Google reshuffles Project Mariner team amid the OpenClaw frenzy
Artificial Intelligence never stops surprising us, and this week brought a move that deserves serious attention from anyone following the industry closely.
Google confirmed it has reorganized part of the team behind Project Mariner, its AI-powered browser agent capable of operating Chrome and carrying out tasks on the web on behalf of users.
According to sources familiar with the matter, confirmed by a company spokesperson, some team members who had been working on the project were redirected to initiatives considered higher priority within Google Labs.
But hold on — this doesn’t mean Google is sitting on the sidelines when it comes to AI agents.
What’s actually happening is way more interesting than a simple layoff or abandoned project.
All the technology developed under Project Mariner is being absorbed into the company’s broader agent strategy, including confirmed integration with Gemini Agent, which was recently launched leveraging the advanced reasoning capabilities of Gemini 3.
And this move is coming at a very specific moment: the entire market has its eyes on OpenClaw 👀, a tool that has become the talk of Silicon Valley — so much so that Nvidia CEO Jensen Huang compared it to a new operating system for agentic computing during the company’s developer conference.
Receive the best innovation content in your email.
All the news, tips, trends, and resources you're looking for, delivered to your inbox.
By subscribing to the newsletter, you agree to receive communications from Método Viral. We are committed to always protecting and respecting your privacy.
The battle between different agent models is just getting started, and understanding what’s behind this reorganization goes a long way toward seeing where AI is headed in the coming months. 🚀
What Project Mariner was and why it mattered so much
Launched as a bold bet from Google Labs, Project Mariner was essentially a browser agent — an Artificial Intelligence system trained to browse the web autonomously, interact with pages, fill out forms, click buttons, and execute complex tasks inside the browser without the user having to do any of it manually. Think about what that means: you ask the agent to buy a product, schedule an appointment, or research information across multiple sites, and it just handles all of it while you focus on something else. That was exactly the level of autonomy the project promised, and it delivered some seriously impressive results during public demos, operating Chrome as if it were a real user with remarkable skill.
Google CEO Sundar Pichai highlighted Project Mariner during last year’s I/O conference, signaling that the company saw browser agents as one of the most promising bets for the future of intelligent assistants. At the time, that genuinely seemed to be where the market was heading, with OpenAI and Perplexity launching consumer-facing browser agents, all promising to automate online tasks. These agents could click, scroll pages, and fill in fields on websites in a way that closely mirrored what a human would do. It felt like we were on the verge of a revolution in how people interact with the internet.
The key differentiator of Project Mariner compared to previous browser automation attempts was how it processed visual and textual context from pages simultaneously. Rather than relying solely on HTML code to understand what was on screen, the agent used the multimodal capabilities of the Gemini model to literally see the interface and make decisions based on what it observed — much like a human would. This made the system far more adaptable to websites that don’t follow strict technical standards, which in practice represents the vast majority of the real web. This approach was closely watched by engineers at other companies because it solved a problem that had been holding back the development of functional browser agents for quite some time.
Beyond the technical side, Project Mariner held enormous strategic value for Google in the race for AI agents. While competitors explored agents focused on text or code tasks, Google was betting on something directly tied to its most valuable product: the browser. Placing an intelligent agent inside Chrome, which has billions of active users, was a move that created a distribution advantage extremely difficult to replicate. And that’s precisely why the team reorganization made so much noise — it touched something the market considered one of the company’s strongest bets for the near-term future of digital assistants.
The numbers that show why browser agents didn’t take off as expected
Despite all the initial excitement, public adoption of browser agents fell well below industry expectations. And the numbers tell the story. Comet, Perplexity’s browser agent, reached just 2.8 million weekly active users in December 2025. Sounds like a lot? Not when you put it in context. Meanwhile, OpenAI’s ChatGPT Agent dropped to under 1 million weekly active users in recent months. Compared to the hundreds of millions of people chatting with ChatGPT every week, browser agent usage basically amounts to a rounding error in the company’s metrics.
These numbers help explain why Google decided to reassess its priorities. It doesn’t make much strategic sense to keep a team exclusively dedicated to a product format that still hasn’t found meaningful traction in the market, especially when the underlying technology can be leveraged more efficiently within other initiatives.
OpenClaw enters the scene and shifts market priorities
If you’re not yet familiar with OpenClaw, it’s time to pay attention, because this name is going to show up more and more in conversations about the future of Artificial Intelligence. It’s a highly capable agent tool that arrived making waves across Silicon Valley. The comparison Jensen Huang, Nvidia’s CEO, drew between OpenClaw and an operating system for agentic computing wasn’t made lightly. At Nvidia’s developer conference this week, he was blunt: every company in the world needs an OpenClaw strategy. That kind of statement from one of the most influential leaders in the tech industry changes the game entirely.
The key advantage of OpenClaw and agents like Claude Code is that they don’t rely on visually navigating web pages. Instead, they control computers through the command line — the terminal. And that has proven to be a far more reliable way to complete tasks. As Kian Katanforoosh, CEO of AI upskilling platform Workera and an AI instructor at Stanford, explains, the reason is relatively straightforward: the terminal is text-based, and Large Language Models are also text-based. That natural compatibility makes the interaction much more efficient.
According to Katanforoosh, working through the terminal can be 10 to 100 times more efficient in terms of steps required to achieve the same results a browser agent would reach by visually navigating pages. That’s because browser agents need to take a series of screenshots, feed those images into an AI model, and then decide what actions to take based on what they see. This process burns through computational resources, is slow, and in many cases, unreliable. Terminal-based interaction strips away that entire visual layer and gets straight to the point, using plain text to execute commands and retrieve results.
