Zhu Xiaohu: "The Next Byte" has been established this year

On September 11, Zhidongxi reported that today, at the opening ceremony of the 2025 Inclusion·Bund Conference, Ant Group CEO Han Xinyi, Xiaomi Group Vice President of Mobile Phones and General Manager of Wearable Division Zhang Lei, and Managing Partner of ZhenFund Zhu Xiaohu held a roundtable dialogue on the theme "Breaking the Deadlock and Reconstruction: The Implementation of Large Models in Industry," discussing their views on super entry points, AI business models, and future opportunities for Chinese enterprises.

Han Xinyi believes that AI is consuming software, and the medical scenarios that Ant Group focuses on are essential needs for users and are of medium to high frequency, so they are not worried about commercialization but are focusing on three major technical challenges: high-quality data, suppressing hallucinations, and medical ethics.

Zhang Lei mentioned that Agents have already replaced Apps in single, clear domains, but complex and highly logical tracks cannot be replaced in the short term. Unlike AI phones and AI PCs, AI glasses have reshaped interaction logic, enabling a shift from passive to active decision-making. He revealed that the frequency of users waking up Xiao Ai to operate on glasses is 6-7 times that on mobile phones.

Zhu Xiaohu believes that AI and software will coexist, and the dirty and tiring work that large companies are unwilling to do presents opportunities for Agent companies. From an investor's perspective, they should first avoid collaborative software, as this market will be small, while more easily commercializable technologies are mature ones, such as last year's meeting minutes and this year's Voice Agents. Chinese entrepreneurs are suitable for B2C, and there will definitely be an explosion of AI applications next year, providing great opportunities for Chinese entrepreneurs.

Finally, the three leaders shared their favored directions for current AI entrepreneurs. Han Xinyi is optimistic about the service industry in specialized fields, but he also mentioned that this can be challenging for young entrepreneurs starting from scratch, requiring certain experience and resource investment; Zhang Lei is optimistic about translation, assistant, and learning service scenarios; Zhu Xiaohu believes that each wave of cycles is quite similar, with last year focusing on hardware, this year on AI infrastructure, and next year will definitely see a big explosion in applications, with the next ByteDance, Kuaishou, and Xiaohongshu likely already established this year.

Here are the highlights from the roundtable dialogue:

Low-code Software Has Been Replaced by AI; Complex Interaction Scenarios May Create Super Apps

a16z General Partner Jennifer Li once stated, "AI is consuming software."

Han Xinyi agrees with this view, believing that software solves deterministic problems in a relatively certain way, which is also what large models excel at. Additionally, there are already software-generated Agents solving tasks, and this trend is very rapid.

Zhang Lei and Zhu Xiaohu have slightly different views, believing that AI and software will coexist.

Zhang Lei mentioned that he currently does not see the possibility of large models eating all software; rather, large models can enhance efficiency in many scenarios without replacing all software. In the later stages, large models and software will undergo a process of integration and coexistence Simple low-code and no-code software will be replaced by AI. Many low-code companies that had high valuations during the bubble period are now basically gone. However, for complex processes and software with strong logic, it is unrealistic for AI to replace them under the Transformer architecture. Zhu Xiaohu's basis for this conclusion is that as long as the Transformer architecture has even 1% of hallucination, AI cannot replace complex process management software; instead, there will be more coexistence and symbiosis.

▲ Zhu Xiaohu, Managing Partner of Jinsha River Ventures

In Han Xinyi's view, there may be no Apps in the future; there will be one or several super entry points, but the exact form is still uncertain.

Regarding the relationship between Agent and App, Zhang Lei mentioned that Agents have already replaced Apps in single, clear domains, but complex interactive scenarios are very difficult in the short term. In the long run, it may lead to the creation of super Apps, but the capabilities required for this are very comprehensive, and there is still a long way to go.

Zhu Xiaohu believes that the emergence of super entry points is inevitable, and the entry points will become increasingly concentrated. This trend is already very clear; Google phones can now directly input voice and provide feedback, combined with multimodal forms for input.

