Snapdragon Summit day 2: how Qualcomm is building personal AI for phones, wearables, and PCs

AI is moving from something you talk to in the cloud into something that quietly lives on your devices, understands your world, and takes action on your behalf. Day 2 of Snapdragon Summit 2026 was all about this shift to the “agentic age” and how Qualcomm plans to power personal AI across phones, glasses, earbuds, cars, and PCs.

From big models to useful agents on your devices

The day opened with Qualcomm CEO Cristiano Amon and Liquid AI cofounder and CEO Ramin Hasani discussing how AI is evolving from large, cloud-hosted models into practical agents that live close to users.

Liquid AI started as an MIT spinout focused on squeezing maximum intelligence into the smallest possible compute units—phones, wearables, cars, and other edge devices. Instead of training huge models in the cloud and then trying to compress them later, Liquid AI designs its “liquid foundation models” (LFMs) from scratch for devices. These models range from 200 million to 24 billion parameters and are already widely downloaded in the open-source community.

Hasani argued that the next wave of AI is about “production-grade agents” that can:

• Take reliable, verifiable actions (not just generate text)
• Respect real-world constraints like privacy, connectivity, and power
• Run across many deployment environments—cloud and edge

To get there, he said five things must change: model quality on devices, shared context across devices, robust “harnesses” around models, hardware–model co-design, and a strong developer ecosystem.

Liquid Context: a shared memory layer for Snapdragon devices

Liquid AI announced “Liquid Context on Snapdragon,” a smart memory layer designed to give AI agents a persistent understanding of your life across devices.

Liquid Context has two main parts:

• Memory engine – with user permission, it can read and compress signals from messages, notes, photos, and app activity to build a private, on-device memory of your preferences and routines.
• Agent layer – local agents running on the Snapdragon NPU use that memory to proactively do work for you, not just respond to prompts.

In Hasani’s demo scenario, the agent quietly manages calendars, drafts social posts, and prepares helpful actions without being asked every time. The key idea: agents should know when to bring you value, not just wait to be called.

Liquid AI is already working with Mercedes-Benz to run fully local in-car intelligence on Snapdragon-based systems, showing how this concept extends from phones to cars.

Why Qualcomm’s heterogeneous hardware matters for agents

Hasani called Snapdragon the “platform of choice” for Liquid AI because Qualcomm’s chips are designed as heterogeneous systems. Liquid AI uses:

• The Hexagon NPU for efficient agent inference
• Low-power sensing blocks for continuous context gathering
• CPU cores for agent orchestration and sampling

Liquid AI even built a generation of its liquid foundation models specifically tuned for Snapdragon, underscoring the trend of co-designing models and hardware together—an idea echoed across other industry events like Intel’s Computex 2026 AI PC announcements.

From model-centric to agent-centric computing

Hasani drew a clear line between “model-based” and “agent-based” systems:

• Model-based: users manually coordinate each step—ask the model, interpret the answer, decide the next step.
• Agent-based: users set a goal, and the agent plans, calls tools, and executes steps end-to-end.

For users, that means less repetition and coordination. For developers, it means shifting from writing one-off glue code to orchestrating goal-based workflows with shared context. For OEMs, it means they can build branded, context-rich experiences that run on their hardware and data, instead of handing all intelligence to the cloud.

Adaptability vs scale: why personal AI must keep learning

Today’s models are mostly fixed snapshots: once trained, they don’t truly evolve in production. Hasani sees this as a major limitation.

He argued that future AI systems need to be:

• Lifelong learners – continuously adapting to new data and user preferences in production
• Personalized – scale gives broad capability, but adaptation gives you an agent that actually feels like yours

This is especially important for “physical AI” on devices you carry everywhere, where your context constantly changes. It also reinforces why the cloud vs edge debate is misguided: personalization will increasingly happen where you are—on your phone, PC, car, or glasses—while the cloud provides heavy-duty reasoning when needed.

Personal AI wearables: glasses, watches, rings, and more

Qualcomm’s Ziad Asghar then shifted the focus to personal AI devices—wearables that are always with you and can sense your world in real time.

He argued that just as the mouse and keyboard defined the PC era, and touch defined the smartphone era, the agentic era will be defined by new, more natural interfaces: voice, vision, and subtle wearables like glasses, earbuds, rings, and pendants.

