Google's Gemini Agent can orchestrate complex tasks on your behalf in the Gemini app

What you need to know

• Google is updating the Gemini app with a new interface and Gemini 3 Pro support.
• Google AI Ultra subscribers can try the Gemini Agent, a tool that can complete actions within the Google ecosystem and on the web on your behalf.
• The Gemini app is also gaining two generative interfaces, which are custom views coded in real time by Gemini 3 based on your request.

Gemini 3 is here, and Google's latest AI model brings better multimodal understanding and enhanced agentic coding abilities. It's available in the Gemini app starting today, powering a new app interface that is AI-generated in real-time to create custom UI elements. The app also adds the Gemini Agent, an experimental tool that can handle multi-step tasks for you without leaving the Gemini app.

It's the evolution of Project Mariner, a research prototype Chrome extension that tested using an AI agent to browse on a user's behalf. Everything Google learned from Project Mariner, plus Gemini 3 smarts, was used to create the Gemini Agent. The feature taps into Google Workspace apps and tools like Deep Research and Canvas to take care of complex requests. Paired with live web browsing support, the Gemini Agent is Google's "next step toward a true generalist agent."

Google provides a few examples of how you might use the Gemini Agent, like to manage Calendar events, create reminders, or organize your Gmail inbox. Although the Gemini Agent can complete actions on your behalf, it'll ask you for confirmation before making important changes, like completing a purchase or deleting an email.

Gemini 3 isn't just an agentic powerhouse, it's also great at vibe coding — the process of describing an interface or application to an AI agent that then codes it for the user. The same vibe coding prowess is now being used to create generative interfaces automatically for the Gemini app. The app now has a My Stuff folder, a redesigned homepage, and a new shopping experience.

The Gemini app has a fresher standard appearance, but it's designed to adapt based on the situation using what Google calls generative interfaces. They're generated by the Gemini 3 model in real time based on your prompt, tailored to your request. The first to kinds of generative interfaces are called visual layouts and dynamic views.

Visual layouts create an "immersive, magazine-style view complete with photos and modules." Meanwhile, dynamic views use Gemini 3 to build an interactive UI in real time that can be used to learn and explore content surfaced by Gemini. Both of these generative interfaces can be tweaked with further Gemini prompts, and they're available starting today. Some users might only see one of these layouts as Google compares the experiments and solicits user feedback.

Users can try Gemini 3 Pro in the Gemini app by selecting the "Thinking" model, and it's available for free users and paid subscribers. Google AI Plus, Pro, and Ultra subscribers have higher rate limits, and the Gemini Agent is only available for Ultra users in the U.S. in the web app for now.