Snapdragon Summit 2026 Day 2: Personal AI, Sound, and PC Keynotes | Snapdragon for the Agentic Age

Please welcome to the stage President and CEO of Qualcomm, Cristiano Amon.

I am the co-founder and CEO of Liquid AI. I started this company 3.5 years ago, it spun out of MIT. So we've been decades-long researching how to maximize the amount of intelligence capabilities you can get into the smallest unit of computing. This is devices. We want to enable physical AI, we want to enable AI agents inside the physical AI. Flash forward, Liquid AI, we are a foundation model company. We are building very efficient models that can be the core of agentic behavior. The technology of our company, the core is a hardware-in-the-loop approach to really design condition models from scratch. We don't do post hoc optimization. You can do post hoc optimization, but we care about what we do beforehand. We designed the models to be inherently designed for the device world. The technology that comes out of it is called Liquid Foundation Models, LFM. We open-weight them. There are about 1.4 million downloads on a weekly basis on Hugging Face. You can see that the models are ranging from 200 million parameters to 24 billion parameters. This is the fabric of enabling all sorts of devices from wearables to cars or maybe even PCs like bigger computers. We built two sets of products at this company around the models. We built something: Model Plus X, Model Plus a harness that is ready to get deployed on top of devices. This is product number one. Product number two is a model development stack that allows us to take these models and give OEMs, in a B2B fashion, the opportunity to really customize these model plus harness for their applications and downstream uses that they want to do. The goal is basically very ambitious. We want to be deployed. We want to have LFM deployed on 50% of boards annually produced devices across wearables, handsets, automotive, PC, robots, and industry devices.

We are seeing AI is becoming very capable. Capabilities are increasing exponentially. We are seeing fundamental math innovations, we are seeing better reasoning models, models that can reason over longer time, and more complex workflows are getting done. But there is a disconnect. You always see that individual bottlenecks, if you look at the metric, individual productivity based on these AI models we have today is way ahead of the enterprise transformation. The gap is always between how are you bringing these capabilities and matching them to value, to actual value. That has been the problem. I feel on the AI journey we have been on, this fundamental question we have to address is production-ready AI. How can we make AI to be useful? How can we increase the amount of use of AI? That is something that is extremely difficult to get. On this journey as well, what does production-ready AI actually mean? For me it means reliable action agents to take reliable actions, quality of agents that are at the level of frontier that we are seeing, verifiable outputs. The output of the system is completely verifiable. At the same time you're satisfying the deployment constraints that you have. Use cases that might need privacy, they need connectivity, need many different aspects of security and sensitivity. Then there is use cases, then there is deployment environments. You have the physical world deployment, you have the cloud deployment. You have to be able to satisfy those things also with the technology. I think the next wave of this technology, basically hardware and kernel and model and agents, all basically co-design of these things together, is something I'm looking forward to and I think we are on the right path to get there.

I think five things have to change. The first thing is quality. We are talking about device intelligence, we need to reduce the gap between frontier-level performance on the cloud and devices. That is something that is absolutely important. If you don't, one of the reasons device AI and local AI revolution has not happened yet, for me, is just the quality bar. Getting to that quality bar is number one. Number two is context. When we are thinking about devices, you're talking about fabric of devices. Users want to use AI on one device and be able to transfer their experiences and preferences to other devices as well. How do you do that? You need to have the context layer. You need to have something that allows you a memory layer that connects these devices together. The third thing I would say is a harness. Harness is all the tooling that goes around the foundation model. When we talk about agentic AI, we're talking about a model plus a harness. That harness, the adaptation of the harness and adaptation of the model should happen at the same time. The fourth thing I would say is understanding the limitations of the physical world. For example, we are talking about real world has constraints, devices have constraints, understanding the constraints of processors and then feeding them back to the model design process. That becomes the co-design process. The last thing I would say is always developers, developers, developers. We need basically an army of developers. We need to open source, we need to give developers access to the technology to build, to really bring the AI building of devices. You are talking about local AI transformation needs attention from independent developers, and for enterprises, data is the most important thing. I believe that model building capability, we should bring it into the hands of enterprises so model building capability becomes an ownership from the enterprise side. So enterprises should own the model building capability. You give them the framework so they can get enabled. When all those things happen.

The next wave, think about where we are today. We are building legacy, massive amount of investment going to the cloud business, which is great because we have exponential amount of demand for AI. There is a lot of investment going into that direction. I think the next wave is going to be local AI. Why? Every year we have more than 2 billion processors getting built that can host the foundation model on top of it around the world. Every year 2 billion devices, this is an untapped opportunity. It takes the qualities I mentioned to really enable that local AI revolution. Often these processors, like Qualcomm, is the best positioned. You own a whole lot of that population of 2 billion. I think that's the next hot wave.