Wearables matter because they are:

• Always on-body – more convenient than reaching for a phone
• Context-rich – cameras see what you see, microphones hear what you hear
• Ideal for multi-device AI – each device adds a different slice of context

The Snapdragon tech stack for personal AI

To make this vision practical, Qualcomm highlighted the full stack of technologies built into Snapdragon platforms for wearables and XR:

• Low-power sensing hubs to constantly track motion, environment, and basic signals without waking the main processor
• Always-sensing cameras that process only what’s needed for context on-device, preserving privacy and saving power
• Camera ISPs tuned for low-light and challenging scenes so AI can “see” clearly
• Hexagon NPUs and micro NPUs for efficient on-device AI, including hybrid cloud-edge workloads
• Video analytics engines with hardened perception functions for low-power interaction
• Integrated micro-power Wi‑Fi so even tiny devices can talk directly to cloud agents, not just via Bluetooth
• Advanced audio and microphone tech so AI can hear you clearly and respond with rich sound

Asghar showed a simple but powerful example: a visually impaired user wearing AI glasses that can answer questions like “What color is this apple?” completely offline, in airplane mode. This is where personal AI becomes not just an interface, but independence.

One-bit models and richer on-device AI

Qualcomm also highlighted support for highly compressed “one-bit” models that dramatically reduce memory and power requirements, enabling larger multimodal models to run on small, battery-constrained devices.

Snapdragon is becoming a key launchpad for these models across XR and wearables. This opens the door to:

• Real-time translation on glasses or earbuds
• Vision-based assistants that understand what you’re looking at
• Faster, more private on-device AI without constant cloud calls

Meta, Snap, and Samsung: building the smart eyewear ecosystem

Qualcomm is working closely with major partners to make AI glasses mainstream:

• Meta – starting with the Ray-Ban Meta smart glasses, Snapdragon provides performance, connectivity, and power efficiency for a wearable form factor. As AI has advanced, these glasses have gained increasingly capable assistants.
• Snap – the newly launched “Specs” AR glasses use Snapdragon to deliver AI assistance, work tools, entertainment, and shared experiences in a lightweight, heads-up device.
• Samsung – beyond phones, Qualcomm and Samsung are extending AI experiences to watches, XR headsets, and intelligent eyewear as part of a broader personal AI ecosystem.

Android XR and Gemini: Google’s vision for context-aware computing

Google joined the stage to outline how it’s bringing Gemini intelligence across wearables and XR, tightly integrated with Snapdragon hardware.

Key points from Google’s segment:

• Wear OS + Gemini – the latest Pixel Watch is “designed for Gemini,” offering proactive help like pulling up a contact when you’re asked to share it.
• Android XR – a new Android platform built for the Gemini era and optimized for Snapdragon XR chips. It supports a range of devices from headsets to glasses.
• Galaxy XR – Samsung’s Snapdragon XR2 Gen 2–powered headset that combines productivity, entertainment, and 3D gaming with Gemini’s contextual help.
• Wired XR glasses – lightweight glasses running Android XR on a separate compute unit, designed for mobile workspaces and private cinemas on the go.
• Intelligent eyewear with Gemini – upcoming glasses from Gentle Monster and other partners, powered by Snapdragon AR1, will offer real-time translation, navigation, and ambient assistance while letting users stay heads-up and present.

Google emphasized a unified vision: personal AI that moves across devices, understands your context, and helps you throughout the day—a theme that also underpins Microsoft’s announcements at Build 2026 and other agent-focused events like Microsoft’s AI and agentic PC strategy.

Snapdragon Start: a fast track for personal AI hardware

To help startups and new device makers get to market faster, Qualcomm introduced “Snapdragon Start,” a program that bundles:

• Compact Snapdragon modules – such as tiny AR1-based boards for glasses
• End-to-end software – spanning device firmware, companion phone apps, and cloud services
• Ecosystem support – reference designs, manufacturing partners, and integration help

The goal is to collapse time-to-market for new personal AI devices, from AR glasses to pins and beyond.

Hands-free AI glasses for the enterprise and safety

One of the standout examples from Snapdragon Start is Inspecs, a company with decades of experience in eyewear manufacturing and safety glasses.

Instead of chasing consumer novelty, Inspecs focused on the world’s largest installed base of compulsory eyewear: safety glasses used in factories, labs, and industrial sites. They asked: if workers already have to wear protective eyewear, why not make it smart?

Inspecs’ first Snapdragon AR1+–based enterprise frame includes:

• Impact-rated protection that meets existing safety standards
• A built-in camera for inspections and remote assistance
• Multiple microphones and speakers for clear voice interaction
• Touch and voice controls
• A high-capacity battery integrated into a familiar frame

On a factory floor or oil rig, a worker can snap a photo or stream live video to a remote expert without putting down tools. This isn’t a gimmick; it’s a direct productivity and safety upgrade.

Inspecs is also developing:

• An audio-only smart frame for always-available voice assistance
• A prescription frame that combines vision correction with hearing assistance, designed to lower the barrier to adopting hearing support

Aware: an agentic operating system with eyewear at the center

Startup Aware is taking a more consumer-focused angle, building what it calls an “agentic operating system” that spans all your devices, with eyewear as the natural hub.

