GuShiio.com Wen| Wang Zhiyuan
When I encountered problems in the past, I was used to using traditional search to solve them.
Now, I rely more on AI assistants: it provides answers quickly, saves the trouble of sifting through information, and is more efficient and more convenient.
Recently, I opened the App Store and found that there have been new changes in the AI assistant market rankings: DeepSeeK ranks first steadily, while the second and third positions repeatedly jump between bean bags and Tencent Yuanbao; it is particularly worth noting that Tencent Yuanbao advertising is everywhere. It can be seen on site B, circle of friends, and QQ.
This makes me wonder: What’s wrong with Tencent? Why did you suddenly enter the tough battle of AI assistants with all your strength?
01
To answer this question, let’s first review what happened over the past year:
Just after the New Year, a private enterprise symposium became popular. The AI circle was all focused on it. The four big shots attending the symposium were:
DeepSeek Sang Wenfeng, Yushu Technology Wang Xingxing, Alibaba Jack Ma and Tencent Ma Huateng. This meeting is not only to cheer up private enterprises, but also to send a clear signal:AI has become the core of the national science and technology strategy.
We don’t know the details after chatting, but the signal is obvious:
From DeepSeek’s reasoning model breakthroughs, to Yushu Technology’s robot innovation, to the participation of Internet giants, AI is being regarded as a key force in promoting industrial upgrading and enhancing national competitiveness.
Therefore, AI is not only a technical arena, but also a vane of the market and policy.
Tencent obviously saw this opportunity. After the meeting, the DeepSeek R1 reasoning model was quickly connected, but the first step did not go smoothly.
There are many feedback problems from users: poor search experience, inability to find public account sources, and even the rhythm of WeChat’s “fast swiping information stream” is incompatible with the “slow chat” model of AI’s in-depth thinking. Tencent decisively adjusted its strategy and diverted traffic from WeChat to Tencent Yuanbao, which was developed by itself.
On February 13, after Yuanbao connected to the DeepSeek-R1 model, user activity increased significantly. Data shows that five major updates were completed within a week, adding functions such as picture understanding and in-depth thinking.
These optimizations not only improve the user experience, but also use the powerful traffic portal of WeChat search to allow Yuanbao’s average daily new downloads to reach 1.6 million, and its ranking on the iOS platform jumped from 200 to second place.
Immediately afterwards, Tencent Yuanbao launched the Thinker (T1) model, which claims to be able to think in depth, specialize in treating complex problems, and search for things in the WeChat ecosystem online. This wave of operations directly squeezed Doubao out.
Why is Tencent so anxious to promote Yuanbao? I think there are two reasons:
On the one hand, Tencent has long been deploying in the AI field and has a deep accumulation of technology. From the early mixed model to the current T1 model, Tencent hopes to prove itself through hard-core technology and put aside those who say that technological innovation is not good.
On the other hand, market pressure is high.
Nowadays, the competition in the AI assistant field is fierce, and DeepSeek has strong reasoning capabilities and is far ahead; ByteDance’s bean bags rely on Short Video platform diversion and advertising bombardment to seize the market.
Ali is not willing to be outdone. Qwen supports the open source market, Quark targets users born in the 2000s, Tongyi Qianwen also deployed AI assistants, and even won smart projects in cooperation with Apple. Baidu’s Wenxin Yiyan is also making steady progress.
In this situation, if Tencent does not quickly find a position to stand firm, it may not even have a place to stay;Therefore, the rise of Tencent Yuanbao is a reflection of technical confidence and a precise card position in market strategy.
02
A good card slot does not mean a good product experience. Let’s talk about the experience of using it.
First, the search capabilities of mobile phones are indeed quite strong.
T1 directly connects into the content pool of WeChat ecosystem and public accounts, capturing all public domain content. Unlike other AI products, which scan information on the Internet everywhere, it focuses on digging for treasures in the WeChat ecosystem.
For example:
I asked it about the QwQ-Max-Preview released by Qwen, and it immediately posted several public account articles, with cases and data, and the organization was very clear. This accuracy is much higher than DeepSeek’s performance based solely on public data, and can basically compete with Kimi’s exploration version.
But the problem comes: AI hallucinations are too serious. It quotes the content of public accounts quickly, but the content may not be reliable.
For example: I asked how much social media advertising will be put in 2024, and it directly threw out a bunch of data, saying that the revenue of China’s Internet advertising market increased by 13.55% to 650.863 billion yuan, and the total advertising and marketing market size is approximately 1.4274 billion yuan.
It sounds pretty bluffing, but on closer inspection, the data sources are complex, and the public account quotes other people’s accounts, which may be made up. I still have to trace the source myself and find authoritative organizations to verify it.
I saw someone on Xiaohongshu saying that Tencent Yuanbao’s main user groups are professionals and students. Professionals value intelligent document processing and AI search functions more, while students prefer intelligent assistant functions.
So, will hallucinations affect specific actual use scenarios? I will.
I want to do business analysis and rely on data to make decisions. Being wrong is no small matter; or writing a report citing numbers, and later it will be exposed that the source is false, and I will lose face. Strong search is an advantage, but there is no verification mechanism, so the speed is inaccurate, and the practicality will be reduced.
Second, the interactive interface on the PC side is too outdated.
Anyone who has used bean buns knows that they have already moved in the direction of the agent, and voice questions can be answered directly, even saving typing. As for Yuan Bao? The traditional “three-piece set for the elderly”: buttons, input boxes, and confirmation buttons. Similar to Kimi’s current state.
