Zhang Peng in conversation with Zhu Xiaohu, Chu Ruisong, and Fu Sheng: In the era of Agentic AI, don't do everything on your own in silence.

As the focus of AI discussions shifts from the amazement of "what it can do" to the profound question of "how should I live," an era of technology - driven narratives is quietly coming to an end, and a new cycle dominated by commercial realities has begun.

In the past year, we have witnessed the rapid iteration of models and the collective anxiety of startups after the press conferences of tech giants. The growth mantras of the mobile - Internet era - network effects, scale effects, and data flywheels - seem to have been weakened or even become ineffective. In a new battlefield where the iteration speed is three times faster, the old "military strategies" can no longer guide the battle. Entrepreneurs and transforming enterprises are all facing fundamental business questions: Who are my customers? How do I charge? Where are my barriers? How long can I survive?

For this reason, Zhang Peng, the founder of GeekPark, had an in - depth conversation with three key figures at the center of the storm. They are: Zhu Xiaohu, the managing partner of GSR Ventures, who holds capital and has a calm and pragmatic perspective; Fu Sheng, the chairman and CEO of Cheetah Mobile, who is leading a company with a long history through a "genetic transformation" on the front line; and Chu Ruisong, the vice - president of Amazon globally and the president of Amazon Web Services Greater China, who, as a technology platform, has insights into the ups and downs of numerous enterprises.

The perspectives of investors, entrepreneurs, and technology platforms are like a prism, reflecting the most real anxieties, struggles, and opportunities of practitioners in the era of Agentic AI. In this conversation, you can see:

• New business paradigm: Deliver results, not just tools. In the era of Agentic AI, the core of the business model is to charge based on results, which may be a new breakthrough point for Chinese software enterprises.

• New survival philosophy: Barriers are dynamic. Traditional network and scale effects are weakening, and data barriers are not that high. The new barriers are either to "grow stealthily" in areas that giants don't care about or to rely on extreme speed and execution.

• New organizational change: Both startups and enterprise transformations need to "ride on the wave." Real transformation starts with ideological and organizational changes. Learn to fully utilize platforms and tools to amplify efficiency. Don't do everything by yourself; don't reinvent the wheel.

• New growth engine: Going global is a must - option. The domestic market is the best training ground, but to make money, you still have to go to the global market. You can even consider going global from day one.

The following is the full text of the conversation, edited and organized by GeekPark.

01

A new breakthrough point: Using Agentic AI

Deliver results and charge based on results

Zhang Peng: Let me start by asking Mr. Zhu Xiaohu. Investors are an important driving force. Apart from the technological changes we've seen, the promotion of capital and the preferences of the capital market are also crucial. You yourself highly recognize the general trend of Agentic AI. But in the direction of this general trend, what specific paths have the most opportunities? What kind of tracks and paths should today's startups or growing enterprises choose to use this technology to create greater value?

Zhu Xiaohu: Agentic AI has indeed spread very rapidly this year. As mentioned in Amazon's previous speech, the most important thing is to directly deliver results using Agentic AI. In the past four or five years, China's software SaaS industry has experienced a rapid rise and then a rapid decline. Both entrepreneurs and investors have found that it's really difficult to charge for software in China. The difference from the US lies in the difficulty of charging.

Now, when we use Agentic AI to deliver results, we are actually not charging for software anymore but for results. This may be a very important point that can bring a new breakthrough to Chinese software enterprises. Therefore, we prefer to see business models that focus on vertical and segmented industries and charge customers based on results. This may be a way for Chinese entrepreneurs and investors to explore new possibilities. Additionally, we also see many cases on the consumer side where AI is used to directly deliver user experiences and results, and the revenue is growing very rapidly.

Zhang Peng: Are there any specific tracks that you are currently optimistic about? Or, having seen many startups, what types of them do you generally look favorably upon, and which types do you tend to avoid?

Zhu Xiaohu: Actually, whether in China or the US today, we only look at one number - the revenue growth rate. For early - stage companies, especially from the angel round to the A - round, a 5 - to 10 - fold revenue growth is a baseline that investors are more likely to accept.

