The “super APP” that once had high hopes seems to have reached a crossroads of development. When ecological expansion encounters regulatory restrictions and user preferences shift to lightweight applications, can the former overlords regain lost speed and control? This article provides an in-depth analysis of the development logic and future dilemmas of super apps, and explores how much time is left for them.
On an ordinary Tuesday afternoon, Xiao Wang, a programmer in Shanghai, said to the mobile phone screen:
“This weekend, find a destination in Jiangsu, Zhejiang and Shanghai that starts from Shanghai, has a per capita budget of less than 2,000, and can see the mountains and rivers, arrange a two-day and one-night itinerary, book high-speed rail tickets and high-rated homestays, and recommend a few local must-eat restaurants.”
Without opening 12306, Ctrip, and Dianping, there is no need to manually switch back and forth between countless apps, copy and paste. A few minutes later, he received a push on his phone: The itinerary has been planned – the high-speed train to Anji at 9:15 a.m. on Saturday can book a “XX B&B” with a detailed guide including three local restaurants. He only needs to click “Confirm” at each key link.
At this moment, we didn’t realize that the super apps that ruled our lives would face a new revolution.
In 2024, if the large model is an arms race in the technology circle, then AI Agent is the Normandy that all forces want to take.
A divided battlefield
AI Agent has allowed China’s Internet giants to show a long-lost collective restlessness mixed with excitement and fear.
At the table, players have different expressions and intriguing bets.
The most devout believer is Baidu. Robin Li believes that 2025 may be the first year of the outbreak of AI agents. Large reasoning models have emerged with amazing deep thinking capabilities, which will promote the implementation of an important application direction of artificial intelligence, that is, “AI agents”.
In recent years, Baidu has almost been “smashing pots and selling iron” all in. From the iteration of the Wenxin model, to the launch of Qianfan AppBuilder, a platform where everyone can “drag and drop” to generate an agent, to landing applications and agents such as Wenxin Kuaixia, Xinxiang, and Baidu Library, Baidu’s strategic intention is very clear: they want to seize the once-in-a-lifetime opportunity of AI Agent and return to the center of the Internet after the era of PC search and mobile APP. This is Baidu’s hard work, and it is also a battle that cannot be lost.
To achieve these three challenges, product managers will only continue to appreciate
Good product managers are very scarce, and product managers who understand users, business, and data are still in demand when they go out of the Internet. On the contrary, if you only do simple communication, inefficient execution, and shallow thinking, I am afraid that you will not be able to go through the torrent of the next 3-5 years.
View details >
The most entangled landlord is Tencent. Sitting on WeChat, the most fertile land in the mobile Internet era, Tencent’s attitude towards AI and AI Agent is like a landlord who is worried that “new agricultural tools” will dig up his ancestral graves. It is not only eager to use the “divine hoe” of Agent to make the land more productive, but also afraid that it will break the existing ecological balance.
Therefore, Tencent’s actions are particularly restrained. The Hunyuan model is the base, but its application is more “empowering” than “subversion”. Help you summarize summaries in WeChat Reading, write minutes in Tencent meetings, and collect information in QQ Browser. These agents are more like “AI patches” for Tencent Empire’s various functional plug-ins, making the moat deeper and the walls higher.
For Tencent, Agent does not seem to be a clarion call for offense, but a fortress for defense.
The most pragmatic shopkeeper is Alibaba. After a series of organizational adjustments, Alibaba has become very realistic. It no longer just looks at the sea of stars, but bows its head to calculate every business. Agents are dismantled into “guys” who can improve efficiency and increase revenue here in Alibaba.
On DingTalk, it transforms into an “AI assistant” to help you write weekly reports and schedules, with the goal of improving the office efficiency of tens of millions of enterprises; On Taobao, it is called “Taobao Ask”, which helps you find goods, compare prices, and provide dressing suggestions. For Alibaba, the value proposition of Agent is extremely simple: Can you help me make more money? Can you help me save some trouble?
This is a kind of pragmatism, and Agent is not a unified belief of the group, but a “Swiss army knife” for each business unit to solve specific problems.
The most mysterious assassin is ByteDance. Byte is the most low-key, but probably the deadliest player in the game. No one can ignore the potential of the byte. It has two huge “information feeders” of Toutiao and Douyin, as well as the world’s most understanding recommendation algorithm. The essence of Agent is “service to find people”, and Byte’s entire business empire is based on the logic of “content to find people”. This genetic fit is unmatched.
Imagine an AI Agent that is deeply integrated with Douyin life services, and when you swipe a skiing video, it can immediately help you plan your entire itinerary to Changbai Mountain. This seamless transition from “planting grass” to “pulling weeds” is commercially explosive enough to make any opponent shudder.
While everyone is talking about model parameters, the real battlefield has quietly shifted to Agent. It is no longer a technical concept, but a path that large manufacturers can indeed see to generate value, whether it is the value within the enterprise, or the commercial value to consumers and the industry. **
Game of Thrones
Why can something that seems like a “little assistant” stir the nerves of the giants? Because its emergence is not a simple technological iteration, but a subversion of the business paradigm of the past decade.
To understand the subversive nature of Agent, we must go back to the coordinates of history.
First, Agent is trying to end the “era of APP silos”.
The past ten years have been the decade of “APP is king”. Our phones are occupied by colorful icons, and each app is a data chimney, which together builds a wall. If you want to take a taxi, you have to turn on Didi; If you want to eat, you have to open Meituan; If you want to shop, you have to open Taobao.
