into the company’s agent strategy in the future. Google has already incorporated some of these features into other agent products, including the recently launched one Agent of Gemini– added the spokesman.
The change comes as Google and other AI labs rush to respond to the emergence of highly proficient agents like OpenClaw. While these tools are currently mostly used by developers, Silicon Valley believes they could soon complement general-purpose assistants for people and businesses. Nvidia CEO Jensen Huang compared the buzzy tool to a recent operating system for agent computers. “Every company in the world today needs to have an OpenClaw strategy,” he said at the company’s developer conference earlier this week.
Google CEO Sundar Pichai discussed Project Mariner at last year’s I/O conference. At the time, browser agents seemed like the next gigantic thing in the industry, and OpenAI and Perplexity launched consumer agents that promised to automate online tasks for users. Agents could click, scroll and fill out forms on the website just like a human would. However, the adoption of these products has not met industry expectations.
In December 2025, Perplexity’s Comet browser agent reached just 2.8 million weekly vigorous users. Meanwhile, OpenAI’s ChatGPT agent is reported to have dropped to less than 1 million weekly active users in recent months. Compared to the hundreds of millions of users chatting with ChatGPT every week, browser agent usage essentially comes down to a rounding error.
Novel agents in town
In the last year, the dynamics in the world of artificial intelligence have changed dramatically towards agents like Claude Code and OpenClaw (whose creator was hired by OpenAI). Unlike web browsing agents, these systems control computers via the command line, which has proven to be a more reliable way to perform tasks. Some of these products include, but are not constrained to, the ability to exploit a computer. In comparison, browser agents currently seem somewhat constrained as a standalone product.
“Claude Code and OpenClaw showed that it’s actually much more efficient to work with the terminal because the terminal is text-based and LLMs are text-based,” Katanforoosh said. “It probably takes 10 to 100 times fewer steps to achieve the same results.”
This doesn’t mean that browser agents aren’t improving or that computer usage research has reached a dead end.
Last month, startup Standard Intelligence published the file computer use model trained from videos, not screenshots. The startup claims to have developed a video encoder that can compress videos into a contextual window AI model that it claims is 50 times more proficient than previous desktop usage models. To showcase the capabilities of its AI model, the startup connected it to a car, a live video feed, and a computer keyboard. The model was able to briefly drive autonomously around San Francisco.