Zhang Peng: So, there still needs to be a very solid and practical value - closed loop.

Zhu Xiaohu: Yes. Generally speaking, we say that you need to create 10 times the value. Only when you create more than 10 times the value for your customers will they adopt your product or service very quickly, ensuring a 5 - to 10 - fold annual growth rate. Why did software rise so rapidly in US enterprises back then? It was growing at a rate of two or three times a year and quickly reached an ARR of $100 million. In today's AI era, the growth rate is even faster than that of software enterprises back then, at least 5 to 10 times, especially in the early stage.

02

Stay away from tech giants or stand on the shoulders of giants?

Zhang Peng: When tech giants hold press conferences, startup teams are very distressed because when their technology is upgraded, many of the efforts made by startups seem to be affected. What's your view on this? Is this a normal situation? How can this problem be avoided?

Zhu Xiaohu: This is actually the same as in the mobile - Internet era. In the early days of the mobile - Internet era, we saw many similar situations. For example, when Android and iOS first emerged, there were many functional apps, such as flashlights, perpetual calendars, email apps, and browsers. Many startups were involved in these areas and were very popular at first, but they have all disappeared later. That is to say, tool - based enterprises have opportunities in the early stage of any cycle because iOS or the underlying models don't have the time to develop these tools. But once they have the time and notice which tool has a large user base, they will develop it themselves, and the startups will be wiped out. In today's AI era, the iteration speed is more than three times that of the mobile - Internet era, so it's even faster. When a tech giant like OpenAI holds a press conference, and then another giant holds one, many tool - based enterprises will be in a difficult situation. Therefore, we say that we should focus on segmented and vertical industries, directly deliver results, and stay away from tech giants. In the mobile - Internet era, we said to "stay one street away from tech giants." Now, in the AI era, one street is not enough; we need to stay three streets away.

Zhang Peng: In the past, everyone wanted to compete for the central market. But today, it means that you'd better not compete with tech giants for the central market and should find a "Shangri - La" to develop.

Chu Ruisong: It's true that many startups may feel anxious when some tech giants hold press conferences. However, I think AWS is different. When Amazon Web Services holds a press conference, entrepreneurs are very excited. Why? When we hold a press conference, we launch new services that can effectively help entrepreneurs accelerate innovation.

Whether it's Bedrock or AgentCore, AWS has always positioned itself as "we do the undifferentiated heavy lifting for customers." We take care of those necessary but non - differentiating tasks for our entrepreneur partners, software enterprises, or various companies, so that they can focus on creating their own value. Therefore, when Amazon holds a press conference, these startups or enterprises will be very excited.

Zhang Peng: I understand. There is a tech giant that is different. Amazon Web Services is still a friendly giant to entrepreneurs.

03

The eve of the Agentic AI explosion has long passed;

It's now 5 a.m.

Zhang Peng: I'd like to follow up. Since you've met many startups and growing companies, Agentic AI has brought many changes and impacts on the development paradigm of the entire software business. How would you summarize these?

Chu Ruisong: In June this year, Amazon Web Services held a summit in Shanghai. At that time, I said that we were on the eve of the Agentic AI explosion. Now, some people joke with me that the eve has long passed, and it's already 5 a.m. Why? First of all, all the elements for the explosion of Agentic AI are now in place. Firstly, large - language models now have human - like thinking abilities. They can independently judge what is the most appropriate next step to complete a task.

Secondly, we have protocols like MCP, which enable Agents in Agentic AI driven by large - language models to easily access existing data and call the APIs of existing applications, allowing them to perform many tasks.

Thirdly, the toolchains for developing Agentic AI applications, whether it's the Strands Agents SDK released by Amazon or toolchains like Bedrock and AgentCore, are becoming more and more perfect. Developing Agentic AI applications will be very convenient. So, it's now a certain time in the early morning for the explosion of Agentic AI.

What impacts does Agentic AI have on the software industry? I think there are several aspects.

Firstly, it will change the way software enterprises deliver value to customers. Previously, software enterprises developed a tool and gave it to customers. Customers had to use this tool well before they could complete what they wanted to do.

