Google is restructuring parts of its AI division, quietly moving team members away from Project Mariner, its experimental browser-based agent. The shift comes as the tech world pivots toward a new wave of AI systems championed by Jensen Huang.
According to reports from Wired, several employees from Google Labs who were working on Mariner have been reassigned to higher-priority initiatives. Google confirmed the move, stating that Mariner’s core features will continue within its broader AI ecosystem, including integrations with its newer Gemini Agent tools.
Project Mariner, unveiled at Google I/O, aimed to create AI agents capable of browsing websites, clicking buttons, and completing tasks automatically. Despite the promise, adoption across the industry remained limited. Competing tools like Perplexity’s Comet and OpenAI’s ChatGPT Agent struggled to reach significant user bases compared to traditional chatbot usage.
A key issue lies in how browser agents operate. They rely on repeatedly capturing and interpreting screenshots, making them slower, resource-intensive, and prone to mistakes. Experts, including Stanford instructor Kian Katanforoosh, note that text-based environments—like terminals—are dramatically more efficient for AI systems.
This shift in thinking has fueled rapid interest in OpenClaw, an open-source platform developed by Peter Steinberger. OpenClaw enables users to deploy autonomous agents directly from a terminal, allowing them to perform complex, multi-step tasks with greater speed and reliability.
At Nvidia GTC, Huang highlighted OpenClaw as a transformative development, comparing its potential impact to foundational technologies like Linux. Nvidia has also introduced NemoClaw, an enterprise-ready version designed with added security and control features.
The platform’s rapid growth has been striking, with widespread adoption globally, including strong backing in China from major players like Baidu and Tencent, as well as government-supported initiatives.
Other AI companies are following suit. Anthropic is expanding its Claude tools, while OpenAI is working to integrate similar capabilities into future versions of ChatGPT. Even Perplexity has begun shifting toward terminal-based solutions.
The broader trend is clear: AI agents operating in text-based environments are proving faster, cheaper, and more dependable than browser-based systems. For Google, scaling back Project Mariner reflects not a failure, but a strategic adjustment as the industry rallies around this new direction.
About the Author