The arrival of OpenClaw with this approach created visible pressure across the entire industry, including inside Google. When a new tool starts getting this kind of attention from names like Huang and begins circulating in investor presentations and technical events, product teams at major companies need to revisit their bets and adjust their roadmaps to avoid falling behind. It’s very likely that the decision to reallocate Project Mariner team members to other initiatives within Google Labs is directly connected to this competitive pressure — not because the project failed, but because the technology built within it needs to be absorbed into something bigger and more aligned with where the competition is happening right now. In tech, timing is everything, and Google clearly understood it couldn’t keep Mariner isolated as a separate experiment while the agent landscape was evolving rapidly all around it.
It’s also worth noting that OpenClaw isn’t competing directly with Project Mariner in terms of specific browser agent functionality. The battle is much broader than that. What’s at stake is which company or platform will define the standard for how Artificial Intelligence agents operate, communicate, and are managed in the years ahead. Whoever manages to establish that orchestration layer as the reference point will wield immense influence over how developers build agentic applications — the same way whoever controls the operating system influences everything that runs on it. Google, with all its infrastructure and the integration between Gemini, Chrome, and its own services, holds very strong cards in this game, and the reorganization around more strategic priorities appears to be a deliberate move within this larger contest. 🎯
Browser agents aren’t dead, but the game has changed
It’s important to be clear that this industry-wide shift in focus doesn’t mean browser agents are dead or that research into computer use has hit a dead end. In fact, there are some pretty interesting advances happening on this front. Last month, startup Standard Intelligence unveiled a computer-use model trained on videos instead of static screenshots. The company developed a video encoder capable of compressing audiovisual content within an AI model’s context window, which it says is 50 times more efficient than previous screenshot-based models.
And to showcase the potential of this technology, Standard Intelligence did something that sounds straight out of science fiction: they connected the model to a car, a live video feed, and a computer keyboard. The result? The model managed to autonomously drive through the streets of San Francisco for a brief period. This shows that AI computer-use research extends far beyond browsing websites, with potential applications reaching into the physical world in ways few people imagined just months ago.
What we’re seeing, then, isn’t the end of browser agents but a market reconfiguration. Agents that operate through the command line have proven more practical and efficient for the current moment, especially for developers and technical professionals. But that doesn’t invalidate research into visual agents. The most likely outcome is that the future of AI assistants will combine both approaches: agents that know how to use the terminal when that’s more efficient and that can navigate visually when a task requires interaction with graphical interfaces.
Where Mariner’s technology goes from here
The most important part of this whole story is the fate of the technology that was built over months of Project Mariner development. Based on what the Google spokesperson confirmed, the project isn’t being discontinued — it’s being integrated into the company’s broader agent strategy, with a particular focus on Gemini Agent. This makes a lot of sense from a product standpoint: maintaining a separate project for browser agents while Gemini evolves to become Google’s central hub for intelligent assistance would be inefficient and create unnecessary redundancies. By merging browser agent capabilities directly into the Gemini Agent framework, the company can deliver these features in a much more integrated and accessible way for users who already rely on Google products every day, without needing a separate, experimental tool.
This integration also opens up interesting doors for enterprise use and for developers working on automation. With Project Mariner’s capabilities inside the Gemini ecosystem, it becomes much more feasible to build workflows where AI browses the web, collects information, interacts with external platforms, and returns results within a single cohesive interface. This is something mid-size and large companies are actively seeking because it reduces the need for complex and expensive integrations with third-party APIs. The entire web effectively becomes an interface the agent knows how to use, and that significantly changes the cost and complexity equation for anyone looking to automate processes that depend on data scattered across different sites and systems. It’s a natural evolution and one that’s well-positioned for what the market is asking for right now.
What will be interesting to watch going forward is how this integrated version of Project Mariner’s capabilities stacks up against what OpenClaw and other competitors are delivering in terms of reliability, speed, and security in autonomous task execution. AI agents that operate browsers still face serious consistency challenges, especially on dynamic sites and in situations where a page changes mid-task. Solving those problems at scale — for millions of simultaneous users — is a very different kind of engineering than making things work in controlled demos. Google has the infrastructure to tackle that challenge, but the pressure from OpenClaw and other players will ensure the pace of evolution needs to be fast.
The agent race is defining AI’s next chapter
If there’s one thing this whole wave of activity makes clear, it’s that AI agents have become the most important battleground in the industry right now. It’s no longer just about which model can generate the best text or the most realistic image. The competition now centers on who will build systems capable of acting autonomously in the digital world, executing complex tasks with minimal human oversight.
Google, OpenAI, Anthropic with Claude Code, and the creators of OpenClaw are all racing in the same direction but with different approaches. This diversity of strategies is healthy for the ecosystem and tends to accelerate the evolution of every product involved. The Project Mariner team reorganization isn’t a sign of weakness from Google — it’s a sign of rapid adaptation to a landscape that shifts practically every week.
For anyone who uses technology daily, this competition is great news. More companies fighting for this space means better, faster, and more accessible products reaching the market. And given the speed at which everything is moving, it wouldn’t be the least bit surprising if within a few months we’re using AI agents to handle tasks that still feel futuristic today. The future of intelligent assistants is being written right now, and it’s being written at breakneck speed. 😊
