AI Agents Are Moving Into Your Pocket and Onto Your Face
At WAIC 2026, AI agents are leaving data centers for consumer devices. STEPX Neo debuts as the first agent-native phone, smart glasses shipments surge 167% YoY, and on-device AI is quietly reshaping how we relate to intelligence.
The AI Agent Isn't in the Cloud Anymore—It's on Your Face
At WAIC 2026 in Shanghai, one trend was unmistakable: AI agents are packing their bags and moving out of the data center. Their new homes? Your pocket and your nose bridge.
For the past two years, "using AI" meant opening a browser tab and chatting with a model. But walking through the exhibition hall this July, the chat box was nowhere to be found. In its place: phones, glasses, and wearable devices that don't wait for you to type—they just understand what you need and get it done.
The Smartphone Gets an Agentic OS
The headline act was STEPX Neo, the world's first "large-model-native" agent phone, built by Chinese AI company Stepfun under its STEPX brand. It runs Step AOS, an agentic-native operating system that embeds a personal AI agent called Amoo directly into the device layer.
This isn't a fancier Siri. Amoo can perceive on-screen context, understand multi-step intentions, and orchestrate across apps. Point it at a travel request—"book me a trip to Beijing next week"—and it coordinates with Ctrip for flights and hotels, checks your calendar for conflicts, and routes payment through Alipay. All without you tapping between apps.
STEPX Neo recently passed China's national L3 certification for AI terminal intelligence and earned WAIC 2026's "Crown Jewel" award. Nubia and Honor also showcased their own system-level agent phones, signaling that the agentic phone isn't a one-off experiment—it's the next battleground.
Smart Glasses: The Second Front
If a phone still feels too distant, glasses are the more radical answer.
IDC reports that global smart glasses shipments grew 167% year-over-year in Q1 2026, with full-year projections hitting 13.6 million units. Omdia expects China to become the second-largest AI glasses market globally, with approximately 1.2 million units shipped in 2026—about 12% of the worldwide total.
The product landscape is heating up fast. Huawei launched its first HarmonyOS AI glasses in April, embedding its Xiaoyi agent for real-time video interaction and glance-to-pay functionality, at just 35.5 grams. Xiaomi's AI glasses pack a "Super Xiaoai" assistant with real-time translation across 10 languages. On the professional side, Rokid leads China's market with 21.2% retail share, followed by Kuake (15.4%), XREAL (11.6%), and Thunderbird Innovation (10.4%), per AVC monitoring data.
Arm has quietly become the foundation underneath it all: 99% of smart glasses on the market today run on Arm architecture.
What Makes This Possible
Cramming a large language model into a device that fits on your face requires real engineering, not just ambition.
At Computex 2026, Qualcomm unveiled the Snapdragon X Elite with a Hexagon NPU hitting 100 TOPS, demonstrating a quantized 70-billion-parameter model running locally on a laptop at 38 tokens per second. At WWDC 2026, Apple announced iOS 20 will ship with a 3B on-device model optimized via MLX, enabling fully offline Siri agent interactions.
One of the more impressive demos at WAIC came from Arm and Alibaba's Qwen team. Using Arm's second-generation Scalable Matrix Extension (SME2), they built the industry's first commercial TTS solution running a full Qwen-Audio model on a CPU—on-device voice cloning, no cloud required.
The macro numbers back this up: IDC estimates over 40% of global AI inference workloads have already shifted from cloud to edge devices. Quantization, distillation, and model compression are no longer research curiosities—they're shipping in consumer hardware.
From Tools to Environment
The real story here isn't "offline AI works now." That's a technical footnote.
The deeper shift is relational. When AI stops being a thing you open a chat window to talk to and starts being a layer embedded in your operating system, hanging on your face, perceiving your context and acting on your intentions—it ceases to be a tool. It becomes infrastructure. Ambience.
You're not "using AI" anymore. AI is just there.
As Arm's Edge AI EVP Chris Bergey put it at WAIC: "The value of AI agent phones is that AI finally starts genuinely serving the user, unlocking new use cases and experiences."
In 2026, the question is no longer "what can AI do?" It's "how close is AI to you?" And the answer keeps shrinking—from a data center, to your pocket, to the bridge of your nose.