AI that can plan and do things for you

Alongside Gemini 3, Google is introducing a new tool called Gemini Agent. First available on desktop for Google AI Ultra subscribers in the US, the new Gemini Agent acts more like a true personal assistant than a simple chatbot.

What can Gemini Agent do?

Instead of simply answering your questions, Gemini Agent can plan, explore, and execute complex tasks on your behalf. However, before it does anything, you’ll need to grant it the necessary permissions, such as access to your browser, Gmail, or Calendar. It’ll also seek permission for things like sending a message or making a purchase. Before you ask, yes, you can delete the information you share with it at any time.

As far as task completion goes, Gemini Agent boasts a broad set of capabilities. It can organize your schedules, manage emails and dig out information from them, create slide decks, prepare personalized summaries for your day, and more. The best part is that you can give multiple instructions in a single prompt using natural language. You can also see when and how the agent is working on your request as well as ask it follow-up questions once it’s done.

Depending on the task, Gemini Agent may even open a self-contained web browser to retrieve necessary data. If it determines this is needed, Gemini will ask you to “allow content sharing” before proceeding. Then, it will open a new pane and load a fresh instance of Chrome in it. It will then automatically scroll and click on links to complete the task it was given. At any time, you can hover over the browser panel and click “Take control” to manually navigate.

If you ask Gemini Agent to perform something it can do with the use of connected Google Apps, then it may not open a browser window at all. For instance, you can tell Gemini Agent something like – “Look for the latest unread message from Hadlee in my inbox. Check my Drive for any files connected to the project he mentioned. Draft a short reply letting him know I’ll review everything and follow up tomorrow,” and it will perform these tasks by directly connecting to your Gmail and Drive accounts.

Some more examples of the kinds of tasks that Gemini Agent can handle for you include:

• Email help: Find recent emails, gather relevant documents, draft responses, or scan unread messages for tasks and add them to Google Tasks.
• Daily briefings: Create morning digests that include important meetings, key emails, local news, or sports updates.
• Booking and ordering: With live browsing, the Gemini Agent can make dinner reservations or purchase tickets online.
• Calendar management: Search and create calendar events, prepare meeting notes, and more.
• Presentation building: Research a topic and automatically turn the findings into a slide deck.

In essence, Gemini Agent works much like modern agentic browsers, such as Perplexity’s Comet or ChatGPT’s Atlas, which not only display websites but also take actions on them. However, unlike Comet or Atlas, Gemini Agent isn’t a fully integrated part of any web browser, which means it doesn’t have access to all your saved logins and browser sessions. This makes it less versatile but also less risky to use, as you won’t be able to automate tasks on your most sensitive online services. Still, Gemini Agent is not immune from prompt injection attacks, so use it with caution.