Qualcomm's post-app blueprint: Don McGuire on Agentic AI, 6G and India's next frontier

Qualcomm's latest Snapdragon Summit was about more than new smartphone processors. Alongside the Snapdragon 8 Elite Extreme Gen 6 and Snapdragon 8 Elite Gen 6 platforms, the company is increasingly focused on how AI will change the way consumers interact with their devices and the services behind them.

Qualcomm’s Executive Vice President and Chief Marketing Officer Don McGuire outlined the company's approach to what he calls the "agentic AI age". That includes moving beyond standalone apps, running more AI workloads directly on devices, making AI agents capable of carrying out tasks on users' behalf, and building enough trust for those agents to eventually handle transactions.

For Qualcomm, the opportunity extends beyond smartphones. The company sees Snapdragon silicon powering an increasingly connected range of devices, from laptops and cars to smart glasses, watches and earbuds.

Must read: Snapdragon 8 Elite Gen 6 series is here: Qualcomm’s new chips target flagship AI smartphones

The shift from generative to agentic AI

"As we transition from the generative age to the agentic age, where AI is not just about answering questions or queries, how do we build trust? How do we build understanding? How do we build advocacy? And then how do we make it useful? Because people are only going to use it if it's useful," McGuire said.

Qualcomm's positioning relies partly on what McGuire calls the "ecosystem of you". Rather than limiting AI to an individual smartphone, the company wants intelligence to work across a portfolio of connected devices.

"We have a real opportunity to do that because of where we live across what we call the 'ecosystem of you', which is your device portfolio, whether it's your watch, your earbuds, your glasses, your phone, or even your car," McGuire said. "We've got such a footprint there that we can actually tell a really nice story."

The bigger challenge, however, is getting consumers to change established behaviour. McGuire said AI agents will only become mainstream if they make everyday tasks easier than existing methods.

Must read: Meet Abhay Karandikar: From IIT professor to NITI Aayog member & the researcher behind India’s 5G and 6G journey

"If it's easier to do with an agent than going to your phone, opening up an app or multiple apps, people will adopt. If it's harder, or more difficult, then they won't. So it has to be less friction, faster, and more efficient than doing it the current way, or else no one's going to do it.

Securing the commercial "last mile"

One of the more complicated parts of the agentic AI proposition is commerce. AI demonstrations can show an agent searching for a flight or hotel and making a booking, but allowing an agent to actually complete a transaction introduces questions around authentication, payments and trust.

"I understand why they're blocking them because they don't have the last mile secured yet," McGuire said, referring to the challenges around automated transactions.

He pointed to demonstrations involving Mastercard executives as an example of how the industry could eventually address the problem. "If you trust the brand, if you trust the company who's telling you it's okay, you can go ahead and move forward, then I think it'll make it easier for people to unlock that last mile of commerce, which is: 'I'm going to allow my agent or this agent to have my secure payment details so that I can make an easy purchase through my agent.'"

McGuire compared the transition to earlier changes in consumer payments. "People got used to not having to go into the bank and going to an ATM. People got used to loading their cards into Google Wallet or Apple Pay and just paying with their phone."

India could be an important market in that transition because of the widespread use of digital payments and UPI. "So if you think about India, for example, who is the most trusted financial brand in India? ... Then maybe you start there with coming up with a solution from a brand that people are like, 'Okay, I trust them with my money.'"

McGuire also stressed that autonomous systems will still need user confirmation and authentication, particularly when an action could have significant consequences. "You need to have confirmation and authentication. I think those are two really important things in the process. So it shouldn't just assume," he said.

Must read: Apple’s iPhone exports from India hit a new high: $13.2 billion iPhones in 5 months

He pointed to a scenario where a user confirms an action through another trusted device. "Imagine if you said yes to your agent to do something, and then they said, 'Well, we sent a confirmation code to one of your devices, your trusted devices; you need to confirm through that device.' In the scenario that we gave, the woman had to confirm on her watch because that was a trusted environment. I think that's really important in the process."

Why on-device AI matters

Latency is another obstacle to agentic AI. If an AI agent takes longer to complete a task than a consumer would by opening an app and doing it manually, there is little reason to use the agent.