Now, in the era of Agentic AI, regardless of the paradigm used, whether it's Agent Embedded, Agent as a Service, or Agent as a Facade, more and more software enterprises are not only delivering tools to customers but also shifting towards delivering value.

For example, if you are a software company providing call - center services, previously you provided tools for call - center staff to use. Now, in many cases, your call center can understand the customer's business and automatically handle many customer - support requirements of enterprises. Only the more complex cases will be transferred to human customer service. Or, if you have a legal Agent Service, the results generated by the Agent are legal documents that can be directly reviewed by lawyers, rather than just helping lawyers draft documents or search for previous cases. Therefore, delivering results is very important.

Secondly, Agentic AI will change the entire software - development production process because the capabilities of AI Agents have become an important production factor in software development. Previously, software enterprises didn't have many heavy assets. What they had were: first, people, including people in different roles, from product managers to architects, development engineers, test engineers, operation and maintenance engineers, as well as marketing, sales, and customer - service personnel; second, knowledge, whether it's knowledge of horizontal CRM or ERP in a certain field or knowledge of a vertical industry, such as the process industry or the medical industry. These were the previous production factors. But in the era of Agentic AI, AI Agents have become a new production factor. AI Agents have knowledge, and more importantly, they can directly participate in the entire software - development production process as an equal partner.

In the AI - DLC paradigm (AI - Driven Development Life Cycle) defined by AWS, where humans and AI collaborate on innovation, AI is actually the driver. It will drive requirements decomposition, architecture design, coding implementation, testing, deployment, operation and maintenance, and the resolution of subsequent service issues. What's the role of humans? Firstly, humans are the ones who clearly put forward business requirements. Secondly, humans need to conduct reviews, make decisions, and judgments throughout the process. Therefore, the entire process is different from the well - known Agile Software Development Life Cycle. AI is an equal partner, the driver, and executor of the entire AI - DLC process, while humans are the ones who put forward requirements and make decisions and judgments.

Zhang Peng: This is indeed an important change. This may also be the first time in history that a technology, whether we use it ourselves or deliver it to customers, is no longer just a tool. It should be an equal partner within our organization and a productive force that can truly deliver results to customers, rather than just a production tool.

Chu Ruisong: If you use the AI - DLC process well, the improvement in overall production efficiency is not just 15% or 30% but three, four, or five times.

Zhang Peng: It's not a linear improvement but an exponential one.

04

Transform people first, then transform products

Zhang Peng: Let me ask Fu Sheng. You may have the most resonance with everyone here today. Everyone has existing businesses and is actively exploring in the AI era. I know that in your exploration process, you first promoted organizational change, which impressed me deeply. Why did you start with organizational change within your company instead of training your own models?

Fu Sheng: I trained models seven or eight years ago, which were voice models at that time. Later, I realized that the investment was huge, the returns were difficult to achieve, and the competition was extremely fierce. I recalled the other day that Cheetah Mobile has been in business for 15 years and has been listed for 10 years. It's a company with a history. However, having a history is not a good thing in this era. History is a burden, and the biggest burden is actually people.

I remember when Mr. Lei Jun was starting Xiaomi, he talked to me. He said that when he was trying to transform Kingsoft into an Internet - based company through the transformation of Joyo, he found it extremely difficult to change thousands of people who were used to developing traditional software to do Internet - related work. He gave an example. One Saturday, he found that their e - commerce website was down. He called an employee and said that the website was down and asked the employee to check it. The employee said, "Mr. Lei, I know. I'll handle it when I come to work on Monday." He said, "Do you know how helpless I felt?" In the view of those doing traditional software, "I'll handle it on Monday" is already good, but the Internet requires 7×24 - hour service. Therefore, I want to say that the biggest burden of people is actually the experience - based thinking. I've always believed that once a tool becomes powerful enough, it will change people's thinking. If you don't change your thinking, it's difficult to use the tool well.

Therefore, our company's AI transformation is divided into three layers: the first layer is ideological change, the second layer is organizational change, and the third layer is product change. It's not the other way around. Firstly, for ideological change, we took several measures. The first thing was that I