Data & Dimensions | Ep. 20 (Video) - Mobile Moves From Apps to Agents

View Full Transcript

Neil Barbour, Analyst, S&P Global Market Intelligence   00:00:00

Qualcomm brought nearly 400 partners, creators, analysts and media to a resort in Maui in late September to make the argument that your phone isn’t going away, but it’s about to start delegating. Two new flagship chips, three days of demos and a bet that the winning strategy is owning the silicon across a web of agentic inputs: the glasses, the watch, the earbuds, the laptop or the car.

Welcome to Data & Dimensions, a podcast at the intersection of industry, entertainment and interactivity. I’m Neil Barbour, an analyst with S&P Global Market Intelligence Kagan, and Qualcomm flew me out to Hawaii for three days at Snapdragon Summit 2026. This is what Qualcomm’s actual hardware and its partners have to say about where personal AI is really headed. CEO Cristiano Amon said in his opening keynote, “We are just now at the beginning of the agentic age, and Snapdragon is the platform built for it.”

Cristiano Amon, President and CEO, Qualcomm   00:00:57

What makes this very interesting is whether the agent runs on the device or is built into the OS, it now needs to do a lot of coordination across apps, services and data. But your experience, to be useful, needs to remain very intuitive, instant, secure and available. That is what I’m going to tell you: how we’re going to see the difference between what you are going to do yourself and what you’re going to rely on the agent to do.

Neil Barbour, Analyst, S&P Global Market Intelligence   00:01:29

He was careful, though, to head off the narrative that has hovered over the industry: that smart glasses or some other AI gizmo might eventually sideline the phone. “Your phone isn’t going anywhere,” he said flatly. His argument is that the phone’s ubiquity and personalization make it uniquely suited to anchor an agent that needs context and connection: something that is with you all day, understands your life and can reach every other device and cloud service on your behalf. Qualcomm Senior Vice President and General Manager of Handsets Chris Patrick put it this way:

Chris Patrick, SVP and General Manager of Handsets, Qualcomm   00:02:05

We believe the future of personal computing will be defined by what Cristiano mentioned: the ecosystem of you, a collection of intelligent devices, each contributing a unique layer of understanding about you and about your world. The PC will understand how you work, how you create, how you communicate. XR devices will see how you engage with the world around you. Audio devices will know what you listen to and will understand the conversations that matter most to you. Wearables will track your health, your routines, your daily activity. Together, these devices will create an increasingly rich and personal understanding of your life. And at the center of it all will be mobile, where these experiences will converge, where context will come together, where intelligence becomes most personal. The smartphone will be the launchpad for the agentic age.

Neil Barbour, Analyst, S&P Global Market Intelligence   00:03:03

When you combine that foundation with intelligent agents and innovation, entirely new experiences become possible. We’re already beginning to see what this looks like in the real world. The silicon arrived in two flavors for the first time: the Snapdragon 8 Elite Gen 6, built on a 2-nanometer process, and a higher-clocked 8 Elite Extreme Gen 6 that Qualcomm says is the first mobile CPU to clear 5 gigahertz. The CPU manages control flow, key decisions, task scheduling and data movement, all of which are required to make the experience responsive.

Chris Patrick, SVP and General Manager of Handsets, Qualcomm   00:03:35

The CPU helps your agents execute. It’s a technology that is responsible for turning intent into action. The faster the CPU can execute the decisions, the more immediate the response will feel. That’s why we built a brand-new generation of Qualcomm Oryon CPUs featuring our most advanced core architecture yet, with two prime cores and six performance cores. I’m proud to share this is the first mobile CPU ever to break the 5 gigahertz barrier.

Neil Barbour, Analyst, S&P Global Market Intelligence   00:04:08

Executives were candid in the Q&A that the two-tier split is a direct response to a fragmenting flagship market. Foldables, gaming phones and other mainstream premium devices all want different things from their silicon. The Extreme chip’s headline change is a new offset packaging design that moves memory beside the processor instead of stacking it on top, pulling heat away from the chip and increasing sustained performance under heavy load, although Qualcomm admits that it asks more of a partner’s thermal engineering to actually use it. On the AI side, the new Hexagon NPU adds a dedicated element-wise accelerator built for the transformer math behind agentic workloads, a 32,000-token context window and more on-chip NPU memory.

