Google Adds Self-Acting AI to Search Across 180+ Countries. Here's What Changes.

Google has rolled out AI Mode—its AI-powered search feature—to more than 180 countries and territories, and it's now adding a new capability: the ability to carry out tasks on your behalf. If you pay for Google AI Ultra (the company's premium tier), you can now tell the AI to book a restaurant reservation, and it will navigate the restaurant's website, find availability, and complete the booking without you manually clicking through each step.

This is a significant shift in how search works. Traditional search has always been passive: you ask a question, get links or answers, and then you do the next action yourself. Google's new approach is active. The AI system, powered by technology called Project Mariner, can browse the web and perform multi-step actions autonomously.

How Google is Rolling This Out Worldwide

Google has been methodical about expanding AI Mode. It started in the United States, India, and the United Kingdom, and now it's spreading to 180-plus additional countries. For now, all of these versions only work in English while Google develops versions for other languages.

The staggered rollout makes sense. When an AI system actively interacts with other companies' websites, it has to follow different laws and rules in different parts of the world. Google needs to work through data protection requirements and make sure it has the right arrangements with those service providers in each region.

The Technology Behind the Scenes

Project Mariner is Google's tool for making AI browse the web the way a person would—reading the page, clicking buttons, filling out forms. Rather than relying on pre-arranged partnerships or direct connections (called APIs), Google's approach works with almost any website on the internet. The downside is that it uses more computing power and is more fragile; if a website changes its layout or uses security tools like captchas, the AI might get confused.

Who Gets Access, and Why It's Limited

For now, only people who subscribe to Google AI Ultra can access these agentic features (agentic is the technical term for systems that act on your behalf). It's also still marked as experimental, available through a feature called Google Labs. This keeps the feature limited to early testers and helps Google manage costs—making an AI book a restaurant reservation requires far more computing power than answering a question.

The subscription model also acts as a natural speed bump. Each agentic action involves multiple requests to websites and complex decision-making that plain search doesn't need. Charging for it helps Google manage how much it uses its servers.

Starting with Restaurant Reservations—But Why?

Google's first real test case is restaurant reservations, and there is logic in that choice. Booking a restaurant is common enough that many people will try it, but it doesn't involve large amounts of money, so users are less anxious about letting AI handle it. Getting this right builds confidence before moving to bigger tasks like airline bookings or medical appointments.

Looking back at how automation has evolved, this pattern repeats. Voice assistants started with simple queries—"What's the weather?"—before they could control your smart lights. Mobile payment apps began with small purchases before people trusted them with larger ones. The principle is the same: prove yourself on low-risk tasks first.

What This Means in a Competitive Field

Google is not the only company working on this. OpenAI's ChatGPT can browse the web, Anthropic's Claude can interact with computer interfaces, and Microsoft is doing similar things with Copilot. The real competition is not over whether to build these features, but over who gets them to work reliably.

Google has some advantages here. It already knows your search history, your location, your restaurant preferences. When booking a reservation, the AI can use that context to make better decisions than a system starting from scratch.

The Real Challenges

Active AI on the web introduces problems that simple search never had to solve. What happens when a restaurant is fully booked? What if a payment fails partway through? The system has to fail gracefully—backing out of the action and explaining to you what went wrong—while leaving a clear record of what it tried to do.

There is also the question of privacy. When AI acts on your behalf on other websites, it leaves traces: logs of your actions, data shared with those services, patterns that those companies now know about you. Those trails are out of Google's direct control.

The experimental label on these features gives Google and its users room to work through these issues before rolling it out more widely.

What Comes Next

The expansion to 180-plus countries gives Google a vast testing ground. It will see how restaurant booking works in Tokyo, Bangkok, London, and São Paulo. That real-world data will show where the AI gets stuck and how to improve it.

Over time, Google will have to decide whether to keep these features exclusive to paid subscribers, expand them more widely, or create different tiers. The restaurant reservation case may be narrow, but it is building the foundation for more complex tasks: booking flights, scheduling doctor appointments, managing multiple steps across several websites.

The question for users is simple: as AI becomes more capable at acting on your behalf, how much of that do you want to let it do? Google is betting that the answer is "more, as long as it works."