Twenty years ago, the emergence of smartphones brought an unforgettable "mobile internet moment." As someone who lived through that history, I remember it vividly. The deep integration of mobile communications and the internet gave birth to the mobile internet, and phones suddenly transformed into all-purpose terminals that reshaped lifestyles in the digital age.

People have been waiting for another such "moment." The fusion of phones and generative AI has made that expectation possible. Numerous phone manufacturers at home and abroad have rolled out on-device generative AI services. Companies call it different things 鈥?phone intelligence, phone AI, AI agent phones, and so on 鈥?while consumers prefer the term "AI phone." Their common features include multimodal interaction capabilities, built on coordination between on-device large models and cloud-based large models, with phone AI agents able to understand human intent and automatically operate to complete tasks.

The "AI moment" for smartphones has quietly arrived. Chinese companies are at the forefront. Phone makers and internet companies have partnered to be the first to achieve underlying coordination across chips, large models, and operating systems, and to launch AI agent phones capable of autonomously completing tasks, enabling cross-app automated operations, moving on-device large models from concept to mass production, achieving breakthroughs in multimodal interaction and on-device inference, and leading application innovation in terminal AI.

We applaud AI phones and look forward to the "AI moment" opening a new stage in the integration of AI and the mobile internet, allowing consumers to enjoy the rich fruits of this fusion while also nurturing a new industry ecosystem.

Pursuing Win-Win Outcomes with an Open Attitude

The integration of generative AI into phones turns them from tools into assistants, shifting from "humans operating phones" to "human-machine collaboration." This is a profound change that will affect the entire mobile internet mechanism built on "humans operating phones."

From the consumer's perspective, the most exciting aspect of AI phones is that AI agents automatically operate various apps based on human intent. This can both handle cumbersome operations on behalf of consumers and help make optimal consumption choices. In this process, AI agents replace the device owner as the operating entity, and the entire user behavior chain is planned and executed by AI agents. This means that the entire commercial logic of the mobile internet 鈥?traffic entry points, traffic value, advertising monetization, and so on 鈥?will be redefined. Under the new paradigm, merchants' competition for user attention will give way to competition for agent invocation rights and execution priority.

This is not only a challenge to app service providers but also to the mobile internet ecosystem. It reminds me of 20 years ago, in the early days of mobile internet development, when a large number of app service providers bypassed telecom operators and directly provided various mobile internet application services to phone users over mobile communication networks. Some called the behavior of app service providers OTT (Over The Top), like an over-the-top pass in basketball.

Telecom operators faced the challenge with an open attitude, dismantled the garden walls, and adapted to the internet business model. The open ecosystem brought prosperity to the mobile internet, creating growth and development opportunities for countless app service providers and content providers. At the same time, telecom operators themselves also developed, with data traffic surging rapidly, surpassing voice and SMS to become the main source of telecom operators' business revenue.

AI agents entering the mobile internet represents an extension of the mobile internet's boundaries. Every previous evolution of mobile communications extended existing boundaries, and each extension brought win-win results.

Cooperation Is the Foundation of the Industry Ecosystem

Even if AI phones solve all technical problems, they will struggle to gain traction without ecosystem cooperation. Cross-app operations in particular involve many issues such as data flow, privacy protection, and attribution of responsibility, which must be resolved through cooperation between enterprises.

Currently, there are several different implementation paths for AI agents' cross-app operations, including protocol-based direct connection (such as MCP and A2A), GUI (graphical user interface) methods, and others. Each path holds enormous innovation potential, but no matter which method is used, it will be difficult to truly implement without cooperation between enterprises.

The most undesirable scenario would be: app service providers strengthen risk control mechanisms and close off entry points; agents then try to bypass such risk controls; then app service providers adopt even stricter risk control measures 鈥?and so on in a vicious cycle.

The integration of artificial intelligence and the mobile internet brings not only technological upgrading but also a restructuring of the industry ecosystem's landscape and a new paradigm. This is both an AI journey running through every link of the ecosystem chain and a concerto that breaks down barriers and promotes symbiosis and win-win outcomes across all links.

The reason the mobile internet has achieved today's prosperity is twofold: first, it is a garden without walls, open to all service providers; second, the mobile internet has formed a series of norms that all parties commonly abide by.

I remember when location-based services using satellite positioning were about to be launched on phones, many users worried about security and personal privacy protection. Subsequently, the implementation of a series of norms for phone location services quickly eliminated people's concerns. These norms included the principle of "transparent consent" (a pop-up must appear requesting permission before using location information) and the principle of "minimum privilege" (location information is only permitted during use of that service). It was precisely these norms that helped navigation, ride-hailing, shared bikes, food delivery, and other services flourish.