Qualcomm says that adds up to a hefty speedup on some models. As agentic workloads become more sophisticated, AI tasks won’t just run in one place. They will move across models, across accelerators and even between the device and the cloud. But still, at the heart of the system is the Hexagon NPU. The NPU helps agents think. It runs the large AI models that perform inference, generate context and make sophisticated decisions, bringing richer intelligence directly onto the device. To power this next generation of AI, Qualcomm reengineered the NPU architecture. At its core is the new Hexagon element-wise accelerator, designed to speed up token processing, improve model performance and support context windows up to 32,000 tokens. This will allow agents to process more information and respond faster and more intelligently.

Neil Barbour, Analyst, S&P Global Market Intelligence   00:05:52

Xiaomi had the honor of going first onstage. The company confirmed its 18 Pro and 18 Pro Max, running the standard and Extreme chipsets, respectively. They would launch in Beijing that same evening, with a global rollout later this year.

Adam Zeng, Senior Vice President, Xiaomi Corporation, and President, International Business Department, Xiaomi   00:06:25

And for our international friends, Xiaomi 18 Pro series will come to global users very soon this year.

Neil Barbour, Analyst, S&P Global Market Intelligence   00:06:43

Xiaomi claimed a 38% reduction in per-frame power consumption during gaming on the new silicon, one of several generation-over-generation efficiency figures partners were eager to attach their names to this week. If day one was about proving the phone can still do more, day two was about proving that it doesn’t have to. Ziad Asghar, Qualcomm’s senior vice president and general manager of XR, wearables and personal AI, laid out a wearable stack: always-on, low-power sensing; the same imaging silicon scaled down; tiered NPUs; and ultra-low-power Wi-Fi for tiny form factors, explicitly framed as the interface layer for the agentic era.

Ziad Asghar, SVP and General Manager of XR, Wearables and Personal AI, Qualcomm   00:07:40

And now let’s see how we can put these technologies into practice. Let’s imagine someone who is visually impaired walking through their day with only their glasses on.

Demo audio   00:07:50

What color is this apple? This is a red apple, perhaps a Gala, a Fuji or a Honeycrisp.

Ziad Asghar, SVP and General Manager of XR, Wearables and Personal AI, Qualcomm   00:07:53

Just notice: no phone in her hand, no screen to navigate, just a lightweight wearable that can see the world around them, understand what’s in front of them and respond naturally through voice. With Snapdragon, we can bring together the camera ISP, low-power sensing, on-device AI, voice and audio and run an AI model directly on the device in airplane mode. That means glasses can help answer simple but meaningful questions in the moment. This is what personal AI can become when the technology disappears into a wearable form factor.

Neil Barbour, Analyst, S&P Global Market Intelligence   00:08:36

Qualcomm said its glasses platform is now the first personal AI hardware to support one-bit models, which sharply shrink the memory footprint needed to run capable AI models on something as small as glasses. The partner roll call included Meta, Snap and Samsung, along with an expanded Google relationship, a new Snapdragon Reality platform for tethered XR glasses and audio-only Gemini glasses being built with eyewear partners including Gentle Monster.

Rick Osterloh, Senior Vice President, Platforms and Devices, Google   00:09:07

It has to understand you. It has to understand your current context, and it also has to understand the long history you have and the preferences you have. Obviously, as Google, we’re blessed to have so much information about our users. They trust us with a lot of their information, and that allows us to add capabilities like personal intelligence. With a user’s consent, we can offer deeply personalized AI experiences so you can get exactly what you want. You don’t have to reintroduce yourself to an AI every time you try to get something done. It knows that you like basketball, Mexican food and the sun. That might describe me, but for you it might be something different. This is what we want to achieve: making it really easy for users to get done the things they want to get done.

Neil Barbour, Analyst, S&P Global Market Intelligence   00:10:01

Qualcomm’s Snapdragon Start program, which gives startups reference silicon, software and a path to retail, produced two of the week’s more interesting stories. Inspecs, an eyewear manufacturer, took a working safety-glasses device from module to hardware in four months. Aware pitched a full agentic operating system built around eyewear instead of a phone screen. Qualcomm previewed a reference design for its standalone 5G AI pin, targeting 2027, and announced a Personal AI Health Alliance aimed at fall detection, ambient clinical note-taking and elder care, all pitched as on-device by necessity given the sensitivity of the data. Audio got its own refresh, too: Snapdragon Sound Elite Gen 2, built so earbuds can run their own voice-first AI agents instead of serving only as phone accessories. Qualcomm also showed a pair of earbuds with a camera, which was interesting to see.

