I received an investment from Sequoia even before my company was registered.

In this issue of Blue hour, we've invited an AI application entrepreneur. When he only had a demo and the company hadn't been registered yet, he received investment term sheets from Sequoia Capital and Huachuang Capital.

To some extent, this is a microcosm of the current AI investment boom—investors have long set their sights on mid - and high - level executives who have left large tech companies and the so - called "Six Tigers" of large models. Recently, I've heard many friends say that big institutions like Sequoia have resumed the strategy of "buying into the entire track". This situation makes it seem like we've traveled back to the era of mobile internet.

The technology is still in the process of maturing, and the form of the product doesn't seem that important because it will probably change. From the decision - makers' perspective, projects have high uncertainty, and leading institutions have an advantage in terms of capital. So, it's reasonable to adopt a strategy similar to that of the mobile internet era. However, no two eras are exactly the same. What will be the end - game of the industry? What opportunities are there for startups? What themes should be laid out at each stage? Investors looking at projects in the market need to fully unleash their imagination. While the strategy can be replicated, in terms of specific investment strategies, AI investment may not simply copy the playbook of the mobile internet era.

Let's get back to our guest, Juzi. Born in 1986, he has worked at Samsung, LightInTheBox, Boss Zhipin, and MiniMax. He is a senior product leader. In May this year, the product he founded, ListenHub, was officially launched. Using a popular term in the AI circle, it is an audio agent that can create a 3 - minute short content based on your prompt. To my surprise, the dialogue presented by two AI hosts was very engaging and human - like.

"Most of the past experience has become invalid. Today, entrepreneurship values the upper limit more than certainty." This was Juzi's answer when I asked about the differences between the two eras he has experienced. In this interview, you'll see his detailed explanation of this question. Concepts such as traffic, personalization, creators, and consumers have also been given different meanings by AI.

Our guest looks gentle and speaks in a soothing tone, making people feel at ease. In his student days, he was like Tao Yuanming, loving solitude and literature. He chose the "Teaching Chinese as a Foreign Language" major to push himself to communicate with the outside world. He is also obsessed with technology products and has maintained the habit of writing technology blogs until now—he runs an AI self - media called "Juzi's Soda Shop" from an insider's perspective. When talking about products, he rarely uses the inflated narratives common in the venture capital circle. He often says, "It's not good enough yet." His vision for the product is not to grab users' attention but to allow people to use it comfortably and accompany them in the long run. This idea undoubtedly has an idealistic tinge. However, beneath his gentleness lies a steady determination and a sense of unwillingness that the mobile internet era has ended before he could fully showcase his abilities.

Chatting with him made me curious about a question: The most successful content products on mobile phones often take advantage of human weaknesses. In the AI - driven startup wave, is it possible for a product that allows human nature to flourish to succeed? Or, to put it another way, is it possible for startups in this era not to pursue extreme scale?

Juzi's favorite color is, of course, orange. His preference for it dates back to the BlackBerry era when the French Telecom version of the phone was orange. But I'd like to assign him another color. Flaxen, as the name suggests, comes from the herbaceous plant flax. It is a tea - colored shade with a gray undertone, reminiscent of dried grass stems or other elements in nature. It is soft, simple, and enduringly appealing.

Here is the content of our conversation:

Release the demo in October, get the TS in November, and register the company in December

Liu Yanqiu: First, tell us about the opportunity that led you to start your business?

Juzi: At that time, Google's NotebookLM had been out for a month. It was a revolutionary product. For the first time, the content generated by AI could be directly consumed without human intervention. In the past, there was a role called the creator. AI provided creators with some options to choose from and edit, and then the creators would produce works for consumers. The content couldn't be completely produced by AI. But with NotebookLM, you just need to give it a file or a topic. The significance is that if you only rely on existing creators to produce content, the market will be very small, covering only 1% of the people. Once you use technology to lower the threshold and turn consumers into creators, the market will expand by at least 100 times. We all thought it was an opportunity. We released the demo in October 2024, met with investors, got the TS in November, and registered the company in December.

Liu Yanqiu: Nowadays, it's basically an era of science and engineering entrepreneurship. How did studying Teaching Chinese as a Foreign Language affect your later career choices and the way you think about business after starting your own business?

Juzi: I have to start from the beginning. I was born in a county in Shijiazhuang, Hebei. I attended primary school, junior high school, and high school there. I was a science student, and my academic performance was not outstanding. I was always in the middle of my class, usually ranking over 100th at the start of school, but I always got first place in the graduation exams. The same was true for the college entrance examination. I was in excellent condition and suddenly got the first place in the whole school. It was the first time I got first place in my entire high - school era.