"If it's faster for you just to go to Uber and book your Uber than it is for your agent to do it, you're not going to do it through your agent. So you've got to get that latency down," McGuire said.

He argued that running appropriately sized AI models directly on devices could reduce the need to send every request to the cloud.

"Right now, it's assuming it has to go to the cloud and come back to make that decision for you, but it really doesn't if they've enabled on-device AI. If the model is small enough, up to now 20, 30, 40 billion parameters, it should be able to do it on your device for you instead of having to go to the cloud, which will cut down on latency."

The shift could also change the role of traditional apps and operating systems.

McGuire suggested that dedicated app stores could eventually become more like directories of agents that communicate with one another. Smart glasses, meanwhile, could provide a different interface altogether. "Having an agentic experience on glasses, where there is no legacy OS, is going to be a lot different because it's going to be primarily voice-based, and seeing what you see and then interpreting that," he said. "So I don't think you have to unlearn anything. I think you just have to learn how to speak."

6G and the rise of AI-native connectivity

Qualcomm is also looking further ahead at how connectivity will need to evolve as AI becomes more embedded in everyday devices.

McGuire argued that future AI experiences, particularly those involving smart glasses and continuous visual or audio data, will require significantly greater upload capacity.

"The one thing that you need to enable all these amazing experiences is, especially if you're having an ongoing conversation, is massive upload, which 5G can't handle. So that's where 6G is really going to help," he said.

"If you're walking around in your glasses and you're livestreaming... that's massive, constant upload that 5G can't handle. That's what 6G will be able to do, transitioning from a mobile connectivity network to an AI-native network."

For India, McGuire believes the country's progress on 5G infrastructure puts it in a strong position for the next generation of connectivity. "India has caught up on the 5G infrastructure with government commitment. Countries have to stop wanting to do their own thing when it comes to connectivity and standardise. If that holds true for 6G, then India will be just as ready as Japan or anywhere else."

Why open-weight models matter

McGuire also discussed the changing economics of AI models and the differences between the US and Chinese AI ecosystems.

He does not believe Chinese open-weight models have overtaken the leading US frontier models in raw capability. Their importance, he argued, lies elsewhere: efficiency and the ability to deploy models for specific workloads without relying exclusively on the most expensive frontier systems.

"I don't think they're ahead of the frontier models," McGuire said regarding Chinese open-weight models. "But open-weight models are driving efficiencies outside of frontier models, which is a good thing. If you had to rely on frontier models for everything, the costs would be prohibitive."

"I don't think they're ahead of the frontier models," McGuire said regarding Chinese open-weight models. "But I think the open-weight models are driving efficiencies, and they're driving more model efficiency outside of frontier models, which is a good thing. Because if you had to go to the frontier model for everything, the costs would be so high."

Qualcomm is also looking to reduce its reliance on external AI infrastructure. "We're bringing everything on-prem too. So we're putting our own data centre infrastructure on site so that we can process it all locally so that we don't have to pay Anthropic millions of dollars in data centre bills," McGuire said.

He sees China as particularly advanced in applying open-weight models to consumer products and agentic experiences.

"On the open-weight models and then getting that to the agentic workloads and the experiences, China is way further ahead. Have you seen what they've done in the car? Every car was agentic... Just walk in, some little dude pops up with his eyes blinking, 'Where do you want to go? What do you want to do?' ... They're so far ahead."

Making AI useful rather than explaining it

For AI to become mainstream, McGuire believes technology companies need to demonstrate its value through products rather than simply explaining the underlying technology.

"A lot of times, it's like proving it to them without them knowing that it's actually happening," he said.

"AI's been working in the background for years, making your photos better, making your music sound better, or whatever. People just didn't know it was AI. They just thought, 'Oh, what a really good camera!'"

The same principle, he argued, applies to trust.

"I think it's show versus tell. You have to show them, prove the trust. You can tell people, but why should they trust you? Again, maybe you have a reputation, like a Mastercard, where you're trusted. But all these AI companies are fairly new, so they have no trust, and they have not helped themselves... because they have CEOs who have literally no media training and don't come across as empathetic."