Launch events for smart hardware are increasingly turning into showcases for agent capabilities. Smartphones are busy handling complex tasks across applications, glasses are attempting to understand the surrounding environment, and earphones are beginning to proactively capture user needs...
The protagonists of this boom are not limited to traditional consumer electronics companies. Well-known large model vendors such as ByteDance, Alibaba, Tencent, and Jieyue Xingchen have also entered the fray, moving from chat windows on screens to consumers' pockets, ears, and the bridge of their noses. This scene is highly reminiscent of the set-top box craze during the smart TV era and the smart speaker craze during the smart home era. But the question is, can this round succeed? Meanwhile, some software companies that previously took a detour with hardware are not insisting on "proprietary hardware" this time. Instead, they are focusing more on gathering hardware ecosystem partners, making the "intersection of software and hardware" between models and terminals in this round even more interesting.
AI Hardware Ushers in the "Cambrian Explosion" as Model Vendors Enter the Fray
In just the past week, the AI hardware sector has been incredibly busy. Alibaba Qwen launched two AI glasses and one AI earphone, aiming to become "another pair of eyes and ears for people"; Jieyue Xingchen announced that pre-sales for the STEPX Neo agent smartphone have begun, priced at CNY 3,999; the second generation of the Doubao phone announced the latest updates on the SAEP protocol (Screen Automation Execution Protocol), stating that third-party application operation permissions will be gradually opened.
Looking back a bit further, in September, Tencent's WorkBuddy open platform was officially launched, introducing over a hundred ecosystem partners in the first batch. Co-branded hardware covers multiple categories including glasses, recording cards, speakers, and earphones. At the Cloud Intelligence AI Hardware Forum, Baidu stated that more than 1,000 AI hardware companies have connected to Baidu AI Cloud. AI hardware has reached the "Cambrian Explosion" moment of a massive burst in species—this is the judgment of Zhang Wei, Vice President of Baidu AI Cloud and General Manager of the Pan-Technology Business Department.
Why exactly now? A key reason is that the evolution of personal agents is changing how large model vendors envision the consumer market.
Fang Qi, Senior Research Director at Gartner, pointed out that personal agents aim to "democratize access to agent capabilities, enabling anyone to delegate multi-step tasks to agents." This means that large model vendors no longer just need to prove how many questions their models can answer; they must also consider how to make it convenient for ordinary users to delegate daily tasks to AI. Hardware thus takes on new significance. Compared to AI applications that users can switch at any time, hardware has the opportunity to cultivate user habits through continuous use, becoming a more stable traffic entry point with greater possibilities.
Beyond this, compared to agents designed for general tasks, personal agents place more emphasis on understanding individual needs. Given the current state of AI, capabilities such as multimodal interaction, tool invocation, and autonomous execution are no longer novel. Long-term management of user information, continuous comprehension of complex tasks, and the ability to reuse past experiences are becoming increasingly critical.
To truly enable models to understand users, relying solely on information from chat windows is clearly insufficient. Wearable devices such as smartphones, glasses, and earphones can provide more information related to users' daily behaviors and surrounding environments. Taking AI glasses as an example, when integrated with personal agents, the devices can attempt to invoke different services based on the user's environment, personal preferences, and real-time needs. Hardware not only becomes an interactive interface but also a crucial sensing node in the personal intelligent service system.
From "Capable of Completion" to "User-Friendly": Product Positioning Shifts Toward Scenario Differentiation
Looking deeper into this boom, there is another significant shift in this wave of new AI hardware products: an increasing number of products are beginning to leverage their form factor advantages, shifting from being all-encompassing to making strategic trade-offs.
Zhu Mingming, founder of Rokid, pointed out that the biggest pain point in the current industry is the lack of dedicated killer applications. Products that merely port smartphone apps cannot unlock the unique value of head-worn devices in terms of all-day wearability and spatial awareness. This issue is equally instructive for other AI hardware: while personal agents can provide similar comprehension and task-processing capabilities, the devices themselves each have their own strengths.
In the past, when discussing the next generation of smart terminals, the industry habitually asked who would be the "next iPhone," expecting an epoch-making ultimate entry point that encompasses everything. Now, enabling different types of hardware to leverage their strengths and thoroughly penetrate specific scenarios is the only way to better retain users.
For instance, glasses are closest to the human line of sight, making them suitable for identifying environments and providing real-time information through first-person perspective visuals; earphones can complete voice interactions without occupying hands or line of sight, fitting scenarios like reminders and translation; smartphones, with their mature operating systems and application ecosystems, are better suited to handle complex tasks involving multiple applications. Devices such as speakers and rings also have the opportunity to play their respective roles in segmented scenarios like home interaction and health perception.
Even within the same category of devices, differentiation exists. Qwen glasses are equipped with high-precision eye-tracking features; Li Weike is collaborating with Tencent Cloud to explore personal memory capabilities; Rokid is expanding usage scenarios around navigation, teleprompting, and first-person perspective shooting; and some Moonix products, in order to achieve imperceptible wearability, have abandoned display modules and certain low-frequency functions, aiming for an ultimate lightweight design.
This means that not every piece of hardware needs to become a fully functional microcomputer. Whoever can push what they do best to the extreme can also become irreplaceable in a specific scenario.
Software and Hardware Vendors "Team Up" to Seize Entry Points, While Business Models Remain to Be Verified
When large model vendors enter the hardware development arena, the changes are not limited to the products themselves. In the past, software platform providers supplied technology, terminal manufacturers handled production, and internet platforms provided application services, with relatively clear divisions of labor. Today, around a single agent hardware product, the boundaries between model companies, terminal manufacturers, and application and service providers are becoming blurred.
Integrating hardware with models is only the first step; the application ecosystem is the unavoidable hurdle. Agents can plan tasks but cannot independently complete operations such as payments and ticket bookings, still relying on third-party application providers. Jieyue Xingchen announced a partnership with Alipay to integrate over 300 life services into STEPX Neo, while also launching Alipay's AI payment product, Vibe Pay. ByteDance, through the SAEP protocol, allows application developers to decide whether to permit the Doubao phone assistant to execute screen automation operations. Behind these different technical paths lies the issue of how the new agent ecosystem can interconnect with the existing internet landscape, and more profoundly, who will formulate the interface rules and how the boundaries of permissions will be defined.
The deep adjustment of cooperative relationships also poses new challenges to business models. Specific issues such as how to charge for service invocations and how to distribute transaction revenues still await market exploration and verification. Traditional internet business models are highly dependent on app traffic, ad displays, and transaction conversions, with the core logic being traffic monetization. STEPX Neo bundles hardware with a one-year model invocation quota, incorporating continuous AI services into the hardware price; Tencent WorkBuddy connects hardware, applications, and developers through an open platform. The speed of new product releases is already fast enough; next, what deserves more attention is whether actual usage and commercial operations can keep pace.
Author | Chen Cun
Editor | Wu Lilin
Art Editor | Ma Liya Supervisor | Lian Xiaodong