AI operating systems are making a concentrated debut, with terminal manufacturers vying for the next-generation entry point

As large language models evolve from answering questions to understanding intentions, planning tasks, and executing operations across applications, the value of operating systems as the intermediate layer is being redefined.

A battle for the right to define the entry point of the AI era is quietly unfolding at the operating system level.

From July 17 to 20, the 2026 World Artificial Intelligence Conference (WAIC) is set to be held in Shanghai. According to previews, over 300 products will make their global debut, including landmark new releases such as the StepVerse Agent Operating System and the world's first AI agent smartphone.

Honor will also make its first appearance on the WAIC stage on July 18, hosting the sub-forum titled "From Digital Screens to Embodied Intelligence" to discuss the evolutionary trend of Agentic OS shifting from "tools" to "partners". The Honor Robot Phone, as the first device to implement partial functions of Agentic OS, will showcase its interaction capabilities, perception scheduling, and decision-making execution capabilities in depth for the first time.

CAIJING New Media observed that since June this year, Huawei has launched HarmonyOS 7, announcing its transition from the "interconnection of all things" to the "Agent era"; Microsoft has declared that Windows will evolve into an Agent operating system, making approximately 1.6 billion Windows PCs worldwide potential platforms for running local AI agents; Rokid released the AIOS smart glasses YodaOS, claiming to lead the industry into a new era driven by native AI operating systems; the openKylin open-source agent operating system, developed under the leadership of the National University of Defense Technology, was also launched in Beijing.

Behind this series of moves, the entire mobile terminal industry is undergoing a paradigm-shifting leap centered on AI technology, with the entire industry simultaneously reshaping product forms, the interaction logic between humans and machines, and the full-link system usage experience.

During his keynote speech at the opening day of MWC Shanghai 2026, Fang Fei, President of Honor's Product Line, pointed out that terminals in the next decade will no longer be "containers for applications" but "stages for agents". Terminals connect users at one end, understanding real people and real scenarios; at the other end, they connect to cloud models, computing power supply, and the entire AI service ecosystem.

Zhu Mingming, founder of Rokid, stated straightforwardly that each era has its representative terminals: the DOS era had PCs, the GUI era had Windows, and the mobile internet era had smartphones. In his view, the AI era will inevitably bring about a new transformation in interaction modes, and AIOS should allow users to directly access native services through AI instead of going through the operational path of traditional apps.

As large language models evolve from answering questions to understanding intentions, planning tasks, and executing operations across applications, the value of operating systems as the intermediate layer is being redefined. Players across different tracks, including smartphone, PC, smart glasses, application, and model manufacturers, are flocking to this race to be "the one awakened first by users".

Terminal Vendors Shift OS to Agent Architecture

At the 2026 China Internet Conference on July 8, Chen Kai, Deputy Director of the Information and Communication Administration Bureau of the Ministry of Industry and Information Technology, revealed that the State Council deployed and implemented the "AI +" action in 2025, explicitly proposing that by 2030, the application penetration rate of new-generation smart terminals and agents will exceed 90%.

Industry insiders generally believe that compared to simply outputting model capabilities, terminals can not only host AI services but also continuously accumulate user data and build an ecosystem. Moreover, compared with API calls and software subscriptions, hardware can establish longer-term business relationships.

Smart terminals are the core carriers for AI to interact with humans. Honor previously predicted that the industry is witnessing three major changes: interaction is evolving from GUI to Agentic UI centered on natural language and intention expression; the value focus is shifting from operating the app ecosystem to understanding users' long-term context; the distribution model is transforming from consumer-facing ToC to agent-oriented ToA.

Fang Fei stated that the terminal itself is a collection of computing power, sensors, interaction, and action capabilities. It is not a downstream entity defined by model capabilities, but the only way for AI to enter real life and reach users. AI will reinvent hardware. With the continuous enrichment of terminal forms, various devices are in urgent need of a unified management framework. In June this year, Honor systematically defined Agentic OS, the next-generation mobile terminal operating system, for the first time — a human-centric system with four core features: intention-driven, natural interaction, proactive intelligence, and inherent cross-device connectivity, serving all-scenario devices.

Huawei HarmonyOS has also fully transitioned to the Agent era. In June this year, Huawei launched HarmonyOS 7, realizing the Agent-based reconstruction of the operating system for the first time. It rebuilt the underlying architecture from four dimensions: communication, interaction, AI, and data, breaking down the full-scenario ecological barriers among smartphones, PCs, vehicle infotainment systems, and smart home devices. According to official data, HarmonyOS is already the second-largest smartphone operating system in the domestic market, with the total installed volume of HarmonyOS devices exceeding 1.3 billion, over 70 million smartphones running HarmonyOS, and more than 11 million registered developers on the platform.

Nubia, a subsidiary of ZTE, has chosen a different path. On July 9, Nubia officially announced that the world's first AI agent smartphone, positioned as a mass-produced flagship model, will make its debut at WAIC on July 17. In March this year, ZTE Corporation stated during institutional research that in terms of technical paths, the company will deeply integrate AI agents into the operating system to create a truly native AI experience, rather than a simple functional plug-in. Currently, the company is further upgrading system-level AI capabilities, integrating ZTE's self-developed Co Claw intelligent scheduling technology to achieve seamless collaboration across applications and ecosystems, support automated task execution in more complex scenarios, and improve user operation efficiency and interaction intelligence.

Notably, on July 9, Xiaomi announced that its open-source IoT operating system openvela has been positioned as an "AI hardware intelligent base". It is reported that Xiaomi Vela technology has been applied on a large scale across 180 million devices in numerous categories, including smartwatches, bands, earphones, smart speakers, and AI glasses.

