What Changed and Why It Matters

China’s smartphone makers are moving past the app grid. They’re shipping agent‑native phones that plan and execute tasks across apps.

ZTE, StepFun, Oppo, Vivo, and Honor have unveiled systems where an AI agent operates your apps for you, not just answers questions. Tencent is pushing agents into WeChat. Even ByteDance is powering phones with its Doubao assistant.

The signal: the phone UI is shifting from “tap apps” to “delegate tasks.” This isn’t just a new feature. It’s a new OS layer with memory, permissions, and cross‑app control baked in.

What most people miss: the agent replaces your app grid, not your apps.

Why now? Two forces align: on‑device AI is finally fast enough, and China’s super‑app ecosystem makes cross‑app automation practical. The result is a credible post‑app model arriving first in China.

The Actual Move

Here’s what shipped and who’s moving:

• ZTE unveiled the NaviX Ultra, an “agent-first” device that ditches the classic app grid for a single AI entry point. The agent orchestrates tasks across installed apps, presenting flows instead of icons.
• StepFun, a Shanghai startup reportedly backed by Tencent, introduced a phone designed as the AI product itself. The agent becomes the primary interface that plans, books, and follows up—cross‑app.
• Oppo, Vivo, and Honor rolled out AI‑agent operating systems that plan multi‑step tasks across apps. Reported features include agent harnesses (safe ways to call app actions), long‑term user memory, and privacy‑aware permissions.
• Tencent confirmed a push to bring AI agents into WeChat, aiming to automate everyday tasks inside China’s dominant super‑app.
• ZTE’s Nubia line surfaced again in coverage of a rapid “72‑hour war,” highlighting a model using ByteDance’s Doubao assistant, paired with Qualcomm’s Snapdragon 8 Elite for on‑device acceleration.
• Industry analysis frames this as part of a broader split: the US leans API‑first, app‑centric agents; China leans device‑native, super‑app‑integrated agents.

Mission shift: from “assistant in an app” to “OS‑level operator that gets things done.”

The Why Behind the Move

This is a deliberate strategy, not a demo cycle.

• Model

Agent‑native OS layers standardize how AI calls app capabilities. Think “intents + permissions + memory” as first‑class primitives. On‑device inference handles fast steps; cloud models handle planning.

• Traction

China’s super apps already coordinate payments, messaging, travel, and services. That shortens the path from intent to action. Early agent flows (book, pay, confirm, remind) feel native on day one.

• Valuation / Funding

Backers like Tencent signal a platform‑level bet. When the interface changes, value pools shift—from app icons and ad inventory to agent‑mediated transactions and subscriptions.

• Distribution

OEM control plus WeChat distribution is a powerful combo. Ship the agent with the device; amplify it inside the country’s default app. Distribution beats model quality when the default is good enough.

The moat isn’t the model—it’s the distribution and default placement.

• Partnerships & Ecosystem Fit

Agent harnesses give developers a new surface: expose safe actions, not full app UIs. Expect SDKs for intents, capability registries, and incentive models tied to completed tasks, not downloads.

• Timing

Smartphone demand needs a reason to upgrade. “Your phone now does the doing” is a clearer pitch than “20% faster CPU.” On‑device NPUs and modern SoCs finally make it viable.

• Competitive Dynamics

• China: OEMs + super apps = fast integration, strong utility. Apple/Google alternatives are less integrated in China, leaving white space.
• US: assistants live inside apps first. Privacy norms and app store economics slow OS‑level agent control.
• Cross‑border: expect divergent UX patterns and SDKs.

• Strategic Risks

• Reliability: agents must be right, reversible, and inspectable.
• Safety/permissions: cross‑app control needs transparent consent and audit logs.
• Ecosystem pushback: if agents reroute demand, app makers will fight for placement and economics.
• International expansion: different privacy regimes complicate export.

What Builders Should Notice

• Design for intents, not screens. Expose actions with clear contracts and guardrails.
• Memory is a feature. Store preferences and state to cut steps and earn trust.
• Distribution compels adoption. Default placement beats marginal model wins.
• Measure completions, not clicks. Agents shift KPIs to outcomes and reliability.
• Safety UX is product‑critical. Permissions, previews, and undo build confidence.

Here’s the part most people miss: the winning agent will explain itself as well as it acts.

Buildloop reflection

Every platform shift begins when someone makes the default simpler.

Sources

Dapta — AI News Smartphone: China’s First AI Agent Phone
Caixin Global — Chinese Smartphone Makers Bet on AI Agents to Revive Demand
OnClick Innovations — China’s Agentic AI Phones Ditch the App Grid Entirely
Cobo — Chinese Smartphone Makers Bet on AI Agents to Redefine Mobile Ecosystem
Beam — AI Agents in 2026: US vs China’s Two Different Futures
YouTube — China’s AI Agent Advantage
Startup Fortune — StepFun Unveils the First Agentic AI Phone Ahead of Apple and OpenAI
AI Agent Economy (Substack) — China’s First AI Phone Sparked a 72-Hour War. Here’s Why
France 24 — China tech giant Tencent bets on AI agents