A day before Computex 2026 kicks off in full, Qualcomm President and CEO Cristiano Amon spoke at length about agentic AI. He says that the company calls this year “the year of the agents”. And the keynote is essentially about the evolution and proliferation of agentic AI, from simply answering prompts to something that can take action on their own.
This will not only be for professional settings either, though that’s probably where it will be most useful. But at home, these AI agents can update your family about things that happened in your day, and what’s up on your schedule. The agent can also feed you information and updates that are specifically relevant to you, and work on things in your schedule to prioritise. And the agent can do all of this proactively, across devices. These will be the new centre of someone’s digital life, the way phones currently are.
On one hand, it sounds very aspirational. Because it’s clear that agentic AI tech is not quite there yet. At least, not for the average consumer market. This is especially when Amon says that these agents will treat all of your tech as end points, letting it move from product to product, from one ecosystem to another. Amon himself says that devices of today were not designed for those experiences. More specifically, devices of today are made for actions initiated by the user, and the market needs devices that allow agents to act proactively and independently.
With all that in mind, for devices to be able to run AI agents that way, it requires not only for devices to have much higher compute capabilities for CPUs, and very efficient NPUs and GPUs since they’ll basically be running all the time, all day. And these agents will also need context, and a lot of these will come in visual form, so bandwidth will also need to be very high, while also maintaining very low bandwidth. This applies to not only smart glasses with cameras on them, but also cars, the latter letting you ask about interesting things you may see while on the road. Assuming the car’s cameras also see it, of course.
On the flip side, some larger scale applications do look a tad scary. On the lower end of the creepy scale is when vision intelligence is used to look at and recognise your number plate, or monitor traffic situations. Then on the other end there’s businesses using it to go through their security camera feed and tracking specific customers, including their precise movement and field of view while they’re in a store. It’s even possible to look for specific people by just describing them using natural language.
Needed to power all this is, as you may expect, 6G. Because not all of these agentic AI processes happen on-device, or what is referred to as edge computing. A lot of the functionality still happens in the cloud. This is probably especially the case to allow said agentic AI to jump from device to device. Then there’s the kicker – the network itself is part of a large data centre, tracking RF signals in real time, creating a digital twin of everything.
The keynote ends with the reveal of a server rack and the name Qualcomm Dragonfly. Naturally, this is something for data centres. But Amon did not elaborate further, as that is saved for another date – 24 June.