Qualcomm isn’t trying to win any of these categories outright. It doesn’t need the AI pin to outsell the phone or the glasses to replace the watch. It needs Snapdragon silicon inside enough of the devices around you that replacing any one piece with something outside that ecosystem becomes annoying. That’s a services-company instinct wearing a chipmaker’s badge, and it’s the real reason the partner list matters more than any individual device. So let’s look at what Qualcomm means when it says its devices can handle agentic AI. Joshua Timothy, a member of Qualcomm’s Snapdragon Insiders creator community, was the first creator invited onto the Summit’s main stage. He showed an on-device workflow using FFmpeg plus a large language model running locally to sort through 500 raw video clips shot around Maui, tag subjects and scenes, build a searchable local index and automatically generate an import script for DaVinci Resolve, all overnight in a hotel room.

Joshua Timothy, Filmmaker and Snapdragon Insider   00:12:00

During the welcome event, I didn’t want to miss the fun, right? I had 500 clips to go through that night, and so I left this in my hotel running, creating the database.

Judd Heape, VP of Product Management, Multimedia, Qualcomm   00:12:09

Again, the part that people hate, right?

Joshua Timothy, Filmmaker and Snapdragon Insider   00:12:11

Exactly.

Joshua Timothy, Filmmaker and Snapdragon Insider   00:12:13

And I had it searching through the footage. But let me show you the power of what your agents can do. Once it analyzes it, it can create a script directly in DaVinci Resolve. And if I just go to my project folder, you’re going to watch all the organized clips that the AI agent did. It imports them into the project, sorts them and I have all my bins. It tells you, you know, we’ve got 32 food clips, 42 gear clips, lifestyle. And right here, they’re all sorted. We can go in and see there’s all my beach footage, all sorted, ready to edit.

Judd Heape, VP of Product Management, Multimedia, Qualcomm   00:12:58

So the language model went through, figured out from all the clips you had, the hundreds of clips, what the categories were that you predefined and grouped them into these folders and different categories. Really cool. How long would that have taken if you had done it manually?

Joshua Timothy, Filmmaker and Snapdragon Insider   00:13:10

I would have missed the party, that’s for sure. Four or five hours.

Judd Heape, VP of Product Management, Multimedia, Qualcomm   00:13:14

I’m glad you didn’t miss it. Really cool.

Neil Barbour, Analyst, S&P Global Market Intelligence   00:13:16

His pitch was that uploading and downloading 500 clips through a cloud service would take too much time to be practical, particularly on unreliable Wi-Fi. AGI Inc., whose advisers include Android co-founder Rich Miner, demoed an agent OS layer that operates third-party apps directly.

Div Garg, CEO and Co-founder, AGI Inc.00:13:34

Just imagine if your agent can act on anything that has info about your day, your notes, your meetings, your note-takers. And the best one we’ve seen is the one Vanesh introduced, Personal Scribe. It’s running on this phone.

Steve Frey, Co-founder, AGI Inc.   00:13:48

So this morning, Div and I were discussing a restaurant that we saw yesterday that we wanted to go to later.

Div Garg, CEO and Co-founder, AGI Inc.   00:13:55

I was supposed to get us all dinner there tonight with Chris Patrick joining us, and I’ve already forgotten the restaurant name, but my phone remembers. Find the restaurant which I talked about this morning and send it to the Qualcomm technology teams group on WhatsApp.

Steve Frey, Co-founder, AGI Inc.   00:14:13

Okay. So now the agent is looking at the local transcript that was recorded by Personal Scribe on the device. It’s found the name of the restaurant, and now our agent is opening WhatsApp and opening the WhatsApp group and pasting in the message that Div said. All right, there we go. There’s the restaurant. Should we send it?

Div Garg, CEO and Co-founder, AGI Inc. 00:14:49

And Chris and Vanesh, if it works, dinner’s on us. Now, our agent works across any app. So we’ll also show you how you can use this from a smart device.

Steve Frey, Co-founder, AGI Inc.   00:15:08

We’ll show you an example of using apps on the user’s phone hands-free. So Div has these headphones that I really like, and I want to get a pair, too. So I’ll have my glasses help me. These glasses are connected to the phone and I’ll say, “Hey, what are these headphones? Find one on Amazon.”

Div Garg, CEO and Co-founder, AGI Inc.00:15:40

So the glasses recognized it. It’s running on this phone. It’s searching. It’s picking. It’s checking out. But I didn’t need to open my phone for this.

