On November 30, ZhenFund hosted a ZhenCampus discussion on AI entrepreneurship at Tsinghua University. ZhenFund Managing Partner Dai Yusen was joined by Zhang Yutong, President of Moonshot AI (Kimi), Zhang Huaiting, Founder of AiEDU, and Zhang Tao, Co-Founder of Manus, for an in-depth conversation on innovation and the future of AI.

This March, AiEDU launched its first AI product, the Aixue app, formally bringing digital teachers into one-on-one lessons. Today, it can deliver entire middle-school lessons through AI, with no human intervention. AiEDU has not only achieved this but has also been the first to do it at scale. Monthly revenue has reached tens of millions of yuan, and more and more students are experiencing an entirely new way to learn.

On March 6, 2023, Zhang Huaiting and Liu Wei, a former colleague at Gaotu and former General Manager of Gaotu Classroom, decided to start a company after a ten-minute phone call. They discussed two questions: Could today's AI fundamentally transform productivity and business models? And was education still worth pursuing?

Their answer to both was yes, and they believed they were the right people to do it.

As AI technology advances rapidly, educational paradigms are changing, though adoption moves at different speeds across fields. Education leaves extremely little room for error. An education company's first obligation is to get things right. Beyond that, AiEDU hopes to deliver genuine personalization: whether a student is shy, confused, or talkative, they should receive appropriate guidance, providing education for everyone and teaching each student according to their needs. Completion rates, accuracy, and renewal rates will tell us whether it works.

From joining Baidu at 27 to founding two companies, Zhang Huaiting also used this event to reflect on where his entrepreneurial journey began. For him, entrepreneurship is a deliberately chosen way of life——a way to witness the beginning of an era and to make tomorrow more exciting than yesterday.

As a former core leader of Baidu's Phoenix Nest platform and co-founder of GSX/Gaotu, Zhang Huaiting may be among the founders in China best positioned to answer the question of AI and education. ZhenFund has had the privilege of supporting AiEDU continuously since its angel round, exploring the future of education together.

The following is the full transcript of the discussion.

01

Two Questions About the Future of Education

Dai Yusen: More than a decade ago, you were one of the leaders of Baidu's Phoenix Nest platform and led a team that won Baidu's highest award. Ten years ago, you co-founded GSX, now Gaotu, as COO. Its market capitalization peaked at more than US$30 billion. Most people would probably consider you already highly successful. Yet in 2023, you decided to start another company. Why do it again, and why then?

Zhang Huaiting : I was afraid of getting old.

Former Israeli Prime Minister Shimon ·Peres once said: If what you will do tomorrow does not excite you more than what you did yesterday, you are getting old. Boredom is more frightening than death.

I have always felt that life is about waking up each morning knowing exactly what you want to do that day and being impatient to get started. As long as you keep that excitement, you remain young in spirit. Conversely, if you wake up feeling there is nothing to do and nothing interests you, it is easy to grow old.

I once read an explanation that if you use your brain every day, brain cells continually regenerate, making it easier to stay young. I came to Beijing to study thirty years ago, and even now I feel excited and eager to do things every day, just like a young person.

In 2019, after my previous company went public, I retired too.

During that period, there was nothing I absolutely had to do each day, and I was not in a particularly good state. A couple of days ago, I spoke with a friend who achieved financial freedom after Baidu went public in 2005. He also went through a long stretch of boredom and emptiness and became increasingly listless, so he started another company. He told me he is now preparing for his third venture.

Entrepreneurship gives you something to be excited about every day, a reason to get out of bed. It is a way of life that can truly keep you young. That is probably what the old saying means: “Adversity sustains us; ease destroys us.”

Dai Yusen : In 2023, education had just gone through the “double reduction” policy, and AI education was still a direction few people agreed on. Why did you choose to start a company in this field that March?

Zhang Huaiting: Many people say you need to think repeatedly about an important decision. But I think that if you cannot work something out in a day or two, you probably will not work it out in a week or a month either.

