Andrii Bidochko

• Updated: March 20, 2026
• 5 min read

Google Restructures Project Mariner Team, Shifts Focus to Terminal‑Based AI Agents

Google has restructured the Project Mariner team, moving key researchers to higher‑priority initiatives and folding Mariner’s browser‑automation capabilities into its broader Gemini Agent strategy.

What’s happening with Google’s Project Mariner?

In a move that signals a strategic pivot, Google’s AI labs have shuffled the talent behind Project Mariner—the experimental AI agent that could navigate Chrome, click buttons, and fill out forms on a user’s behalf. According to Wired, several core engineers have been reassigned to projects deemed more critical to Google’s long‑term AI roadmap.

The company confirmed the changes, emphasizing that the underlying “computer‑use” technology will not disappear but will be integrated into the upcoming Enterprise AI platform by UBOS and Google’s own Gemini Agent suite.

Project Mariner: The original vision

Launched in early 2024, Project Mariner was Google’s answer to a wave of “web‑browsing agents” that promised to automate everyday online tasks. The prototype could:

• Open a web page, scroll, and locate relevant information.
• Interact with forms, submit data, and even complete purchases.
• Summarize the results and present them in a concise chat window.

At the 2023 I/O conference, Sundar Pichai highlighted Mariner as a potential “new operating system for the web,” positioning it alongside OpenAI’s OpenAI ChatGPT integration and Perplexity’s Comet browser agent.

The restructuring: Who’s moving where?

According to two insiders familiar with the internal shuffle, the following changes are taking place:

1. Key researchers from the Mariner prototype have been transferred to the Gemini Agent team, where they will focus on multimodal reasoning and tighter integration with Google Search.
2. Product managers are now overseeing a cross‑functional “AI Agent Platform” that consolidates browser‑automation, terminal‑based agents, and code‑generation tools.
3. Engineering resources previously dedicated to Mariner’s UI layer are being redirected to improve the Workflow automation studio, enabling enterprises to build custom agent workflows without writing code.

The shift reflects a broader industry trend: moving from visual, screenshot‑driven agents toward text‑centric, terminal‑based assistants that are more computationally efficient.

Why the industry is favoring terminal‑based agents

Since late 2024, AI labs have observed that agents which operate directly in a command‑line environment—such as AI YouTube Comment Analysis tool and Claude Code—achieve higher reliability with fewer computational steps. The reasons are threefold:

• Text‑first architecture: Large language models excel at processing plain text, eliminating the need for costly image‑to‑text conversion pipelines.
• Reduced latency: Terminal commands can be executed in milliseconds, whereas screenshot‑based pipelines often require multiple seconds of image encoding and decoding.
• Scalability: Text‑only interactions consume far less GPU memory, allowing providers to serve millions of concurrent users.

Industry leaders like Nvidia’s Jensen Huang have called this shift “the new operating system for agentic computers,” urging every company to develop an AI news strategy around terminal agents.

Analysts weigh in on Google’s move

Market analysts see Google’s restructuring as a pragmatic response to the “adoption gap” that plagued earlier browser agents. Kian Katanforoosh, CEO of Workera, noted that “the massive computational overhead of screenshot pipelines makes them unsuitable for mass‑market products.”

Competitors are also adjusting. Perplexity’s recent launch of a “Personal Computer” agent attempts to blend terminal efficiency with a lightweight GUI, while OpenAI is doubling down on Codex‑powered coding assistants for enterprise workflows.

What this means for developers and enterprises

For developers building on Google’s AI stack, the restructuring offers several concrete benefits:

• Unified API surface: The Gemini Agent will expose a single endpoint for both web‑automation and terminal commands, simplifying integration.
• Better tooling: Access to the Web app editor on UBOS means rapid prototyping of agent‑driven workflows without deep knowledge of Chrome internals.
• Enterprise‑grade security: By consolidating agents under the UBOS partner program, Google can enforce stricter data‑privacy controls for SMBs and large enterprises alike.

Enterprises that rely on repetitive web tasks—such as invoice processing, data entry, or market‑research scraping—can now look forward to more reliable, cost‑effective solutions that leverage text‑based automation.

Key takeaway from the Wired report

“Google’s decision to fold Mariner’s capabilities into Gemini reflects a broader industry realization that terminal‑centric agents are 10‑to‑100× more efficient than their screenshot‑driven counterparts, and that the future of AI assistants will likely be a hybrid of web‑automation and code‑generation.”

Looking ahead: The next chapter for AI agents

While the browser‑automation hype may be cooling, the underlying technology is far from obsolete. Startups like Standard Intelligence are pioneering video‑based computer‑use models that could bridge the gap between visual and text agents, promising up to 50× efficiency gains.

Google’s integration of Mariner’s tech into Gemini suggests a future where agents can seamlessly switch between a web UI and a terminal, choosing the most efficient interface for each task. This hybrid approach could finally deliver the “general‑purpose assistant” that Sundar Pichai envisioned.

How UBOS can help you stay ahead

If you’re a product manager or AI researcher looking to experiment with next‑gen agents, UBOS offers a suite of tools that align perfectly with Google’s new direction:

• Explore the UBOS templates for quick start, including a pre‑built “AI Agent” template that combines web‑scraping and terminal commands.
• Leverage the AI Video Generator to create demo videos of your agent workflows.
• Use the AI SEO Analyzer to ensure your agent‑powered web content ranks well.
• Join the UBOS solutions for SMBs to get affordable, enterprise‑grade AI automation.

Ready to build the next generation of AI assistants? Visit the UBOS homepage and start prototyping today.

Read the original story on Wired.

Andrii Bidochko

CTO UBOS

Andrii Bidochko is an AI entrepreneur and researcher focused on AI agents, reinforcement learning, and autonomous systems. He writes about the technologies shaping the future of machine intelligence, from frontier models and agent architectures to real-world AI applications.