The smartphone industry has spent years adding AI features. China’s newest devices are attempting something more ambitious: removing the need to think about individual apps at all.

At the 2026 World Artificial Intelligence Conference in Shanghai, StepFun, Nubia and Honor presented phones built around “agentic” AI—systems designed to understand a goal, divide it into smaller actions and complete those actions across multiple services.

The pitch is simple. Instead of opening five apps and tapping through 20 screens, tell the phone what outcome you want and let it handle the workflow.

That sounds like a small interface change. It could become a major shift in who controls mobile commerce, travel bookings, payments and everyday digital services.

From command to outcome

StepFun offered one of the clearest demonstrations.

At its WAIC booth, a staff member asked the company’s StepX Neo smartphone to find a nearby electric vehicle charger and arrange for his usual coffee to arrive at home as the charging session was ending.

The phone produced separate service cards for the charger and coffee order, using stored information such as the vehicle model, preferred charging speed, usual charging time, coffee choice and home address. Final confirmation remained with the user, while it was not clear whether the charging reservation itself had been completed.

That last detail matters.

READ MORE: Travala Bets on Agentic AI for Onchain Travel Booking

Agentic AI is impressive when it understands the relationship between two requests. It becomes genuinely useful only when it can reliably finish both.

StepFun says its Android-compatible Step AOS connects with major Chinese platforms including Alipay, Trip.com, Didi Chuxing and Meituan. This is the real foundation of the product. The model may reason, but the operating system needs permission to access payments, transport, travel inventory, delivery networks and personal preferences before it can turn reasoning into action.

StepFun has not yet disclosed the StepX Neo’s price or release date.

Three versions of the AI phone

Nubia used WAIC to present the NaviX Ultra, built around ByteDance’s Doubao assistant.

The company describes it as a mass-produced AI agent phone capable of handling schedules, information searches and multi-app commands through voice or text. However, the device was not available for hands-on testing at the event, while official pricing and launch details remain limited.

Honor took a more physical route.

Its Robot Phone combines an agentic operating system with a mechanical camera gimbal that can move independently, track people and respond through gestures. Honor also demonstrated cross-app tasks such as ordering a cake, booking a ride and reserving a karaoke room as one connected birthday plan.

READ MORE: How Agentic AI Is Rewiring Travel — and Why APIs Must Catch Up

Reservations have opened in China, but the product is still closer to an early category-defining experiment than a proven mainstream handset.

Honor chief AI scientist Huang Fei captured the wider ambition well:

“The essence of an agentic OS isn’t adding an AI assistant to an operating system. It’s rebuilding the operating system around users’ intentions and tasks.”

The ecosystem problem

The challenge is not simply whether the AI understands a request. It is whether airlines, hotels, mobility providers, payment platforms, restaurants and other services allow the agent to act on the user’s behalf.

“An agentic operating system cannot be built by any single company alone,” Alibaba vice president Xu Zhuhong said at WAIC.

“It requires an open ecosystem and shared standards rather than a closed product,” he added. “Companies across the industry chain, from foundation models to platforms and applications, must work together.”

This is where the category could slow down.

Users will not tolerate a phone that occasionally books the wrong train, orders food to an old address or confirms a payment without making the next step obvious. Better transaction receipts, clearer permission controls, visible cancellation options and stronger explanations of why an agent selected one service over another would improve these early concepts considerably.

Agentic phones are also not for everyone. People who prefer manual control, avoid storing personal preferences or feel uncomfortable giving AI broad access to payments, location and communications may find the experience more intrusive than convenient.

Conclusion

China’s agentic phones are not emerging in isolation.

Google is bringing multi-step app automation to Android and has introduced AppFunctions, allowing developers to expose app capabilities directly to AI agents. Samsung is positioning Bixby, Gemini and Perplexity as agents that can coordinate actions across Galaxy apps. Apple’s new Siri AI is being designed around personal context, onscreen awareness and system-wide app actions.

The difference is emphasis.

Google, Samsung and Apple are extending powerful existing ecosystems. StepFun, Nubia and Honor are using new hardware to argue that the app-by-app smartphone model itself is becoming outdated.

For travel and connectivity companies, that could be the real story.

In an agentic interface, being visible in an app store is no longer enough. A service must be structured, trusted and technically accessible enough for an AI agent to select, compare and purchase it automatically.

That changes the competitive question. An eSIM provider, airline, hotel platform or mobility company may no longer be competing only for search rankings, advertising clicks or home-screen visibility. It may be competing to become the preferred service inside an AI agent’s decision-making system.

The next battle will not necessarily be for space on the smartphone screen. It may be for the right to become the invisible service an AI chooses before the traveller ever sees a list of options.