For my first venture, I had known my partner for ten days. After four conversations, we decided to start a company.

This time, in 2023, Liu Wei——my former colleague at Gaotu and former General Manager of Gaotu Classroom——and I decided after a ten-minute call. Coincidentally, just over an hour after we hung up, Yusen messaged me. He said ZhenFund's research suggested that education might be the lowest-hanging fruit among AI applications and a promising area for startups, and asked whether I had any thoughts.

I remember clearly that it was March 6, 2023. Wei and I discussed just two questions.

First, how was this AI wave different from the previous one? Could it genuinely transform productivity and business models?

Second, was education still worth pursuing? Should a market-driven team bring AI technology into education in a meaningful way? Were we the right people to do it?

Both answers were yes. We confirmed our agreement in ten minutes.

Zhang Huaiting speaking at the event

Looking at the situation today, AI is at an explosive inflection point. Internationally, the giants are competing; in China, the overall direction is positive, but progress is uneven. The real competition in this period is over the paradigm and efficiency of developing talent. From that perspective, I believe AI education has enormous potential.

What is traditional education? In-person teaching is a broadcast: when students have questions, immediate interaction is difficult, and knowledge absorption is inefficient. Online education expanded the reach of that broadcast and addressed the issue of access, but without enough time and people, it still could not tailor teaching to each student.

Generative AI, however, may turn a service industry into a manufacturing industry, allowing every individual to receive immediate attention, immediate interaction, and immediate improvement. This is a transformation of an educational paradigm that has lasted thousands of years. For the first time, we have a chance to overcome the impossible triangle of low cost, high quality, and large scale.

We can do it, it is worth doing, and we have a chance to truly change a long-established educational paradigm.

02

Commit at the Technology's Tipping Point

Dai Yusen: At Baidu and GSX, later renamed Gaotu, you were very good at scaling organizations. Now that you are starting again from zero, what advantages do startups have in the AI era? How do they compete with large technology companies?

Zhang Huaiting: In 2020, two leaders of ByteDance's education business invited my partner and me to meet at Manet's Meadow. We discussed what education would look like in the future. Our competition in online education was fairly intense then, but that did not stop us from discussing the future educational paradigm.

At the time, I felt that recommendation systems could not deliver true personalization as generative AI can today, let alone provide personalized educational services at scale. So I said the direction was right, but the technology's time had not yet come. To run a real business, you need the PEST conditions to align: politics, economics, social habits, and technology. At that point, the technology simply was not ready.

In 2023, we saw GPT-4 launch and realized the technology was approaching a tipping point. Of course, it was not fully mature. But it is precisely at such a tipping point that you have to commit if you want to capture the opportunities of a changing era.

Online education used to be highly labor-intensive and depended heavily on organizational capabilities——essentially, people managing people. In the AI era, our internal requirement is to embed the organizational ability to manage tens of thousands of people into a system of intelligent agents, gradually replacing management-driven operations with technology-driven ones.

In fields such as education, healthcare, and law, which do not exist purely in a virtual world, it is difficult to launch an AI product with a hundred million users from day one. Users' tolerance for error varies enormously by scenario. For example, a mistaken ad or short-video recommendation is no great problem. But if an educational product entrusted with your child tells you it is 80% accurate and may be wrong 20% of the time, would you dare use it? Even in the same AI era, different contexts and tolerance for error naturally produce different development timelines.

At this stage, our strategy is still intelligence-driven, but humans must provide a safety net. As the intelligence improves, human involvement will gradually decrease. Today we have “intelligent systems plus human management”; over time, we will move toward relying primarily on intelligent systems. First humans, then humans plus intelligence, then intelligence plus humans, and ultimately full intelligence.

Dai Yusen: AiEDU created the world's first truly AI-native course, with monthly revenue exceeding ten million yuan. What technical challenges did you overcome to get there?

