StepFun's AI smartphone play: Yin Qi is betting on commercializing hardware-software integration

Article author: Zheng Minfang

Source: Wall Street News

On July 13, StepFun unveiled its AI terminal brand STEPX in Shanghai, simultaneously launching Step AOS—an agent-native operating system—its next-generation personal agent Amoo, and its first large-model-native agent smartphone, the STEPX Neo.

Rather than simply launching a new smartphone, StepFun aims to test whether integrating foundational models, operating systems, agents, and hardware terminals into a cohesive software-hardware ecosystem can offer a viable path for commercializing large AI models in China.

Amoo, built into the STEPX Neo, serves as the unified entry point for personal agents. Users only need to state their needs, and the agent will break down tasks, select appropriate tools, and coordinate system capabilities and external services to execute them.

According to StepFun’s vision, users can ask Amoo to plan trips, book hotels and transportation, manage documents, schedule appointments, screen resumes, or automatically handle routine repetitive tasks.

Underpinning this capability is Step AOS, which builds a runtime environment for agents from the ground up based on the principle of human-agent symbiosis. Its underlying AgenticOS atomic engine operates across multiple ecosystems, including Android, Linux, and RTOS.

However, whether an agent can truly complete tasks for users depends not only on the model and operating system but also on its ability to access today’s highly walled-off internet service ecosystems.

Currently, STEPX has already completed integrations with initial partners including Ctrip, Alipay, Didi, Meituan, WPS, and CapCut, covering scenarios such as travel, payments, local services, office productivity, and content creation. For instance, when a user expresses a travel need, the agent can invoke Ctrip’s travel services, Didi’s mobility capabilities, and Alipay’s payment functions to complete the entire task chain—from itinerary planning to booking and payment.

This ecosystem approach differs significantly from the previously noted Doubao mobile assistant.

The Doubao mobile assistant primarily uses a GUI-based approach—identifying on-screen content and simulating user taps, swipes, and inputs to complete tasks within existing app interfaces.

Its advantage lies in not having to wait for individual app developers to open up APIs, theoretically enabling faster coverage across more applications; however, because it bypasses apps' native interaction and distribution systems, it is also more likely to face restrictions from platforms.

Jieyue Star has opted for a comparatively moderate API collaboration route.

STEPX directly calls service APIs provided by partner developers, allowing app vendors to decide which capabilities to expose to agents and jointly define terms regarding user authorization, data usage, payment confirmation, and liability boundaries.

Although this model entails slower initial integration, it rarely infringes directly on app developers’ core interests and thus stands a better chance of fostering stable, long-term partnerships.

The trade-off is that nearly all the pressure of ecosystem development falls squarely on Jieyue Star.

Currently, STEPX has integrated only a subset of leading applications. Whether high-frequency super apps such as WeChat, Taobao, Douyin, and Xiaohongshu are willing to open sufficiently deep APIs will directly determine how many tasks Amoo can genuinely accomplish.

According to All Weather Tech, Jieyue Star plans to co-develop agent invocation standards with partners and is also in discussions with Tencent to explore potential collaboration models for integrating WeChat and other nationally dominant applications.

Hardware itself presents additional challenges.

Agents need to continuously process multimodal information such as audio, images, and text. Running multiple agents in parallel further increases demands on chip computing power, energy consumption, and thermal management. StepFun stated that the company is already working with partners across the industrial chain to adapt edge-side chips and sensing hardware, aiming to alleviate issues of insufficient edge computing capacity and excessive power consumption.

Meanwhile, StepFun is also proactively building up the talent pool and organizational capabilities required for its terminal business.

Currently, StepFun’s terminal business is led by President Ni Jiayue, who previously served as Vice President of Huawei Consumer Business Group (China), President of Honor’s Global All-Scenario Business Unit, President of Honor China, and President of Honor’s Global New Retail Business Unit, bringing extensive experience in smart devices and retail operations.

However, StepFun does not intend to adopt the traditional smartphone manufacturers’ profit model.

StepFun stated that STEPX will not rely on hardware sales margins or monetize through pre-installed advertisements. The company has not set any rigid shipment targets at this stage; its top priority is refining the agent user experience and building user trust in AI agents.

According to its roadmap, STEPX will launch a user 'Wish Pool' next week, where engineers will work one-on-one with early adopters to help them build personalized AI agents. On August 20, it will also partner with Bilibili to launch the 'Imagination Program,' collaborating with content creators (UPs) to explore real-world use cases for AI agents.

StepFun hopes to identify the right product form and monetization logic for the age of AI agents through co-creation with users.

In Yin Qi’s view, hardware in the future will merely serve as a carrier for agent-based services.

"People may need to redefine what hardware means in the future. Hardware is just one part of the entire system—it’s merely a container. Ultimately, what users are really using is the Agent service," Yin Qi said.

Compared with relying solely on model API fees, integrating software and hardware gives StepFun at least one direct channel to reach users, validate its products, and explore viable monetization models.

But as StepStar moves from being a model company to reaching end consumers, the challenges it faces are no longer purely technical. The real test of this hardware-software integration experiment lies in whether it can convince internet platforms to open up their services, persuade users to grant critical permissions, and establish a sustainable subscription-based business model beyond hardware sales.

265 Views