The year is 2025, and Google is now rolling out its own agent AI feature in the Gemini app. While the company has discussed agent AI prototypes before , it now appears ready to take them mainstream. At the Google I/O 2025 keynote, Google discussed how a new feature could go online on its own and perform tasks for you. Like OpenAI’s Operator , it could take a hint, create a checklist of things to do, and then perform them for you.

According to Google, Agent Mode combines features like real-time web browsing and deep research with the integration of data from Google apps to perform its online tasks. The model is supposedly capable of performing multi-step actions, from start to finish, with minimal user oversight.

We don’t know much about how exactly this feature will work yet, but Google showed us an example on stage. Here, Sundar Pichai asked Gemini to find a new apartment to rent on a budget and with access to an in-room laundry. Gemini then generated a list of tasks to do, like opening a browser, going to Zillow, finding matching listings, and even booking a tour. This is all possible because Google uses MCP in the background. Model Context Protocol (introduced by Anthropic) is a new industry-wide protocol that web developers and apps can use to directly integrate with AI tools. In this example, Google can search Zillow and book a tour using the protocol, which is much more reliable than launching a web browser and asking AI to click a few buttons for you.

Agentic’s capabilities aren’t limited to just the Gemini app’s agent mode. Google also introduces a more modest version of these capabilities in Chrome and Google Search. For example, Agentic’s AI-mode features can help you search for game tickets in the background .

According to Google , Agent mode will soon be available in the U.S. as an early preview of Google’s new AI Ultra plan, which costs $250 per month. There’s no word yet on broader availability.

More…

The year is 2025, and Google is now rolling out its own agent AI feature in the Gemini app. While the company has discussed agent AI prototypes before , it now appears ready to take them mainstream. At the Google I/O 2025 keynote, Google discussed how a new feature could go online on its own and perform tasks for you. Like OpenAI’s Operator , it could take a hint, create a checklist of things to do, and then perform them for you.

According to Google, Agent Mode combines features like real-time web browsing and deep research with the integration of data from Google apps to perform its online tasks. The model is supposedly capable of performing multi-step actions, from start to finish, with minimal user oversight.

We don’t know much about how exactly this feature will work yet, but Google showed us an example on stage. Here, Sundar Pichai asked Gemini to find a new apartment to rent on a budget and with access to an in-room laundry. Gemini then generated a list of tasks to do, like opening a browser, going to Zillow, finding matching listings, and even booking a tour. This is all possible because Google uses MCP in the background. Model Context Protocol (introduced by Anthropic) is a new industry-wide protocol that web developers and apps can use to directly integrate with AI tools. In this example, Google can search Zillow and book a tour using the protocol, which is much more reliable than launching a web browser and asking AI to click a few buttons for you.

Agentic’s capabilities aren’t limited to just the Gemini app’s agent mode. Google also introduces a more modest version of these capabilities in Chrome and Google Search. For example, Agentic’s AI-mode features can help you search for game tickets in the background .

According to Google , Agent mode will soon be available in the U.S. as an early preview of Google’s new AI Ultra plan, which costs $250 per month. There’s no word yet on broader availability.

More…