Steve Frey, Co-founder, AGI Inc.   00:16:10

And these are also connected to my watch. You can see this here. Once it’s ready, it’s going to send a confirmation. Let’s see. Here we go. All right, great. Should we buy the headphones? Okay, here we go. Let’s buy the headphones. All right, great.

Neil Barbour, Analyst, S&P Global Market Intelligence   00:16:40

Memories AI showed an agent called Lucy, built on its own compact video-language models, that continuously indexes what a user’s glasses, laptop and phone see and hear into a private on-device memory graph, complete with a tap-to-forget control. In the demo, it pulled rental-car directions from a screen viewed days earlier and drafted a follow-up message from a dinner conversation that nobody had typed into a note. Liquid AI also had a compelling demo to show.

Jeffrey Li, Chief Operating Officer, Liquid AI   00:17:12

So here you’ll see it noticed what I said with Personal Scribe, that I went for a run. It saw my run behavior. It saw that I slowed down. It detected asymmetry in my gait from my smartwatch. It recognized that I was in pain, researched symptoms about knee recovery and then found that I have a lifting session tomorrow. And that’s how it came up with this text proactively for me. So the point here is that my device picked up the moment I slowed down from running to walking based on my typical pace. You’ll see here that the first two miles I was clocking a hearty 8:30 pace, and my last mile I slowed down with asymmetry from my smartwatch. It also heard what I said via Personal Scribe on the sensing hub running on the Qualcomm Snapdragon, the expletive that I said explicitly. And finally, it overheard that I mentioned knee pain when I was calling my mother.

Liquid Context stitched these two signals together to understand not just that I stopped running, but specifically I stopped running because I got hurt. And that context is what the agent then acted on proactively. It looked up how to recover. It looked up my text messages to see that on Fridays I regularly work out with my friend, even though it’s not on my calendar, and drafted a proactive text letting him know that we’re off this week. Now that’s an example of proactive AI that responds to real-world physical context that’s only able to be captured based on the devices in and around you that accompany you every day.

Neil Barbour, Analyst, S&P Global Market Intelligence   00:18:56

There was a lot of talk at the show about AI being able to buy things, which is what we’re all doing on the internet anyway. To that end, Mastercard CEO Michael Miebach appeared by video, and Chief Marketing and Communications Officer Raja Rajamannar joined Amon onstage to argue that agentic commerce needs what they called a trust layer before consumers will actually let an agent spend money on their behalf.

Michael Miebach, CEO, Mastercard   00:19:29

Artificial intelligence and the use of agents is already changing how we discover, make decisions and complete purchases. These transformational technologies are creating entirely new ways for people to interact with brands, services and one another. Agent-driven transactions are already taking place around the world, with incredible potential for future growth. But while people are always interested in new technologies, we know they will only use what they trust. People need to be confident that their payments and data are secure and that they control the process. Trust is the foundation of that confidence and must be there on day one. That’s where Mastercard comes in. We’ve delivered trust for more than 60 years. Every time you see our logo, you have an expectation of secure, simple, reliable payments and protections, and we have a commitment to deliver exactly that. Our teams are innovating and building the trust layer for agentic commerce. That trust layer brings everything you expect when you pay with a card into a world where AI agents act on your behalf. We’re excited to work with innovators like Qualcomm to build this next chapter of commerce. Thank you for the strong partnership, Cristiano.

Neil Barbour, Analyst, S&P Global Market Intelligence   00:20:53

As they put it bluntly onstage, the level of trust you need in your agent is different once money actually moves, and an established brand’s reputation is, for now, doing work the technology itself cannot yet do on its own.

So where does that leave us? Qualcomm spent the week supporting this new interface paradigm with concrete evidence, even if the foundation it laid was a little uneven. The partner demos made a credible case that AI could be distributed across a wide array of devices rather than confined to a single app or cloud service.

They also showed why Qualcomm’s strength in connectivity and power-efficient computing gives it a plausible role at the center of that shift. The harder question is how quickly consumers and developers will move with it. Qualcomm’s vision assumes a meaningful hardware-upgrade cycle across several device categories, while users invested in Apple’s ecosystem may see less benefit from a fabric built largely around Snapdragon-powered Android and Windows devices. Windows on Arm has moved well beyond its early compatibility problems, but legacy applications, drivers and specialized workloads can still create friction for users coming from x86 PCs. There is also a limit to what can reasonably run on personal hardware. Qualcomm is well positioned on performance per watt, but the largest models and most compute-intensive agent workloads will continue to rely on the cloud.