After the college entrance examination, I still wanted to study languages because I'm an introverted person who doesn't like to communicate much. So, I thought learning languages was quite important and hoped to change myself through language learning. I also love literature. At that time, the best Teaching Chinese as a Foreign Language major in China was at Beijing Language and Culture University. Although I scored over 600 in the college entrance examination and could apply to most universities except Tsinghua, Peking, and Zhejiang Universities, I still chose Beijing Language and Culture University. It's a pity there was no Zhang Xuefeng back then. If there had been, I might not have applied (laughs).

Liu Yanqiu: Which writers and works do you like?

Juzi: I like One Hundred Years of Solitude and Chinese magic - realist works in the vein of One Hundred Years of Solitude, such as On the Banks of the Eerguna River. I'd recommend One Hundred Years of Solitude to all my friends. If they don't read it, I'll recommend it to them every year. Several of them finally read it after I recommended it three times, and they all told me the book was great.

Liu Yanqiu: What's the reason for your special admiration for One Hundred Years of Solitude? Do you like the relationships between the characters, the literary atmosphere, or the message it conveys?

Juzi: One Hundred Years of Solitude presents over a hundred years of history in a fast - paced book. A man leaves his village to build a new one, and through generations of development, it experiences the process from ancient times to the modern era. It also shows people's reactions to great changes and many worldly affairs. What's remarkable is that the state of people is depicted very realistically, but it's presented in a fictional environment, which is very engaging. It's a bit like a game written by the author.

Liu Yanqiu: Now, let's get back to your career path. How did you transition from Beijing Language and Culture University to becoming a product manager?

Juzi: The campus atmosphere at Beijing Language and Culture University is very unique. About half of the students are foreigners, and the overall environment is very free and inclusive, which is a very rare experience among domestic universities. During my internship in my junior year, I started teaching Chinese at a language school. Most of the students were foreigners around my age. I had a great time teaching for three weeks, but I also realized that this might not be the life I wanted. I didn't want to repeat the same teaching content for my whole life. How boring that would be.

Since I'm very interested in technology, I started writing blogs at that time. One of my best - written blogs was called "Berry Culture", which was about BlackBerry smartphones. It ranked second among similar blogs in China at that time. I also served as a moderator in many digital forums and made many friends. Later, I worked with a friend in Nanjing to develop BlackBerry software and sold it to the United States and Canada, earning one or two million yuan. There were many talented people in our circle, and my life experiences were closely related to theirs. Later, BlackBerry declined. I asked a friend what my next career choice should be. He said I had two options: becoming a product manager or an interaction designer. I read all the books on product interaction and product management that I could buy on the market. After that, I decided it was better to become a product manager.

Today, the product success rate can be as low as 50%. What matters is the upper limit

Liu Yanqiu: I tried using Listenhub. The realism of the human voice was beyond my expectations, but I'm not sure where to use it. Did you have any ideas about the target users of this product before?

Juzi: Ultimately, it's more related to the usage scenarios and can cover a wide range of people. You can listen to it while driving or exercising. In short, when you're not convenient to use a computer but have some free time, you're suitable to use this product to obtain various kinds of information, and it may even include music. At the same time, it will be related to smart hardware, such as smart headphones, smart glasses, smart cars, and smart cockpits, because people interact through voice in these scenarios. In the short term, we want to attract users to help us produce some content. At this stage, we're looking for users who are well - educated, hungry for information, and willing to share it.

Liu Yanqiu: Will people remember things more deeply by learning through "listening"?

Juzi: At least all languages are definitely learned through a lot of listening. Many people learn English by listening to Friends a lot. Listening forces you to linearly access information. The Transformer architecture is a linear thing. When the model is trained, it reads line by line. In earlier times, such as in the CV and NLP eras, it was all about pictures and reading one by one. There are various efficient ways to input information. For example, some people say they want to develop a diffusion - based model. A non - Transformer architecture can generate an article quickly, but why is the Transformer the most successful today? Because it mimics humans. Humans intake information in this way.