The opportunities for Agent vendors can be compared to Uber, DoorDash, and Airbnb during the American mobile internet era. These three companies operate in areas that large companies are unwilling to engage in, which is also an opportunity for AI.

User Retention is the Standard for Growth Potential of Agent Companies; AI Glasses May Become a New Entry Point

As an investor, Zhu Xiaohu believes that in the software industry, the demand for editing and collaboration software will significantly decrease. Projects that previously required hundreds of people to collaborate may now only need ten due to AI, which will greatly reduce the requirements for collaborative software and have a significant impact on the market. Just as the previously popular Figma has now seen a decline in interest.

From another perspective, software is not being replaced; rather, the number of users is decreasing. If the user base decreases by 10% from the previous volume, the impact is enormous. Therefore, from an investment perspective, they will definitely avoid collaborative software, as this market will be much smaller in the future.

For companies like Figma, Zhu Xiaohu mentioned that their business focus must change from previous collaboration to now focusing on delivering better results. However, even if the generative capabilities are very strong, there is still a 5% to 10% gap in the final commercial stage. How to edit this part and quickly achieve the quality of commercial delivery is where startups can create value.

Regarding the criteria for judging the growth potential of companies in the AI and Agent era, Zhu Xiaohu believes that user retention remains the key indicator, from the PC internet, mobile internet to the AI era.

Currently, many AI companies do not have user retention. This is because when products are first released, users are eager to try them, but after trying, they may not necessarily pay. In the mobile internet era, recalling users requires more than ten times the cost, which is almost impossible Therefore, whether retention is good or not can prove whether the company has development potential.

He mentioned that, in fact, many AI companies are now afraid to talk about retention in subsequent financing stages, only mentioning ARR (Annual Recurring Revenue).

ANT GROUP focuses on medical scenarios, and regarding user retention considerations, Han Xinyi believes this is a user necessity and a medium to high-frequency demand. Therefore, when AI in healthcare reaches its peak, there may be no need to focus on user retention anymore. The most critical point is whether the application can truly understand users and provide suggestions like a professional physician, which is also the ultimate goal of AI in healthcare.

At the same time, general large models cannot replace such vertical scenarios because strong professional capabilities in medical scenarios create a moat, understanding users better the more they are used.

However, AI in healthcare will not replace doctors for a considerable time; it is more like an assistant to doctors. It cannot have a clear understanding of the human body like a person, but AI can help doctors better, more comprehensively, and systematically understand patients, transforming specialists into general practitioners.

One major scenario where AI is difficult to replace humans is in treatments requiring physical contact, so Han Xinyi mentioned that the only way out is human-machine collaboration. Nowadays, the role of AI is to help doctors manage their time better, allowing them to focus more on research and complex cases, providing excellent assistance to more grassroots doctors.

▲ANT GROUP CEO Han Xinyi

In the hardware sector, Zhang Lei discussed thoughts on terminals that will truly become super entry points in the future.

He believes that with the support of AI, all hardware should be rethought. For future entry opportunities, it still needs to return to the capabilities of AI hardware. Only by making users unable to live without the device can the terminal have the opportunity to switch to an entry point.

In the long term, AI glasses have the potential to become a new entry point.

Currently, devices like AI phones and AI PCs focus on enhancing user experience without changing the basic interaction logic, remaining passive and requiring user touch decisions. However, glasses shift from passive to active, fully presenting what users think, see, and feel, and then making judgments and executing based on that. In Zhang Lei's view, this could be a carrier for the future change in interaction paradigms.

Taking Xiaomi AI glasses as an example, the current high-frequency usage scenarios for users include making calls, listening to music, instant snapshots, and image-based Q&A. He added that the frequency of waking up Xiao Ai on the glasses is 6-7 times that of the mobile phone.

Currently, AI glasses are still in the early stages of industrial development; they are merely an extension of mobile phones, and there is still a long way to go in improving display, interaction, and other capabilities.

Mature Technology Suitable for Commercialization; Medical Scenarios Face Data, Hallucination, and Ethical Challenges

Which industries have commercialization potential? Zhu Xiaohu believes: "The ones truly suitable for commercialization are all Boring Technology." Last year, the most successful commercialization was the ability to create minutes, such as Plaud's AI minute-taking hardware, DingTalk's software, and tools for recording conversations between doctors and patients. This technology is not difficult; pursuing commercialization requires relatively stable technology.