Very excited to announce today a product that I'm really proud of: Liquid Context on Snapdragon. This is a smart memory layer that allows you to read, compress, and write every sort of experience that you have throughout your active engagement with the device. And across devices, this context layer is going to enable AI systems to go from one device to another device. That is what we are announcing today. Systems that are running locally, very very close to the user. This Liquid Context we are announcing has two major components. One of them is a memory component that allows you to gather information from all sources, abstracting away, like preferences of the user. Building that memory layer for an AI agent or army of AI agents that are running locally on Snapdragon chip to really assess those information and perform jobs. These two products together, the memory layer and the agent layer are coming together. Let me show you a demo. I'd like to inspire you. This is a B2B product, so OEMs will use this. Let me show you a product to inspire how it looks like. Here is my agent, this is running on a device, this is running on a Snapdragon, it's running on an NPU. With my permission, it consumes all whatever I give it access to. It is basically sensing everything that I give it, my messages, my notes, my photos, everything that I do in my life, I give access to this. Do I trust this agent? Because it runs on my device, I give it access. It actually enables agents that are running on the NPU and proactive jobs for us. The innovation here is making agents understand when to bring you value. This is proactiveness, this is not reactive. You're not asking the system to do something for you, this is the old generation. You're talking about proactive. The system actually designs for us, it manages my calendar, and at the same time it drafts social media messages. It knows I want to post something on X all the time. It actually drafts the messages and makes it ready for me to go. That's an example of what the OEMs can do as the final kind of product. But the whole goal here is we want to prepare these as a product together to really get to market to enable enterprises across cars, mobile phones, laptops, wearables to be able to access context and very powerful agents running on Snapdragon to really enable a full experience for the user.

Mercedes-Benz, we have a partnership with those guys. You are running completely local experience of enabling in-car intelligence, that's a reality. We make it real. This is production-grade AI deployment, actual deployment of these things on Snapdragon chips inside cars.

The degree of sophistication of this use case itself, I will tell you that drives you next to the device. You have to be on the device. You got to be having a powerful system that runs with the user. The user has to always have access to a system that is continuously monitoring and continuously capturing information. Snapdragon enables that. The other thing I will tell you is that the heterogeneous design of the platform, we use all components of the processor. We are using the Hexagon for running the agents very efficiently, we are using for sensing, to abstract away, to bring all the components, we can basically read whatever device metrics we can. We use the CPU for agentic sampling. There is all sorts of elements of Snapdragon that enables this, which is really beautiful for us. Also one more thing to tell you, when we started Liquid AI, one of the most popular CPUs in the market was Qualcomm CPUs. We actually designed the second generation of Liquid Foundation Models, like four of them, for Snapdragon. The models are truly optimized already for the Qualcomm chips.

A model gives you a core capability to really interact with. Then the user is going to run through coordination. So every step you ask a question from the model, you then conclude from the results that are coming your way. This is a model-based kind of design. If you build agents around models, the user has to run with the model and really operate the model one step at a time. With the agents, the whole idea is the world around the agent becomes goal-oriented. You give a goal to an agent and the agent takes care of the operations. It's very important to note that this is where the quality of the models are becoming important. The availability of execution of an agent towards accomplishing a goal, that matters. That is the world building around agents. More automation for everyone.

For the user, I can tell you: less repetition, less coordination. The system connects information. You basically have a lot more agents for the users. For developers, I can tell you that shared context is going to improve their life. They can become more of an orchestrator, not a low-level programmer. That is what is changing for them. The thing I would say, there is a term going on, I think Tobi mentioned, the CEO of Shopify, he said slop grenades. Enterprises are dropping grenades around in the code bases. That is actually a problem because I have a feeling like a goal-based approach, like an agentic-based approach for developers, it makes a lot more sense. Why? Because they can actually compact the slop and monitor the slop in verification loops. It is like loop-based programming. The OEMs, they can build basically experience around their own data, around their own devices. Basically OEMs can own their own intelligence if you give them this capability of agentic AI. The components have to be there, which is basically what we are talking about. We have the perfect hardware and we have basically a layer on top of that to enable the perfect agentic AI experience.

If you think about AI, the way we design foundation models, we train the models on a corpus of data, the model will come out and the model is fixed. We have a snapshot in time during the training process, we put this model in production, then this model can receive input and generate an answer. The model itself is fixed. New experiences come, new preferences, new data streams are coming throughout your model. From time to time you have to actually update that model. That is suboptimal. We want to go to systems that can self-evolve. We want to have systems that can be in production and in production become lifelong learners. Like humans, the way you experience the world is the same way. We want agents to become that. That is the reality. If you think about the word 'liquid,' liquid is for flexibility. This is how we've been thinking about intelligence. We've been thinking about designing efficient and adaptable intelligence. This has been the core mission of our company. Really enabling continuously evolving systems that can match the preferences of the users at a certain time. This is kind of the direction that things are going.