Zhang Huaiting: We can now teach Chinese, mathematics, English, physics, and chemistry in grades seven, eight, and nine entirely through AI tutors. A single lesson can last one to two hours, with no human involvement at all. From explaining definitions and working through questions interactively to helping students apply concepts to new problems and correcting mistakes in real time, the AI tutor can adjust the mode and pace of interaction and the difficulty of questions based on each student's performance as it happens.

At present, we are the only company in the world to have achieved this at scale. Tens of thousands of students interact on the platform every week. Our concurrency is so high that Volcano Engine engineers work on site. It is genuinely very difficult, because we have to ensure:

First, we must not teach a question incorrectly.

Second, students are in many different states. Some are shy; some cannot understand no matter how you explain; some like chatting; some do not like studying. How should you teach them? Set the pace? Keep them engaged? Give them a sense of achievement? Our analysis found that if a student gets either 100% of the answers right or 100% wrong during a lesson, they will not come back. So during the interaction, we need to probe the limits of a student's ability and stretch them just a little, so they feel a sense of achievement and want to keep learning.

We also have to handle concurrency, latency, and other issues in real-time generation. We use realistic digital humans, not cartoons, so lip movements must be synchronized. Teaching English requires correcting pronunciation. When a large model cannot understand a specialized term, we need contextual algorithms to compensate. Learning environments may also be noisy, requiring voiceprint recognition to distinguish whether it is the same speaker or assess their emotional state and decide whether the pace of interaction needs to change. These are all practical requirements.

Finally, concurrency creates cost issues. Startups must be sensitive to costs. Cloud computing is generally billed according to peak concurrency, so we have to smooth out peaks and troughs to meet users' needs while lowering costs. I imagine that at university you mostly study theory and technology, and rarely consider the relationships between finance, legal matters, cost optimization, user behavior, and business logic.

If you are interested in the future, you could join a startup team or build a project yourselves. The real experience is completely different from a textbook.

Dai Yusen: In 2005, you joined Baidu, which listed on Nasdaq. You are now on your second venture. How do you weigh the choice between joining a startup and a large company?

Zhang Huaiting: Do you think someone who achieves complete financial freedom gets there by earning a little more each year over fifty or sixty years, or by earning decades' worth of money at a particular moment?

Look at today's rich lists. If you accumulated wealth by earning one million in the first year, two million in the second, and three million in the third, you could never reach those levels——especially since many people on the lists are still young. The key is to move with the times and seize the opportunities an era offers us. The real difference often lies in how you think and when you act.

After graduating in 2003, I joined a listed company, but after two years I felt it was not right for me. Its thinking and organizational style were fairly traditional. I could even foresee what the next twenty years would look like and how much I would earn. So I decided to look elsewhere for opportunities.

The internet was just taking off. When I interviewed at Baidu, I noticed all the meeting rooms had transparent glass walls. I saw someone in baggy shorts and slippers shuffling into a meeting room with a group of engineers and thought it looked fun. Baidu had few people then, but it felt both relaxed and energetic.

I had five offers: Microsoft, IBM, Huawei, Saisi, and Baidu. The others were in traditional software; only Baidu was an internet company, and it offered me a little stock. Startups had little money then, and I did not know what to do with the stock. I just remembered the Nasdaq bubble had burst and NetEase's share price had dropped below one dollar. I was afraid too, but felt this company represented a new way of thinking, so I joined.

I was 27 when I joined Baidu and stayed almost ten years. By 2014, when BAT's market capitalizations were at their highest, the mobile internet had arrived, and I felt I could go out and do something again. The mobile internet was beginning to connect with traditional industries. I went into education, using technology to transform it, because I felt the direction was right.

In 2019, after the company listed, I felt my passion had diminished. The next productivity revolution seemed to have hit a bottleneck, so I left to invest.

In 2023, I saw generative AI arrive. Its capabilities were completely different from the recommendation systems of the past. I felt we could do something again, and returned to entrepreneurship.