For example, the best way to motivate yourself to read a book is to understand it thoroughly and then share it. I read articles like this now. I'll type them out while reading, which is very slow but helpful for memory. Many people tell me, "You read too slowly. I can read ten lines at a glance." I want to ask, "Can you remember it?" Why do we spend a long time reading a novel? For example, The Hitchhiker's Guide to the Galaxy. An hour after you finish reading it, if you're asked what the book is about, you probably won't remember anything except "the ultimate answer to the universe is 42". You remember the catchphrases and forget everything else, but during the whole reading process, a lot of things have unconsciously been stored in your mind.

The same is true for listening to voices. We've tested that the things you remember after spending 5 minutes listening to a topic are completely different from those after spending 5 minutes reading about it. We still hope that people can remember some important things.

Liu Yanqiu: What a clumsy way of learning for humans.

Juzi: The clumsy way is the most effective and flexible one. Didn't Duan Yongping always say that slow is fast? Maybe it's true.

Liu Yanqiu: You've relatively comprehensively experienced the mobile - internet era. Nowadays, many people like to draw analogies between AI applications and the era when mobile - internet products boomed. In what aspects do you think they are similar, and where are the differences?

Juzi: In fact, a lot of previous experience has become invalid. For example, in the mobile - internet era, people pursued certainty in many things. It's different today. Today, the success rate of your product in completing tasks can be as low as 50%. What matters is the upper limit. You need to have a 10% chance of creating something that scores 100 points. Generating some content that catches people's attention and spreads is also very important. If you create a product that scores 60 points and only improve its accuracy, it doesn't make sense today.

Because the technology has uncertainty and is highly related to the users, the overall engineering standards have changed. In the past, users in a product could be well - guided on what to do next. What you had to do was to make as few mistakes as possible and guide users through the user journey. Today, the user journey is completely different because users are facing an input box, and they start the user journey themselves. So, the requirements for users are very high.

The technology is still undergoing huge changes. There will be significant changes every year in the future. When you look back at today's data and functions, they may seem inappropriate. Since the biggest variable is technology, a good strategy is to make good use of this variable and let the product grow and evolve on its own. Each module is completed by AI, so when the model is upgraded, the product can also improve itself instead of me patching the model.

Liu Yanqiu: To achieve what you've mentioned, what are the differences between the organization of an AI company and that of the previous - generation companies?

Juzi: Everyone in the team must incorporate the AI variable into their work. If an engineer in the team writes code without using AI today, he will definitely be eliminated. The same goes for product managers. The whole organization must realize that AI will definitely surpass them. The most terrifying thing for a team today is that people think AI is still not good enough. That may be the case now, but it's evolving at an extremely fast pace. The biggest change in the past year has been AI coding. A year ago, if you said AI coding was no good and you didn't use it, and you still don't use it today, your efficiency will be very low. Another typical example is MCP. MCP was useless a year ago, but today it's a standard. We must embrace uncertain things and variables because they will get better for sure.

Moreover, the work process has also changed. For example, in the past, a requirement would go from the user to the product manager, be discussed, and then go to the R & D department. This process is wrong. In today's era of technological variables, many requirements should actually come from engineers. Whether they're reading papers or using new products, they naturally think about how to improve things without going through the product manager. Everyone's skills should be diverse. Product managers should learn AI coding, and engineers should also think about how to improve the product. Everyone is getting stronger and more versatile.

Liu Yanqiu: You've been in business for more than half a year. What's the most anxiety - inducing thing for you?

Juzi: It's whether the organization can become an AI - native organization. Because the technology is changing so fast, there's a major change every six months. If you use the experience from six months ago to guide today's work, it's easy to become outdated.

At MiniMax, we tried using the experience of the mobile - internet era to guide our work throughout 2023, and it was all wrong. The lessons were painful. 2024 was a year of reflection. Finally, people started to realize the importance of the web - end and began to make up for things related to computers. Of course, there are more mobile users, but as I said, we must respect technology, and that's how it is.

The Agent is mainly used on computers this year. We're far from reaching the stage comparable to the mobile - internet era. Around 2014, Redmi brought Android phones into millions of households. Location - based offline delivery and other things became popular along with Android phones. Before Redmi, there were mainly feature phones. At that time, talking about the mobile internet didn't make any sense.

Previously, personalization meant distributing pre - made content. Now, it means creating content on - demand

Liu Yanqiu: To be honest, in the mobile - internet era, audio didn't seem to receive much attention. If we look at consumers and creators separately, do you think there will be many C - end audio creators?

Juzi: This is actually a spectrum. Just like before Douyin emerged, no one thought there would be so many video creators. Now, 10 million videos are produced every day. Even apps like Jianying, which