We Invested in a Gen-Z "Veteran" Born in 2003 | Yunqi Capital Y Talk

What does 22 mean? Usually graduation, naivety, and endless second-guessing about the future. But in an era when Gen-AI is reshaping how we work and create, that age is getting answers far faster than before.

For Yang Bolin (William), born in 2003, 22 has been a journey of using AI to transform how knowledge is acquired. Three days to vibe-code a product demo from zero to one. Rapid validation, rapid iteration, rapid team-building... In a time when everyone's a builder, William and his education agent "Cuflow" (pronounced like Qflow, from curiosity and flow) have launched into entrepreneurship at a pace that feels distinctly AI-native.

But "fast" doesn't mean a light, straight line. Long before this, William treated the world as something to be tested firsthand:

In third grade, inspired by a magazine story, he rode buses back and forth across his county seat in Guizhou selling newspapers. At 16, while studying abroad, he returned home abruptly to help handle the family business. He went on to work in brand globalization, online education, and more — exploring different fields, getting results, and moving on. These experiences forged in him an ability to deconstruct complex problems and a tolerance for uncertainty.

In 2025, Yunqi Capital's investment team met this "veteran teenager" with unusual iteration speed. Cuflow became one of the first investments in Yunqi's post-'98 investment program, "Y Transformers."

"Full Throttle: Yunqi New Year Collection" Vol. 02 brings you excerpts from the debut episode of "Y Talk," a column on Yunqi's brand podcast "Attent!on" dedicated to the Y Transformers program.

Scan the QR code above or follow "Attent!on" on Xiaoyuzhou to listen to this episode

In this conversation, you'll hear about:

• Generational collision: How investors (post-'90s/'95s) and a post-'03 founder define entrepreneurial traits in the Gen-AI era, and "founder drive" in an age of divergence
• Product philosophy: Why an AI classroom shouldn't just be a chat box, but a "full-modal space" with perception and orchestration capabilities
• Growth profile: A maturity model for a serial entrepreneur born in 2003 — objective self-awareness, ability to deconstruct complex problems, emotional stability in uncertainty
• Industry insight: With large models iterating so rapidly today, why is "user editing" the best gift developers could ask for?

Guests:

William Yang — Founder & CEO of Cuflow, an education agent product

Han Yi — Lead investor, Yunqi Y Transformers

He Zeqing (Cobo) — Lead investor, Yunqi Y Transformers

Host:

Linda — Managing Director, Yunqi Capital

*Excerpts edited for clarity

01

Why Are "Post-'98s" Naturally AI Native?

Linda:

Since last October, our Y Transformers program has met many post-'98 and even post-'00 AI founders. As investors, what's the most striking trait you've observed in this generation?

Cobo:

The most obvious thing is how AI-native they are. From day one, whether it's building products, internal collaboration, or validating ideas, using AI tools is almost a default behavior for them.

Another thing is iteration speed. The gap from idea to product is much shorter, and in terms of business sense, many teams are naturally oriented toward global markets from the start.

One more interesting point: their understanding of self-media. The post-'98 generation is almost native to social media — extremely skilled at building influence through content and personal expression, which actually matters a lot in the early stages of entrepreneurship.

Linda:

They really do seem to have grown up on social media. Han Yi, what's your take?

Han Yi:

I'll add something I personally value highly, which I call "drive." We're in an age of divergence — lots of opportunities, lots of risks. In this environment, whether a founder has a clear sense of "I want this" is crucial.

This "drive" isn't sloganeering. It's in the actions: I want to build a great product, I want to be more AI-native, I want to lead diverse teams forward, and I want to keep making decisions even amid uncertainty.

Linda:

William, as a member of this post-'98 founder cohort, do you agree with what the investors just said?

William:

Pretty much. But for us, we probably don't actively think "am I AI native?" The world just seems like it's supposed to be this way — we're just moving forward with the environment.

Maybe because information and tools are so abundant, there's a natural mindset: since there are so many opportunities, why not reach for more?

02

"Veteran Teenager":

Crashing Into the Real World Young

Linda:

The investment team had a comment about you, William — that you have a "maturity beyond your years." How do you see that?

William:

Honestly, I'd never thought about "maturity" before fundraising. Several investors gave similar feedback, so I started to take it seriously: what exactly are they talking about?

I later broke it down into roughly three levels. First, relatively objective self-awareness — knowing your boundaries and knowing what you don't know. Second, the ability to deconstruct and handle complex problems. Third, emotional stability in the face of uncertainty.

Linda:

How do you think this "maturity" formed?

William:

More like a mutual shaping process. I probably always had some curiosity about the world and an impulse to change things, and some relatively early, weighty experiences reinforced those things.

Looking back, those experiences were high-intensity collisions with the real world — inputs that filled in my "value function" earlier than usual.

Cobo:

When I first met William, I described him with three words: leverage, teenager, veteran.