In the ecosystem of phone AI agents, higher requirements are placed on the application of norms. Existing phone user agreements, privacy protection measures, and authorization systems are all designed for "humans directly operating phones" 鈥?not a single one is prepared for "AI agents operating on behalf of humans." Therefore, relevant standards systems and interconnection protocols need to be established under the premise of safety, reliability, and orderly regulation.

These standards and protocols will involve the system layer (how should the operating system's perception and execution capabilities be opened to agents through standardized interfaces?), the application layer (how should apps open application entry points to agents?), and the data layer (what basic user information and usage status should agents be allowed to know?).

The establishment and implementation of new norms require joint efforts and close coordination from every link in the ecosystem chain 鈥?phone manufacturers, chip developers, operating system providers, large model providers, app service providers, and telecom operators.

Exploring New Business Models

Like every leap in the history of mobile communications evolution, the integration of artificial intelligence and the mobile internet will produce new business forms and models.

A notable feature of the mobile internet business model is the "free model" supported by "backward charging and cross-subsidization," built on the characteristics of low marginal operating costs and obvious economies of scale in internet operations. Traffic is the concentrated embodiment of this scale effect.

Compared with the "free model" of the mobile internet, the business model of AI large models has already shown its differences. The model commonly adopted by large models is: for B-end users, API access and embedded applications charged by Token; for C-end users, a combination of free and subscription. The root of the difference lies in the fact that for the internet, one more user or one more application has a negligible impact on cost, while each task executed by a large model requires corresponding computing power costs.

It is worth noting that AI phone applications will affect the value of traffic. Traffic is a key element in internet commercial logic. Today's traffic is no longer the quantity of data transmission (bits or bytes) expressed in telecom industry terminology. Traffic as referred to by the internet and social media includes visitor numbers, visit counts, and browsing time, as well as the number of likes, shares, follows, comments, and bookmarks. The flow trajectory of traffic forms the basis of big data analysis. Traffic has become a measure of user attention and a bellwether for attracting investment. Traffic generates value, and traffic can even directly determine the survival of developers, producers, and operators.

Traffic generated by Token interactions is being given new meanings. Agents autonomously executing tasks according to goals not only convenience consumers but also give merchants more opportunities to be discovered. The way AI agents select products differs from the ranking logic of the internet. Simulation experiments on AI shopping agents show that "the position of product placement has a weak influence on AI agents' product choices." For AI agents, the effectiveness of methods in the internet traffic economy that grab front positions to attract attention drops sharply, and merchants will place more emphasis on the completeness of product information and degree of demand matching. This may improve the situation under the "attention economy" where "even good wine fears a deep alley," allowing product value to be fully reflected.

In the mobile internet process, merchants can place ads at every node through which traffic passes, and users may be attracted by ads at any time throughout their browsing 鈥?this is precisely where the value of traffic lies. But when AI agents browse the web on behalf of device owners, the role of advertising changes, and the value of traffic changes as well.

Today we are still full of unknowns about new business models. In what way will AI agents reshape traffic value? Can Token billing support a sustainable commercial closed loop? Where is the cost boundary between on-device inference and cloud coordination? How will interests be distributed among app service providers, merchants, and agent developers? These questions have no ready-made answers.

A new model suited to artificial intelligence is about to emerge, and it will be more efficient than the original mobile internet model. Rather than waiting for a mature answer, it is better to actively throw oneself into this exploration. The history of mobile internet development tells us that changes in business models are never designed 鈥?they are gradually honed into shape through practice.

The "AI moment" for smartphones will not arrive automatically. It requires both continued breakthroughs in hardcore technology and the support of updated concepts and industry coordination. Under the mobile internet wave, Chinese companies persisted in technological innovation and business model innovation, achieving world-renowned accomplishments in e-commerce, social media, mobile payments, short videos, ride-hailing, and other fields. Facing the AI tide, we look forward to Chinese companies continuing their innovative momentum and creating new glories on the new AI phone track.

At present, it is still difficult for us to accurately foresee what magnificent waves AI phones will stir up, nor can we depict what entirely new business landscape phone AI agents will bring. But what can be certain is that the new business forms spawned by the deep integration of generative AI and phones have boundless potential, opening up highly imaginative growth space and development opportunities for the information and communications industry.