Physical Address
304 North Cardinal St.
Dorchester Center, MA 02124
Google is shaking The team arrives behind Mariner Projecthis own AI agent that can navigate through… Chrome browser and completing tasks on the user’s behalf, WIRED has learned. In recent months, some Google Labs employees who worked on the search prototype have moved on to higher-priority projects, according to two people familiar with the matter.
A Google spokesperson confirmed the changes, but said the computing capabilities developed under Project Mariner will be integrated into the company’s agent strategy moving forward. Google has already integrated some of these capabilities into other proxy products, including the recently launched one Gemini agentThe spokesman added.
This change comes as Google and other AI labs rush to respond to the emergence of highly capable agents such as OpenClaw. While these tools are mostly used by developers today, Silicon Valley believes they may soon be able to power general-purpose assistants for people and businesses. Nvidia CEO Jensen Huang compared the buzzy tool to a new operating system for proxy computers. “Every company in the world today needs an OpenClaw strategy,” he said at the company’s developer conference earlier this week.
Google CEO Sundar Pichai highlighted Project Mariner during the conference I/O conference last year. At the time, browser proxies seemed like the industry’s next big bet, with OpenAI and Perplexity launching consumer proxies that promised to automate online tasks for users. Agents can click, scroll, and fill out forms on a web page, just like a human. However, the adoption of these products has faced difficulties to meet industry expectations.
Perplexity’s Comet browser agent has just arrived 2.8 million weekly active users In December 2025. Meanwhile, OpenAI’s ChatGPT proxy was reportedly down Less than 1 million weekly active users In recent months. Compared to the hundreds of millions of users who talk to ChatGPT weekly, using a browser proxy essentially amounts to a rounding error.
New agents in town
Momentum in the AI world has shifted dramatically in the past year toward clients like Claude Code and OpenClaw (whose creator was hired by OpenAI). Unlike web browsing agents, these systems control computers through the command line, which has proven to be a more reliable way to complete tasks. Some of these products include the use of a computer as a feature, among other agent capabilities. In comparison, browser proxies now seem to be somewhat limited as a standalone product.
One reason computer-based agents haven’t taken off is their huge computational requirements, says Kian Katanforoush, CEO of AI optimization platform Workera and who lectures on AI at Stanford University. Most of these agents work by taking a series of screenshots of a web page, feeding them into an AI model, and then taking actions based on what they see. Processing all this information can be slow and clunky at times Unreliable.
“What CloudCode and OpenClaw have shown is that it’s actually much more efficient to work with the terminal, because the terminal is text-based and LLLMs are text-based,” Katanforous said. “It would probably take 10 to 100 times fewer steps to get the same results.”
This does not mean that browser agents are not improving, or that research into computer usage has reached a dead end.
Last month, startup Standard Intelligence released a dossier Computer usage model They were trained on videos, rather than screenshots. The startup says it has developed a video encoder that can compress videos in the context window of an AI model, which it claims is 50 times more efficient than previous computer usage models. To demonstrate the capabilities of its AI model, the startup connected it to a car, a live video stream, and a computer keyboard. The model was able to briefly Drive independently throughout San Francisco.