This year, he feels that Voice Agent has already been widely commercialized, with applications in customer service, sales, AI toys, and other extensions of this scenario.

In specific medical scenarios, Han Xinyi summarized three major challenges for the implementation of large models: high-quality data, hallucination suppression, and medical ethics.

Regarding data and hallucinations, Ant Group has already found relatively good methods. In June of this year, Ant launched the AI health application AQ, which has shown good results in medical health consultations.

First, accumulating high-quality data requires significant investment; a single data point may cost hundreds of dollars or even more, and it requires data at the level of chief physicians to train effectively. Secondly, hallucinations pose a challenge in suppressing them without reducing model capability. The third point is medical ethics; Ant is establishing a medical ethics advisory committee to involve top experts from the medical field for improvement.

Han Xinyi added that they are not worried about commercialization. First, the medical health industry is extremely large. Secondly, the business model is relatively clear, not relying on advertising or transactions; it is inherently about healthcare, pharmaceuticals, and insurance, which are similar across different eras. The third point is that in the next 1-2 years, they should not focus on commercialization, as there is a market and a path, making further exploration unnecessary. Instead, they should concentrate on the three technical challenges mentioned earlier.

Next Year AI Applications Will Explode, Chinese Entrepreneurs Have Great Opportunities

In the global competitive landscape of AI, Zhang Lei believes that China's advantages lie in the cost and efficiency of the supply chain and the AI ecosystem advantages.

For Chinese companies to effectively leverage the cost and efficiency advantages of the supply chain, they must return to product strength. Companies need to quickly identify opportunities, rapidly layout, diversify product lines, iterate SKUs quickly, capture all markets, and achieve sales growth, all of which require corresponding decision-making.

Additionally, under the advantage of the AI ecosystem, finding truly needed user scenarios to implement, enhance user experience, create stickiness, and then improve product strength are all considerations that companies need to focus on based on these advantages.

▲ Zhang Lei, Vice President of Xiaomi Group's Mobile Division and General Manager of the Wearable Division

Regarding the advantages of Chinese companies, Zhu Xiaohu reviewed the AI companies that Jinsha River Ventures has invested in, noting that among the top 10, 6 are B2B companies founded by foreigners, while the remaining 4 are Chinese companies focused on B2C.

He believes that Chinese entrepreneurs are well-suited for B2C, and there will definitely be an explosion of AI applications next year, presenting great opportunities for Chinese entrepreneurs. This is because the B2B track requires local sales teams once it scales up Even though the capabilities of general large models are similar at present, Chinese entrepreneurs excel at building differentiated user experiences outside of AI. These product differentiators have been widely proven in past cycles; human needs do not change, and now AI has enhanced user experiences without altering those needs.

Conclusion: Agent Becomes the Core Trend for AI Implementation! Chinese Entrepreneurs Focus on Differentiated Strategies

The development of the large model industry is rapidly changing. Since the beginning of this year, the enhancement of individual Agent capabilities and the collaboration of multiple Agents for widespread application have become a significant trend in AI implementation. Behind this, Chinese entrepreneurs are striving to find more differentiated experiences and innovative approaches for users on top of Agents, even when the underlying models have similar performance.

At the same time, as mentioned by the three AI leaders above, to find effective business implementation models, it is essential not only to identify user demand scenarios but also to have mature technology and product capabilities to ensure user retention and enhance product stickiness. Chinese AI startups, with advantages in supply chain costs, efficiency, and AI ecosystem, are transforming these advantages into product upgrades and driving global expansion, creating super applications for the AI era.

Author: Cheng Qian, Source: Zhidx, Original Title: "Zhu Xiaohu: 'The Next Byte' has been established this year"

Risk Warning and Disclaimer

The market has risks, and investment requires caution. This article does not constitute personal investment advice and does not take into account the specific investment goals, financial conditions, or needs of individual users. Users should consider whether any opinions, views, or conclusions in this article align with their specific circumstances. Investment based on this is at one's own risk