A document from the China Electronics Information Industry Development Research Institute pointed out that the demand for AI-native devices and cross-terminal collaboration is driving operating systems to transform into global intelligent service hubs, with core features reflected in cross-device collaboration, cloud-edge-end computing power scheduling, and deep AI-native integration. In practical application scenarios, next-generation operating systems can support seamless application migration, real-time data synchronization, and unified intelligent control across devices such as smartphones, tablets, PCs, wearable devices, smart TVs, and smart cars, creating a new ecosystem of "devices as services".

Cross-Industry Players Enter the Operating System Arena

Application vendors and large model vendors are also focusing on operating systems and venturing into hardware terminal layout.

In December 2025, DingTalk launched Agent OS, the world's first working intelligent operating system built for AI agents, as well as DingTalk Real, an enterprise-level AI hardware serving as the intelligent terminal of Agent OS. According to introductions, DingTalk Agent OS is a fully open AI ecosystem, with the core of running and collaborating agents. DingTalk Real requires internal network deployment, can access internal enterprise data, public data, and internal and external networks, and has various DingTalk office agents built in.

According to disclosure from the Sci-Tech Innovation Board Daily, the content to be presented by StepVerse at WAIC is not a single hardware product, but a complete set of terminal layouts for the Agent era, including a newly launched AI terminal brand, a supporting agent operating system, and the brand's first AI agent smartphone.

At the same time, AI glasses are evolving from "hardware terminals" to "system platforms".

On June 26, Rokid released YodaOS, the first AIOS operating system for smart glasses, aiming to enable AI glasses to achieve a full-link intelligent closed loop of "perception-understanding-execution-interaction", rather than just serving as an entry point for a passively responding voice assistant. YodaOS fully opens up full-stack AI development capabilities, allowing developers to carry out development and creation through the platform in a one-stop manner, and jointly build a new ecosystem for AI smart wearables.

The core strategy of YodaOS is "open ecosystem". Zhu Mingming thanked partners including Alipay, WeChat, Amap, and JD.com at the launch event, stating that these partners have opened up underlying APIs instead of simply installing an app on the glasses. Rokid's glasses already support functions such as scanning codes for payment, hailing a taxi, and querying medicine information — tasks that require three or four apps on a smartphone can now be completed with a single voice command.

A research report from Huayuan Securities believes that this represents that AI glasses are beginning to evolve from "hardware terminals" to "system platforms". In the past, the AI glasses industry mainly iterated around hardware functions such as weight, battery life, shooting, and display. In the future, the industry's development focus is expected to shift to operating systems, AI capabilities, and developer ecosystems.

The concurrently launched Rokid AI Assistant 2.0 further confirms this direction: it adds automatic travel itinerary recording and scheduled proactive reminders, enabling AI to move from passive response to proactive service; its self-developed AIUI fully leverages the hardware capabilities of the glasses, replacing the traditional single text output with graphical immersive interaction, transforming AI glasses from a tool attribute to proactive and immersive intelligent interaction.

Huayuan Securities believes that there are still a large number of interesting, high-frequency, and niche unmet demands in the C-end market. In the past, the fulfillment of scenarios such as travel, translation, content creation, and local life services mostly relied on super apps. In the future, VibeCoding and the agent ecosystem may enable rapid development of applications targeting niche scenarios at lower thresholds, allowing more previously unfulfilled long-tail demands to be met; as the ecological carrier for these applications, AI glasses are expected to witness accelerated popularization.

Zhu Mingming stated that smart AI glasses will not replace smartphones, but they will change them. Smartphones may become computing centers, communication centers, and storage centers, turning into devices that people keep around but do not frequently use. When there is a richer and more detailed interactive interface available anytime, anywhere in front of users, people may no longer need to rely on smartphone screens, and interactions will inevitably migrate to glasses on a large scale.

It is reported that Rokid is cooperating with the world's most renowned large model companies to deploy AIOS capabilities on terminals, especially on glasses. In Zhu Mingming's view, the glasses form factor needs to meet the requirements of lightweight, multi-modality, and all-day use, which is exactly the important reason why it is expected to become the best carrier in the AI era.

Zhu Mingming compared the current industry stage to the "BlackBerry era before the birth of the iPhone". The race has not yet started, but the door has been opened. The winners of the BlackBerry era may not necessarily be the winners of the iPhone era. It does not matter who currently holds 60% or 70% of the market share; the key is whether the industry can leap from the BlackBerry era to the iPhone era.

The iPhone 1 era mainly promoted the multi-touch screen, with the most iconic representative being the game "Fruit Ninja". When it comes to the latest interaction technologies, people always think of killer apps that bring unique experiences specific to those technologies. Zhu Mingming believes that the basic platform is ready, and underlying capabilities similar to AIOS are in place, but the ecosystem and user habits have not yet been fully established, and the killer app exclusive to AI glasses has not yet appeared. Once it arrives, it will definitely ignite the market.

Zhu Mingming admitted that the company has made many attempts but has not yet found the explosive breakthrough point. Currently, 70% of its energy is spent on compatibility with the old ecosystem, at least to ensure that existing scenarios are "useful". But he is more looking forward to the birth of a unique technical experience for AI glasses in the next one or two years, so that users will buy glasses specifically for this function. At that time, the "iOS era" of AI glasses will truly arrive.

Currently, the operating system layouts of various manufacturers are still in the early stage of technical verification and ecological construction. Whether it is smartphones, PCs, or glasses, whether they can eventually become the mainstream entry points in the AI era is closely related to the collaborative maturity of model capabilities, hardware form factors, and the developer ecosystem. The answer to this question remains to be tested by the market and time.

This article is from the WeChat public account "Caijing" (ID: mycaijing), written by Liu Fen, and published by 36Kr with authorization.