Great one! I think they are complementary. Scale is going to give you more and more capability, maximizing capability on the intelligence front. But what adaptation is going to give you is personalization. That is something that you don't get from scale. Scale is the sheer volume of capable systems that you can build. Adaptation is moving you towards having an experience that is personalized to you. That is what we are seeing happening. I don't want to have a generic agent that is in my car, on my mobile phone, on my watch, on my wearables, every kind of device that I interact with. I'd like to have a personalized experience. That personalization comes from adaptation, not scale.

I can tell you the way we're seeing, as I told you, we are a B2B company. We see opportunities in automotive, really tangible and really transforming what can happen inside of the car. Bringing static experiences, connected world, we are seeing that fundamental work. That is something happening with automotive. Pursue that across mobile and wearables as well. Little more mature on that side. I would say it is going in a very very interesting direction. The point I think we are bringing on the table now is the idea of shared context across devices and at the same time the active, proactive agents. Agents basically sitting and doing jobs for you so you can focus on other things. That is kind of the things we are seeing, the biggest opportunities, like in the mobile and wearables as well. As you mentioned, I see the hybrid world would be the future. I think a lot of attention, there is an unbalanced attention to the cloud right now, which is needed. We need to host the maximum amount of intelligence, but at the same time I think we can balance it out. We can bring local AI to be really the shared experience. That kind of hybrid world, I think, is going to be the next opportunity. We see a massive demand from each of the sectors I mentioned. I can tell you that it moves us towards our goal, I told you, 50% of devices annually produced. I think it is pushing us, by 2029, if our estimations are correct, we can be on billions of devices enabling these type of experiences. An opportunity to really bring these two worlds of the cloud and local AI together.

The agents, as I mentioned, what our life future is going to look like in a way, we focus a lot less on technological progress and also managing technology rather. It enables us to focus on the things that matter to us. I feel like it will free us of our time to really think about what matters in life, like apart from work. I think that is something that I actually see happening. Proactive agents is definitely one of those directions we are going in. As I mentioned, the next revolution happening is recursive self-improvement. As AIs start to design the next generation of themselves, not just by updating the weights of their model or designing a model from scratch, but the software surrounding the model, adaptation can happen around the harness of a model. You can do updates to model, updates to the harness itself to really create experiences the users want. Then again, I can totally see every single person being a business in the future. I can imagine you can have your own products bringing it to market because we have a lot more agency with this new revolution of agentic AI and hopefully recursive self-improving AI.

What color is this apple?

This is a red apple, perhaps a Gala, a Fuji, or a Honeycrisp.

Thank you so much and good morning everyone. It's an absolute pleasure to be onstage today with our long-term partners at Qualcomm. For well over a decade, Google and Qualcomm have shared a deep conviction about the future of wearable computing. When we look back at the evolution of Wear OS, it's remarkable how far this journey has taken us. Wearables began as a glanceable display for notifications. Together, over the years of continuous innovation, Wear OS has become a smartwatch platform that keeps you connected to your health and to the people and information you care about the most. And now we've added to that foundation by bringing Gemini intelligence to Wear OS. Enabled by Snapdragon, the new Pixel Watch 3 is designed for Gemini intelligence. Your watch delivers personalized proactive help. Like if a friend asks for your visitor's contact details, Gemini intelligence instantaneously pulls it up on your watch, ready to send with a pinch. The watch is one part of an ecosystem of devices you use throughout the day, with your phone at the epicenter and with helpful experiences enabled by Gemini across all of them. The new part of this ecosystem is extended reality, or XR. Since XR devices can blend your digital and physical worlds, Gemini can understand your context and intent in new ways to help you throughout your day. To power this next frontier, we built Android XR with Samsung and optimized it for Snapdragon XR platforms with Qualcomm. It is the first Android platform built in the Gemini era. It supports a broad spectrum of products in different use cases, from headsets to glasses and everything in between.

