WayToClawEarn
High impactThe Verge / Wired

Google shuts down Project Mariner: Browser AI Agent experiment ends

Google has shut down its experimental browser AI Agent Project Mariner. This feature will be first released in December 2024 and supports operations such as automatically booking hotels and organizing mailboxes. Its core capabilities have been integrated into Gemini Agent and AI Mode. The incident occurred on May 4, 2026, heralding the transition of AI agent tools from experiments to product integration.

WayToClawEarn EditorialPublished May 7, 2026Updated Aug 8, 2026

Editorial review of public sources · AI-assisted drafting. How we work · Original source

Core conclusion

On May 4, 2026, Google officially closed the experimental browser AI Agent project Project Mariner. This product was once regarded as Google's key product to benchmark OpenAI Operator and Perplexity in the field of AI agent tools. Its core technical capabilities have been migrated to formal product lines such as Gemini Agent and AI Mode.

Key Points

  • Closing: May 4, 2026
  • Scope of Impact: All users using Project Mariner must switch to Gemini Agent or AI Mode
  • Core changes: Google integrates independent experimental products into a unified AI platform strategy, marking the move of AI Agent from "experimental functions" to "infrastructure-level capabilities"
  • Meanings for automation practitioners: AI Agent’s technology stack is accelerating its convergence to a few main platforms

Event Review: From 2024 to 2026

Project Mariner first became available as an experimental feature in Google Chrome in December 2024. It is based on the Gemini model and is able to simulate human actions in the browser - clicking buttons, filling out forms, managing inboxes, etc. In 2025, Google launched an upgraded version that supports 10 tasks simultaneously.

Over the past year, Project Mariner’s core capabilities have been gradually absorbed into Google’s other AI products:

  • Gemini Agent: can archive emails and book hotels on your behalf
  • AI Mode: AI-powered features in Google Search
  • Auto-Browse: A Chrome feature launching in early 2026 that supports multi-step tasks like automatically researching flight prices

The farewell message on the Landing Page reads: "Thank you for using Project Mariner, which was shut down on May 4, 2026 and its technology has been shipped to other Google products."

Google Project Mariner timeline

Key impact table

DimensionsChangesImpact on automation practitionersRecommended actions
Product positioningFrom independent experimental project → built-in platform capabilitiesBrowser AI Agent is no longer an independent trackEvaluate the compatibility of Gemini Agent with existing workflows
Agent capabilitiesMulti-step tasks (10 simultaneous) → integrated into GeminiSimilar capabilities are still there, but the entrance has changedLearn Gemini Agent API and limitations
Third-party comparisonOpenAI Operator and Perplexity are still developing independentlyGoogle has chosen integration rather than independent productsFocus on independent Agent product iterations on other platforms
Technology evolutionMariner→Auto-Browse migration completedThe underlying architecture is more mature and stablePay attention to the AI release at Google I/O 2026 (May 19)
User impactNo need to enable the experience separatelyFunctions are deeply integrated into existing toolsNo separate learning required, reducing tool switching costs

Why this matters for AI automation

The closure of Project Mariner is not a negative sign, but a sign of a mature industry. When the browser AI Agent capability changes from "experimental function" to "default capability", it means:

  1. Technology stack convergence: Instead of maintaining independent projects, it is better to embed Agent capabilities into core products such as search, browsers and assistants
  2. Scale Delivery: Users no longer need to actively "start experiments" - the capabilities are naturally there
  3. Competitive product benchmarking is clear: Google’s Auto-Browse directly benchmarks against OpenAI Operator, Perplexity’s browsing agent, and workflow tools such as OpenClaw

For content automation practitioners, this means that when building workflows in the future, the browser agent layer will be a default capability that can be relied on, rather than an experimental feature that requires additional configuration.

Next up: Prelude to Google I/O 2026

The shutdown of Project Mariner comes exactly two weeks before Google I/O 2026 (May 19). This is a strong hint that Google will showcase its integrated AI Agent strategy at I/O. What you may see include:

  • In-depth upgrade of Gemini Agent
  • Auto-Browse API is open
  • Deeper integration with Android / ChromeOS

Google I/O 2026 AI preview

Tool entry

The following tool brands that appear naturally in this article will automatically match the maintained tools library on the platform side: Google, Gemini, OpenAI, ChatGPT, Perplexity, OpenClaw, Claude, Chrome

Related extended information

Internal link guidance

View source →

Disclaimer: this site shares educational insights only, for inspiration and reference. No outcome guarantee; external execution and decisions are your own responsibility.