The essence of this model is “people looking for services”. Users need to actively and consciously migrate between different information islands, which creates the core contradiction in the second half of the Internet –Traffic anxiety。 Large manufacturers are sparing no effort to launch subsidy wars and information flow wars locally, essentially to compete for the “first look” of which APP you open. Tens of billions of marketing expenses are spent on this link every year.
The emergence of Agent is a salary draw from the bottom of this model. Its core logic is “service to find people”. Just like Xiao Wang’s example at the beginning of the article, he doesn’t need to care whether the service is provided by Ctrip, Fliggy or Tongcheng, he only cares about the result.What the agent has to do is to dismantle the enclosure of these gardens, and theoretically, it can return the power from the operator of the APP to the user.
This undoubtedly poses a huge potential threat to business models that rely on “traffic distribution” and “advertising bidding” to make a living. When users no longer need to “visit” the APP, what is the value of the APP’s advertising space? When the agent can directly call the underlying service, where will the “traffic broker” in the middle go?
Secondly, this is a game of thrones about “entrance”, and history is strikingly similar.
Looking back at the history of power change on the Internet, it is a history of “entrance” competition.
In the PC era, the entrance is a search box (Baidu, Google), and “information” is the core resource; In the mobile era, the entrance is a super APP (WeChat, Alipay), and “social/transactional relationship” is the core resource. Every change in entrance will lead to the rise of one group of giants and the decline of another.
Now, Agent is becoming the “super entrance” of the next era. This entrance may not be just a box, nor just an APP, but a “dialogue interface” – an “all-round butler” that can understand human language and mobilize thousands of services, of course, its form may also be a box and an APP.But it is worth noting that this “all-round butler” itself may also become a new and more centralized giant.
And it now seems more likely to exist in a certain terminal, such as a mobile phone, such as a computer, or some other device, this imaginary scene, as early as a few years ago, smart speakers were reflected in the smart speaker, through the smart speaker to place an order on the e-commerce platform, of course, in the end did not become a real all-round butler.
Finally, we must recognize that the bottleneck of agents is far more than “IQ”, but also “emotional intelligence”, “hands-on ability” and “reliability”.
Many people think that the strength of the agent depends entirely on the intelligence of the large model behind it. This is a misconception. A strong Agent requires at least four pillars:
- Planning and reasoning skills (brain): This is the core capability of large models, which is responsible for understanding tasks and dismantling steps. At present, this pillar is relatively the thickest.
- Tool Call Capabilities (Both Hands): This is the ability of the agent to interact with the physical and digital worlds, such as calling an app’s API to book tickets or controlling a smart home.
- Long-term memory ability (memory): Remember user preferences, habits, and historical conversations to provide a truly personalized service.
- Trust and Safety (Character): Whether users are willing to hand over their data, privacy, and even payment authorization to an AI. The establishment of this trust is far longer and more fragile than the implementation of technology.
Compare the big model to an intern who has just graduated and has an IQ of 180, he is very smart, but to become a capable assistant, you have to give him a mobile phone that can make calls (tool calls), a notebook that can record your preferences (long-term memory), and authorization to allow him to help you do things (permissions).And you have to believe in your heart that he will not mess up or leak secrets (trust and safety).
At present, the biggest bottleneck of Agent is precisely in “hands” and “character”. “Tool calling” requires a large, stable, and standardized API ecosystem, which is extremely cumbersome engineering and business docking work. The establishment of “trust” faces the triple challenge of technology, privacy security and user psychology.
The model is an engine, but without a reliable drivetrain and wheels, no matter how strong the engine is, it can only roar in place. This transmission system is the ecology of “tool calling”. At this point, WeChat, which has a massive mini program ecosystem, and Alibaba and Meituan, which have mature merchant APIs, have innate “engineering advantages”.
Who will lift the table?
While the coachman was still complacent about one more whip, Ford’s assembly line was already in motion. The wheel of history rolls forward, never stopping for a moment because of whose size is huge.
I have three bold judgments about the future of AI Agents:
First, the twilight of “super APP” has begun, and the prototype of “super agent” will rise on their ecological soil. In the next 5 to 10 years, when we turn on our mobile phones, the first thing we look for may no longer be an APP, but the only agent entrance. Our experience of jumping between different apps today will gradually become history.But this will not be a complete destruction, but more like a transfer of power – from the shell of the APP to the core service capabilities provided by the APP.
Second, the “Agent Store” will complement or even reconstruct the “App Store” and become the new ecological core. Just like the Apple App Store today, there will be an Agent skill store in the future. Users can “plug and plug” various professional sub-agents (such as a “tax return agent” and a “fitness coach agent”) for their main agent according to their own needs, so as to build a “super individual” with thousands of faces. Developers will also shift from “developing apps” to “developing agent skills”.
Third, the ultimate winner is not necessarily the strongest model, but must be the most widely ecological and trusted by users. The competition of agents looks at model technology in the short term, tool ecology in the medium term, and user trust in the long term. Whoever can be the first to solve the last mile of tool call and build an unbreakable flywheel of trust will win the endgame.
From this point of view, the outcome of this war is far from certain.
But the conclusion is clear: we are on the eve of a great paradigm shift. Those Internet giants who once called for wind and rain are now standing at the crossroads of fate.
They may take the initiative to break the old world they have built with their own hands and embrace a new future full of uncertainty; Or stick to the walled garden in front of you, waiting to be swept away by the torrent of the times.
This is not an upgrade, this is a revolution. At the power table of the big factory, the Agent is not a new card, but the one who has the potential to turn the table upside down.
