Google Kills Project Mariner — And That’s Not Bad News
Google's experimental AI browser agent, Project Mariner, is no more. Google has shut down Project Mariner — its AI agent that could perform web tasks such as booking hotels or organising email — saying the technology has moved into other Google products, including Gemini Agent and AI Mode.
On the surface, this looks like a retreat. It is not.
What Project Mariner Was
Project Mariner was Google's attempt to build an AI that could operate a web browser on your behalf — clicking links, filling forms, booking tickets, reading emails. An AI that didn't just answer questions, but took action.
It was experimental. It was imperfect. And it raised more questions than it answered — about what it means to hand an AI agent access to your inbox, your accounts, your digital life.
everything on the web starts with the domain
Why Google Is Consolidating, Not Retreating
Google is growing its cloud business fast by helping companies build and manage thousands of AI agents. Project Mariner was one experiment among many. Absorbing it into Gemini Agent and AI Mode is not a failure — it is a product decision. Google is moving away from standalone AI experiments and toward embedding agentic features inside products that billions of people already use every day.
That is a more dangerous move than it sounds. A standalone experiment is easy to ignore. An AI agent built into Gmail, Chrome and Search is not.
The Real Question: How Much Autonomy Are Users Ready to Give?
AI has been integrated into Google Search, Gmail, Calendar, Maps, Docs, and Photos — meaning that people who aren't even seeking out AI are now engaging with it. No other company delivers AI to so many people in so many places.
Project Mariner's death is less significant than what replaces it. When Google embeds agentic AI into products used by three billion people, the question of trust, permissions and control becomes urgent — and largely invisible to most users.
What This Means for GAFAM
Google is not the only company building AI agents. Microsoft's Copilot, Meta's AI assistant and Amazon's Alexa are all moving in the same direction. The race is no longer about which company has the best AI model. It is about which company's AI agent becomes the default layer between humans and the internet.
Google, with its unmatched distribution across Search, Gmail and Chrome, is the most dangerous player in that race.
The European Perspective
The EU AI Act's provisions on transparency and human oversight apply directly to agentic AI systems. An AI that books your hotel, reads your email and fills your forms on your behalf is precisely the kind of high-risk interaction European regulators have been watching. As Google embeds agents deeper into its products, Brussels will be paying close attention. gafam.ai will be watching.