CEO Ash described the shift this way: if AI makes computers intelligent, why are we still poking at screens with our fingers? We should be talking to them, and they should see and hear what we do.

Aware’s “Agentic System One” aims to:

• Let users issue natural language goals (e.g., “Find a great Italian near the hotel at 7:30, book it, send details to my friend, and get me there on time”)
• Handle all the app-hopping and coordination in the background
• Move context and visuals between eyewear, phone, and car automatically
• Store user memories cryptographically so only the user can access them

Aware is working with leading eyewear brands, using Snapdragon AR1 modules small enough to disappear inside stylish frames, so the tech doesn’t compromise design.

Personal AI pins: a new wearable category

Qualcomm also previewed a reference design for a “personal AI pin” powered by Snapdragon Elite. This compact wearable packs:

• Standalone 5G connectivity
• Dual cameras
• On-device AI acceleration

The reference design gives partners a blueprint to explore this emerging category—pioneered by early products like Humane’s AI pin—and experiment with different form factors and use cases.

The Personal AI Health Alliance

One of the most impactful announcements was the formation of the Personal AI Health Alliance—a global community of healthcare and technology organizations focused on on-device AI for real-world medical needs.

Example areas the alliance is targeting:

• Fall prediction and prevention – using intelligent wearables to anticipate fall risks in older adults before they happen, enabling proactive interventions.
• Reducing clinician “pajama time” – ambient AI that listens during patient visits, drafts documentation, flags warning signs, and reduces hours spent on electronic health records after hours.
• Real-world patient monitoring – combining wearable sensors, clinically-informed digital measures, and on-device analysis to better understand daily patient experiences.

Crucially, these solutions emphasize on-device AI for privacy and real-time responsiveness—exactly the kind of hybrid cloud-edge architecture Qualcomm is pushing across sectors.

Snapdragon Sound Elite Gen 2: intelligent hearables for the AI era

Audio is another major frontier for personal AI. Qualcomm’s Andy Fisher introduced Snapdragon Sound Elite Gen 2, a premium audio platform built specifically for AI-first hearables.

Key improvements include:

• 30% smaller silicon footprint
• 2x compute and AI performance
• 40% lower power consumption
• Integrated Bluetooth and micro-power Wi‑Fi for always-connected earbuds

This means earbuds and headphones can:

• Run on-device AI agents that understand voice commands, context, and environment
• Adapt noise cancellation in real time to sudden loud sounds
• Deliver high-fidelity audio and ultra-low latency gaming
• Connect directly to services, not just relay audio from a phone

Snapdragon Sound apps and agents

To support these new capabilities, Qualcomm is launching an ecosystem of “Snapdragon Sound apps and agents.” Think of them as native “apps” for your earbuds, created by service providers and developers, and updated over time just like phone apps.

Examples include:

• Music and entertainment services streaming directly to earbuds over Wi‑Fi
• AI assistants and translation agents that run locally or in the cloud
• Context-aware experiences that blend audio, voice, and even visual cues

Qualcomm is working with partners like HP and Bose on new Snapdragon Sound Elite–based products, and with an unnamed partner on a novel “audio + visual” hearable that adds a camera—not for recording content, but to give the AI agent an extra layer of context.

Modular and Mojo: a unified AI software stack for Snapdragon

On the software side, a major move was Qualcomm’s acquisition of Modular, whose cofounder Chris Lattner (also known for LLVM and Swift) is now EVP of Advanced AI Software at Qualcomm.

Modular has built a unified AI stack consisting of:

• Mojo – a high-performance programming language that combines Python-like syntax with C++-level speed and control, now open-sourced under Apache.
• Max – a serving framework for running AI models efficiently across data centers and local machines.
• Modular Cloud – infrastructure for scaling AI workloads across heterogeneous accelerators, handling provisioning, scheduling, optimization, and uptime.

The stack already supports six different accelerator architectures, including Qualcomm’s Dragonfly data center NPUs, and is used to serve trillions of tokens in production.

The plan is to bring this entire software platform to the Snapdragon portfolio—phones, PCs, wearables, and more—so developers can target a single stack and still get top performance on Qualcomm hardware.

Mojo on Windows and Snapdragon PCs

Microsoft joined via video to announce native Windows support for Mojo. This means developers can:

• Write high-performance AI and systems code in a Python-like language
• Run it efficiently on Windows devices, including Snapdragon-powered PCs
• Reuse code across cloud, edge, and local environments

This is part of a broader push to make Snapdragon PCs a first-class platform for AI development, not just consumption.

Snapdragon X series: AI PCs reach critical mass

Kedar Kondap from Qualcomm’s compute team provided an update on Snapdragon X series PCs and the broader AI PC landscape.