He's very good at leveraging more experienced partners and resources — a rare ability in young teams. At the same time, he has that strong post-'00 energy and fresh approach, but he's not naive when it comes to goals and decisions.

03

"Building Is Easy;

Starting Up Is Still Hard"

Linda:

William, before Cuflow, you already had several different types of entrepreneurial experiences — brand globalization, medical supplies, online education, and now an education agent. Some people might think that in today's environment, it's become especially "smooth sailing" for young people to start companies?

William:

If we're just talking about building, then yes, entrepreneurship itself is simpler. For example, with Cuflow, I initially vibe-coded alone from zero to a working demo in about three days.

But entrepreneurship, at its core, is still very hard. It's not about "acquiring some production factors" — it's a sustained drain on mental energy. Often you feel like there's nowhere left to retreat — as Jensen Huang put it, chewing glass while staring into the abyss.

Linda:

Looking back, which of these entrepreneurial experiences influenced you most?

William:

The previous online education venture. I observed roughly two thousand class hours, and what surprised me was: a great class doesn't depend on information volume, but on engagement. Later I abstracted a metric called Aha moment per minute. Good teaching should constantly generate "I get it" moments, not one-way teacher output.

Han Yi:

I actually agree with viewing serial entrepreneurship as a leveling-up process. Everyone knows that real success might only need one shot; the earlier experiences mostly accumulate understanding of business and how the world works.

Especially for this post-'98 generation of founders, this process is completely normal. Even if they're all early-stage, even if many start from running a small business, these experiences gradually build judgment about rules, human nature, and resource allocation. When you truly need to do something longer-term and more complex, these become foundational capabilities.

04

What Kind of AI Tool

Can Actually Teach Someone?

Linda:

Back to Cuflow — what problem are you fundamentally trying to solve? What category of needs do you most want to address?

William:

We actually thought about many traditional education problems at first, but ultimately abstracted it: we're essentially more of an information organization company. The core question is simply this: how should information be efficiently organized and transmitted to actually be understood?

The future we see: in the post-AI era, most skill-learning and general education will trend toward bottom-up, self-directed completion by learners. Schools may increasingly focus on guidance, social practice, and human connection. What we want to provide is a platform where learners can complete a self-centered learning experience.

Two key reasons drove us to commit to Cuflow.

First, from my previous online education experience, I observed: great teachers are hard to scale. We roughly saw a 1:9 phenomenon — 90% of students want to take classes with 10% of star teachers, but star teachers' time is limited; even when the remaining teachers work very hard, their share of class hours and income is quite limited. So "good teachers" themselves are supply-constrained.

We also examined nearly all mainstream AI education products on the market, domestic and overseas. Our ultimate conclusion: not satisfied. At the time, there were mainly three categories: photo-based problem solving, learning tools, and text-dialogue products built through prompt engineering. But learning itself is highly multimodal — a large portion of the cerebral cortex handles visual processing, and text is essentially a heavily acquired capability layered on top of the visual system. To put it more "model-like": vision is parallel processing, text is more sequential. Naturally, what the eye directly sees and perceives gets processed more efficiently.

So we believe the real learning experience shouldn't just be "answering questions," but stronger visualization, more immediate interaction — especially solving for "delivering a great class." Because the key in learning is whether aha moments keep happening, whether real interaction actually occurs.

The second reason is more personal: I grew up in an ethnic minority area of Guizhou, then went to study in the UK — I've experienced relatively poor educational environments and relatively good ones. I feel the difference isn't in IQ, but in the information delivery system being broken: information appears in forms unsuitable for absorption, not matched to the learner's cognitive bandwidth. The value of a great teacher is that they can adapt to your cognitive bandwidth, using the most suitable methods and multiple tools to make things clear. AI makes it possible to personalize and scale "the abilities of very expensive teachers," and we really want to do this.

Linda:

So you're emphasizing a "full-modal AI classroom."

William:

Yes. From the start, we didn't position Cuflow as a chat-based learning tool, but as a generative AI classroom with adaptive full-modal capabilities.

The most accessible understanding of full-modal is: students can see, hear, read, write, draw, drag — and AI can also speak, draw, demonstrate, generate code, charts, and video within a "classroom." But what we think matters more: full-modal isn't just "generating different modalities," but having modal reasoning and modal orchestration capabilities — reasoning over graphics, reasoning over spatial relationships, and correcting reasoning through interactive feedback.

For example: you draw a wrong formula or diagram on a real-time whiteboard. The AI doesn't just say "you're wrong" in words; it traces back the structure of what you drew, aligns the formula, highlights the wrong part, and uses animation to show the correct process so you clearly understand where you erred. Simultaneously, it continuously assesses your learning state and dynamically decides: at this moment, should it explain with text, with diagrams, with animation, or run an experiment — which modality best advances understanding? We call this modal adaptivity.

