Agentic AI is the buzzword of the moment among generative artificial intelligence developers. An "AI agent" is a system capable not only of answering questions, but also of acting autonomously to accomplish tasks. Specifically, it can make decisions, perform several actions, and interact with applications or services—such as sending a message, booking a ticket, or changing a setting—without the user having to do everything manually.
An AI agent in your pocket
Google is obviously on the case with the Astra project, which we got a taste of last year with prototype glasses (nothing to do with Android XR glasses, but who knows!). Same old story for the 2025 edition of Google I/O, without the glasses but with a smartphone equipped with an AI agent that takes control of the device at the user's request.
The user in question has both hands busy repairing a bike. All interaction with Gemini is therefore voice-based, and the assistant effectively takes the lead: it can go to a page of a PDF manual or find videos by scrolling alone on YouTube, as if the invisible man were touching the screen.
With its multimodal capabilities, the assistant is also able to locate the right nut on a workbench. It also knows how to make a phone call to a DIY store, ask if a part is available, and summarize the answer for the user.
All this is very impressive... and still not available. Astra's demonstration last year was even more impressive because it used glasses; here, it's more "ordinary" if you will, since the demo is done on a smartphone. But Google's idea is above all to show what an AI agent can be used for in practice and with common hardware.
0 Comments