However, are these two flaws fatal? Not yet. Search and reasoning are the hard foundation of T1 and can last for a while in the short term; the AI illusion problem can be improved through verification, and the interactive interface can also be upgraded. Tencent has this resource and ability to optimize it.
But the problem is, time waits for no one. Qwen has launched QwQ-Max-Preview, DeepSeek’s new reasoning version is about to be released, and bean bags are even more eager to try. How long can T1 ‘s advantages and market opportunities last?
If the user experience and product power do not improve and rely solely on advertising to destroy traffic, the Retention rate will definitely be low.
03
Writing this, I can’t help but think: What will future intelligent product look like?
Looking at the development of agents in the future, we can’t just focus on technological progress, but also wonder how it will change our usage habits and scenarios. Recently, there is an idea in front-end development called “separation of data and views”. I think this will become a future. A direction for intelligent assistants (Agents).
What is data view separation?
For example: Two users searched for the same data through the same agent, but their needs were different.
One user searched for sales data for a quarter, and he hoped to turn the data into intuitive charts for easy display at meetings; another user searched for the same data, but he needed to organize the data into posters for offline promotion.
This shows that the same data can be presented in different ways, which is the core of “separation of data and view.”
Now the APP interface is basically fixed. The interfaces for Taobao, Weibo, and B sites are all the same; users trigger operations by searching, clicking, etc., and then the backend transmits the data and displays it on the screen in the set format.
Although this model is stable, it lacks flexibility and cannot meet the diverse individual needs of users.
It would be very interesting to apply the idea of “separating data and views” to future smart assistants.Data is the core, and views can be ever-changing. Agents can be presented in different forms for different people and different needs.
For example:
The APP may just be a display interface for the Agent, rather than a separate product. If this continues, fixed apps may be eliminated in the future and replaced by a data-centered and agent-centered interactive model.
By then, users no longer need to open fixed apps and can communicate directly with the Agent naturally. Agent will automatically select the most suitable display method based on your needs.
- You want a report, and the Agent directly generates a PPT
- If you want promotional materials, the Agent directly generates a poster
This kind of flexibility and personalization is simply impossible for traditional apps.Therefore, I think that the future strategic opportunities for C-end products such as Doubao, Kimi, and Tencent Hunyuan may be here.
In layman’s terms, whether it is foreign ChatGPT, XAI, Perplexity, or domestic bean buns, Kimi, although they look different and have different starting points, in essence, everyone is competing for the same territory, and many products are actually wearing “Google Chrome” coat.
In this case, what will the product become?
In the end, different paths may end in the same direction and converge in the same direction: under the narrative of “the next Google”, the goal is not to defeat a certain competitor, but how to surpass Google itself;Therefore, a product may be a task engine, task container, task companion system, or even an intelligent partner.
Think about it, when the time comes, we will also be smart partners. What will the gap be?The key is still the crowd and segmentation scenarios.
I have a strong voice ability. If I want to translate videos or convert voice to text, I will think of it as soon as possible. It is based on this entrance and gives me the various forms I want;Kimi has outstanding ability to process text, like a PDF super assistant, solving problems from analysis to generation in one stop.
Essentially, these are organized around one task, which mobilizes all workflows, and the agent is the best assistant to help you complete the task.
In short, future agent products are partners that dynamically adjust according to user needs. This flexibility and personalization are the core competitiveness of future agents.
04
Who will win? Let’s look at the two key aspects of hardware and software.
Why start from these two aspects?Because both the interactive experience of hardware and the intelligent functions of software are key factors that determine the success or failure of future agent products.
Hardware is the “foundation” that determines the upper limit of device performance and the scalability of software; while software “injects soul” into the hardware, enhancing the value of the device by optimizing algorithms, user experience and functional innovation.
This trend can be seen in Apple’s “Apple Intelligence”.
You can say to your mobile phone: Siri, help me check the recent weather and remind me to bring an umbrella tomorrow. Siri can not only complete queries, but also set reminders for you directly in the calendar. This model is already a more intelligent advanced form.
In contrast, smart assistants such as Tencent’s Kimi and Baidu’s Wenxinyiyan are currently mostly running on Android or Apple phones, which are more like “task containers.”
But in the future, this pattern may change significantly. When you say to your mobile phone: Help me search the price of a certain product on Taobao, the smart assistant directly calls Taobao’s functions to help you complete the search, price comparison and even place an order.
In other words, future agent products may evolve from “task containers” to “task mobilization systems” or even become your “digital partners”; the future opportunities for companies such as Tencent and Baidu may lie in how to combine software and hardware to create smarter and more flexible super assistants.
Judging from market trends, this model of combining software and hardware has begun to emerge.Smart wearable devices (such as headphones, glasses, smart slightly, etc.) are becoming important hardware carriers and even entrances for AI implementation.
If you don’t believe me, look at the group data again:
QuestMobile data shows that in October 2024, the monthly active users of AI native applications have reached 89.76 million, a year-on-year increase of 373.0%, and the growth momentum is very strong.
This means that the development of agents in the future will no longer be limited to the software level, but will achieve a more natural and convenient interactive experience through hardware carriers.
Therefore, only by combining hardware and software can we find out whether Tencent Yuanbao has the opportunity to overtake corners in market competition.
Ali has already obtained a super entry from Apple in hardware; Apple and Google are also exploring the transformation from “software-independent applications” to “intelligent integrated assistants” through agents to integrate user needs.”
What about Tencent?No matter what, getting users here first is the best way.
Multimodal technology is fundamental to this interactive revolution. The rest is a bonus for user experience, so when will major manufacturers come up with an AI assistant product that you can carry around? What I say about recording directly, so I don’t have to look at the screen every day.
Hey. I’ll fantasize first.What do you think?