In short: AI is no longer just answering questions, but understanding that "learning" itself is a multimodal cognitive process, and at the right moment using the right expression to make understanding actually happen. Understanding occurs in the process.

William:

Another module extremely important to us is Flow Notes. It's a streaming note system. We want to call upon the right modality at the right moment in a full-modal learning space, and the prerequisite for this is the system truly understanding the user's current learning context.

In the internet era, context capture was often passive — requiring users to actively upload materials, repeatedly supplement information. We wanted this to become more natural.

So we designed Flow Notes. As users attend classes, read, converse with AI, or watch videos, it automatically captures and structurally organizes multimodal learning content, connecting knowledge points from different sources, generating a personally-marked stream of learning notes that help users gradually organize fragmented information into a traceable, evolvable knowledge system.

More importantly, this system isn't just for users — it also serves the AI. The AI gains insights from these session-dimension notes, using them as input for subsequent explanations and interactions. In a sense, it's an externalized memory management system that both humans and agents can use.

---

05

AI Applications: The Hard Part Is After the Model

Linda:

Han Yi, you've repeatedly emphasized internally that many models and applications, when actually moving toward commercial scenarios, encounter massive engineering problems. The memory system, tagging, and so on that William just mentioned — are these the "engineering work" you're talking about? How important is this?

Han Yi:

The improvement of AI capabilities now does partly come from foundation models and frontier models themselves — that's undeniable. But when these capabilities need to land in specific scenarios, there will definitely be massive work beyond the model itself.

Take the memory system William mentioned — it contains a whole set of design thinking and philosophy: when to store? What to store? How to store?

Different companies make completely different choices on these questions, and the resulting product experience differences don't correspond one-to-one with model updates. So I understand this as a very typical manifestation of engineering.

Linda:

You actually used a vivid analogy before.

Han Yi:

Right. I often use "electricity" as an analogy. The invention of electricity itself greatly improved humanity's efficiency in using and transmitting energy, but what truly determines the experience is how electricity gets used — how electric cars use it, how power grids use it, how light bulbs and phones use it. To make these things work, we need different batteries, solve charge-discharge logic, and consider safety, efficiency, and stability. This work itself is on a different level from "whether electricity exists," but it determines whether electricity can actually be used well.

Linda:

So in AI applications, this kind of engineering isn't "nice to have" — it's mandatory.

Han Yi:

Absolutely mandatory. Especially in real business scenarios, without this engineering-level work, model capabilities are very difficult to release stably and sustainably. Take Manus — what people see is a smooth product, but behind it is massive engineering work on memory systems, information I/O, and spatial management.

---

06

Why Cuflow?

Linda:

From an investor's perspective, the AI education direction has been discussed many times, and you've seen many projects. So why Cuflow in the end?

Han Yi:

I generally look at two levels. First, our investment mindset toward AI applications itself. We're in an age of divergence — many projects adjust direction or even pivot at different stages, which is very common in AI applications. So in this environment, I don't get too attached to the day-one product form, but focus more on the team itself and the founder's traits.

As I mentioned earlier, I highly value "drive" in a founder. When many things are unclear, this internal motivation matters even more.

The second level is the education market itself. Education is a long-standing large market with strong willingness to pay — people always need to learn and improve. In every cycle, new technologies and product forms emerge to improve how people acquire and connect knowledge.

What we're more betting on is: in this large market, through a product or service form, to further enhance people's ability to acquire and connect knowledge.

Linda:

Given these premises, what do you value most in Cuflow specifically?

Han Yi:

Their iteration speed. AI applications are no longer simply "software plus AI" — forms keep changing, even getting overturned by new technologies. So what we want to see more is a capability: can you continuously keep up with cutting-edge technological changes and translate them into product capabilities, rather than getting locked into one form?

If fundamentally new interaction methods emerge in the future, or entirely different ways of connecting people with knowledge, today's product forms might all change. In that situation, what matters is whether capabilities can keep adapting and reconstructing.

Linda:

Cobo, you were the first from Yunqi to meet William. When did you sense this change?

Cobo:

Actually, changes across different time slices. When I first met William, I saw him as a young entrepreneur with strong entrepreneurial potential, very clear about what problems he wanted to solve, but still relatively early in technical and product depth.

After some time, meeting again, I could clearly sense his deeper understanding of large models and engineering details, and he was developing clearer product sense. What we saw was someone learning and iterating extremely fast, and this change was continuous.

Linda:

So what you saw wasn't a "finished" entrepreneur.

Cobo:

Right, but someone in motion. Most recently, talking again, he's become more restrained in product philosophy — converging from a bigger, more comprehensive vision toward a more elegant, more sustainable product direction. Many products that truly go the distance are polished this way, step by step.

---

For the full conversation, search and subscribe to "Attent!on" on the Xiaoyuzhou platform to listen to the latest episode.

The Y Transformers program is also continuously accepting applications. Scan the QR code below to create with us!