Highlights:

• Over 7,300 apps now run on Windows on Snapdragon, making early compatibility concerns largely a thing of the past.
• More than 170 PC designs from major OEMs are in market or in the pipeline.
• Over 1,800 ecosystem devices have been tested for interoperability.
• Devices are sold in 12,000+ stores across 120 global retailers, with 150,000 sellers trained.

Qualcomm’s view is that AI is becoming a distributed fabric across cloud, edge, and devices. The key questions are no longer “how big is the model?” but “where should this inference run?”

The PC, with its thermal headroom, memory, and compute, becomes the anchor node for many agent workloads—especially those that need to reason across multiple tasks, apps, and devices. Snapdragon X series chips combine powerful CPUs, GPUs, and NPUs to run large models locally, all day on battery, without constant cloud round-trips.

ISV momentum: on-device AI apps

Kondap highlighted independent software vendors building on Snapdragon PCs, including:

• Local LLM apps that run entirely on-device for privacy and speed
• Productivity tools that use on-device models for summarization, drafting, and search
• Developer tools that allow fast, offline experimentation with AI models

The message: Snapdragon PCs are no longer just “ARM experiments”—they are becoming serious AI development and productivity machines.

Microsoft Surface: new Snapdragon X series devices

Microsoft’s Brett Ostrum joined onstage to unveil new Surface devices powered by Snapdragon X2 series:

• Surface Laptop (13-inch) with Snapdragon X2 Plus – up to 22 hours of battery life and 95% faster local AI inference compared to the previous generation.
• Surface Pro (12-inch) with Snapdragon X2 Plus – 25% brighter display, over 60% faster graphics, and an optional 5G model for always-connected computing.

Both devices are available for preorder and are designed to keep more AI workloads on-device for performance, privacy, and offline use.

Microsoft also teased a new Surface mouse with haptic feedback, extending the brand’s focus on tactile, delightful user experiences.

Lenovo, HP, and the new Google Book category

Lenovo and HP also showcased their Snapdragon-powered devices and deepening collaboration with Qualcomm:

• Lenovo IdeaPad Vibe – a new Snapdragon-based laptop with bold colors and a focus on personal expression and mobile performance.
• HP Google Book 14 – part of a brand-new category Google announced: Google Book, a laptop platform that blends Chrome OS and Android foundations and is “the choice for Android phone users.” It runs on Snapdragon X Elite and offers up to 19 hours of real-world battery life, full desktop Chrome, and native Android apps (including offline support for titles like Adobe and Netflix).

Google Book is built on the same Android tech stack as phones, which allows tight cross-device continuity (shared files, apps, and workflows) and deep integration with Gemini. It’s a new take on what a laptop can be in an Android-first world.

Linux on Snapdragon X2: a big win for developers

Responding to years of developer requests, Qualcomm announced official Linux support on Snapdragon X2 series platforms.

Key pieces include:

• Canonical Ubuntu certification – Snapdragon X2 platforms will be certified to run Ubuntu, with full validation for performance and stability, targeted for the first half of next year.
• OEM Linux support – ASUS is bringing Ubuntu to Snapdragon X2 devices like the Zenbook A14 and A16, making them attractive options for developers and enthusiasts.
• Humane’s Ultra Horizon PC – an upcoming Linux-based agentic PC powered by Snapdragon X2 Elite, built around Humane’s Agentic OS.

This gives developers a full stack of options on Snapdragon PCs: Windows, Google Book, and Linux.

Partnerships, fandom, and the Manchester United tie-in

To highlight the importance of partnerships, Qualcomm and Microsoft also showed how tech, brands, and fandom can intersect.

Qualcomm is a sponsor of Manchester United, and Microsoft Surface is now a back-of-shirt sponsor. Together, they unveiled an exclusive Manchester United–branded Surface Laptop powered by Snapdragon, which will be available only to a limited number of fans.

While it’s a marketing move, it underscores a bigger theme of the summit: AI and devices are becoming part of personal identity and culture, not just tools.

The big picture: a distributed, personal, agentic future

Across all the announcements, a consistent vision emerged:

• Agents, not just models – goal-oriented AI that can plan and act across apps and devices.
• Distributed intelligence – cloud, edge, and devices working together, with orchestration deciding where each inference should run.
• Personal context – memory layers like Liquid Context, wearables that see and hear your world, and PCs that hold your long-term work context.
• Privacy and locality – more AI running on-device, with user-controlled memories and fewer round-trips to the cloud.
• Unified software stacks – tools like Mojo, Max, and Modular Cloud making it easier for developers to target heterogeneous hardware.

Qualcomm’s bet is clear: the agentic age will be built on a fabric of Snapdragon-powered devices—phones, glasses, earbuds, cars, and PCs—all sharing context and intelligence. And if Day 2 of Snapdragon Summit 2026 is any indication, the ecosystem is lining up to make that happen.