Looking back, each of my three choices came at the beginning of an era, and each led me to a place that matched my values and culture. At Baidu, GSX——later Gaotu——and AiEDU, I have wanted to create a technology-driven organizational culture that young people enjoy and that seeks to change the world.

So in both ventures, we set our mission, vision, and values from the outset. We told the team where we were going, how we would get there, and what kind of people we were.

If you want to grow quickly and accumulate greater wealth, the most important thing is to move with the times. Either try it yourself or join a company you believe is truly good.

03

Build Barriers at the Intersection of Fields

Dai Yusen: What are AiEDU's values?

Zhang Huaiting : Our mission is “Love and AI help everyone become a better version of themselves.” Our vision is “To become a trusted growth partner for everyone.” Our values are “Love yourself, love your partners, love the world.” Only when you take care of yourself can you lead a team and influence those around you. When an organization has enough strength, it can try to change the world.

Dai Yusen: How do you attract technical talent?

Zhang Huaiting: I think three things matter most to technical people.

First, sufficient respect. We provide an environment that encourages innovation, tolerates failure, and truly values technology, allowing it to make a difference.

Second, I believe technical people all have a dream of being a hero. Our generation grew up reading Jin Yong and Gu Long and imagining the world of martial-arts heroes. Technology is that world today, and everyone has the chance to leave their own story in it. We hope to give people that space and opportunity.

Third, let people know the organization is dependable.

Attracting talent is somewhat easier for me today, because many AI professionals in China trace their roots to the model team of Baidu's Phoenix Nest platform: the founding team of Fourth Paradigm, Kuaishou's founders and core team, and Xiaohongshu's core technology leaders, among others. So I tell people to focus on doing their work. We can change the world together and be rewarded for it too.

Dai Yusen: How can startups face fierce competition from large technology companies?

Zhang Huaiting : People who work across fields are actually the most likely to find a new path. Although I have a background in machine learning, I did not choose to build models directly when I started my company, because combining models with education creates substantial business barriers in itself.

Look back at the online education era: Baidu had Chuanke, Taobao had Taobao Education, Tencent had Tencent Classroom, and ByteDance had Dali Education. Yet none of these large companies ultimately made it work. The company that broke through was GSX, later renamed Gaotu.

There used to be a theory of organizational “DNA”: what an organization understands and its underlying makeup determine what it can do. Money and traffic do not guarantee that you will do something well. What matters most is understanding the intersection of fields deeply enough, moving fast enough, and continually building barriers.

Dai Yusen: A product cannot rely solely on improvements in model capabilities, right?

Zhang Huaiting: Today, everyone can ask Doubao questions. It can answer, and I can judge whether its answer is right. But which parent would entrust their child's education to Doubao?

Education allows extremely little room for error. We must optimize both the model and the business specifically for educational scenarios and ensure we do not make mistakes before we can truly deliver for parents. That is the most fundamental business logic of education.

Dai Yusen: What kind of talent are you looking for now?

Zhang Huaiting: Our goal is to become a top AI education company globally. If not number one, then at least in the top three. We welcome people working on speech, digital humans, large models, strategy, and engineering.

Q: Will AiEDU offer early education for children aged three –to five?

Zhang Huaiting: If you download the Aixue app, five-year-olds can already learn phonics through interactive, game-based AI teaching.

But we may not build every product ourselves in the future. We hope to make the paradigm work and invite more people to build it together. Perhaps on the Aixue platform, different people will contribute different capabilities and teach one another, turning it into a truly decentralized platform for learning and interaction across all age groups. That would be more interesting. No single company can build every educational product.

Q: How do you get students and parents to accept AI education and be willing to pay for it?

Zhang Huaiting: Look at the learning outcomes.

We are not a utility product that someone uses today and abandons tomorrow. Our customers are genuinely willing to spend real money to study on the platform over the long term.

Ultimately, you only need to look at a few metrics: Completion rates, interaction, answer accuracy, and renewal rates. All the answers are there.