It started with Samsung's Galaxy XR running on Snapdragon XR Gen 2. It lets you immerse yourself in an infinite workspace, enjoy cinema-scale entertainment, and experience rich 3D gaming, all with Gemini by your side. You can ask questions about what you see, have Gemini pull up videos and websites about what you are exploring, or even get a little help navigating. And soon we are extending these immersive experiences and helpfulness of Gemini to new wired XR glasses. These will be the first device running on the Snapdragon Reality platform. They run Android XR on a separate chip, allowing for a sleek form factor without sacrificing battery life or performance. That means you can take your spatial workspace or private cinema on the go, like on a plane or at the coffee shop. It can go one step further. You can also plug in devices from your pocket like your laptop or handheld gaming system, then pull them into view as a spatial window. Wouldn't it be cool to just use Gemini for help when you're working in Photoshop? Both these devices help you during deep focus sessions. Our collaboration with Malcolm is also expanding to intelligent eyewear.

These glasses are designed to give you help from Gemini that is spoken right into your ear. Whether it is translating a conversation in real time, providing turn-by-turn navigation, or answering questions about your surroundings. They also let you stay hands-free and heads-up for things like listening to music, taking photos, making calls, or tapping into your phone apps. All without reaching into your pocket. By understanding the world around you and assisting you quietly in the background, these glasses allow you to stay fully present in life, while having helpful intelligence always within reach. We are working with Qualcomm and eyewear partners Gentle Monster and others to release these glasses later this year. And of course they are running Snapdragon AR1. This progress points to a broader vision. As you heard from Cristiano and Rick in their fireside chat yesterday, future personal computing is not defined by a single device. It is all about personal AI that moves across your devices, understanding your context and intentions as you move throughout the entire day. To enable this vision, we need a unified platform that integrates silicon, compute, and AI. And so we are incredibly proud of what Google and Qualcomm are building together, and we cannot wait to show you more in the coming months. Thank you Ziad, and thank you Qualcomm, and enjoy the rest of the summit.

Thank you, I'm excited to be here at my first Snapdragon Summit. In my role, I lead the software team that brings personal AI devices to life. The first time these glasses really hit me was not in a demo. It was holding my 6-month-old in the pool, capturing the moment when both of my hands were busy keeping him afloat. It is these types of moments that could not have been captured otherwise. They really change how we interact with the world around us. If we look at the Android landscape today, one thing is clear: AI is changing the products, making them agents.

Hello, what can I help you with today?

Before any of you say it, yes, another smart glasses.

I am not from your industry. Turns out that is the point. I spent 38 years in eyewear. My industry last saw genuine breakthroughs both happen before I arrived: the spring hinge, anti-reflective coating. The industry is still very proud of the hinge. Meanwhile, Inspecs became one of the few companies on earth that designs and manufactures eyewear at volume. We own our factories. Our frames reach 80 countries through 75,000 optical and retail outlets. We are not a technology company that makes glasses or interesting form factors. We are the people who really make the things on your face. We just got tired of the hinge.

Hundreds of millions of people, let me emphasize that, hundreds of millions of people put on eyewear before a shift starts. Not because it's cool, not because the brand told them to do so, because of regulation. Success. Nobody has to persuade them. Nobody has to make it aspirational. Safety eyewear is the largest installed base of compulsory eyewear on earth. For over a century, its entire job description has been one sentence: protect the eyes. We thought it could do a whole lot more.

Here is what we found. In a defined work environment, a factory floor, an oil rig, a lab, a camera is not a novelty. It is already there, already trusted, already part of how people work. A worker snaps a photo for inspection without ever putting the wrench down. A specialist 4,000 miles away watches the same feed live and talks them through the repair in real time. That is not a feature bolted on. That's the reason the camera belongs on the space.

Which does turn out to be the harder problem. It's the only problem. The technology stopped being the obstacle the day you handed us the module and the software stack. Your kickoff to a working device, I'm sorry, the kickoff to a working device happened within four months. No eyewear company should have been able to do that alone. For that matter, no chip company should have been able to do that alone either. So, this is it. The world's first, launched here. This is our enterprise frame.

It's built for impact protection. A camera, camera isn't the bridge. Multiple microphones, speakers, voice and touch control. A bespoke high-capacity battery. All of it inside a frame that passes the same standards as the pair sitting in your toolbox right now.

That's the idea. Same protection your eyes already need. Plus hands-free AI for the questions that come up in the middle of the task, the complex jobs that need walking through. And a live feed to supervisors who are nowhere near the floor that sees exactly what you see.

There is a family. The second is audio-only. Your assistant one word away in a frame you would have bought anyway. No camera. The third is the one I'm proudest of. Prescription lenses and hearing assistance in the same pair. People wait an average of nine years before they do anything about hearing loss. It's not the cost, it's not the technology, it is that hearing loss arrives slowly, creeps up on you. There's a moment in everyone's life when a restaurant just gets too loud. It's usually not the restaurant. Glasses stopped being purely a medical device years ago. Hearing aids never did. This is the product that carries them across.

Thank you very much.