Dropping Out of Middle School at 14 vs. Writing Code at 13: The Unconventional Path of a Non-Standard AI Founder
When the endgame is unclear, what do you set out with?
Born in 1996, Xie Yang has the unmistakable markings of a hardcore geek: he started programming at 13, served as a core hacker at ByteDance, and launched his first company at 18. A serial entrepreneur, he founded Authing, a leading identity cloud company in China's SaaS space that was named a Technology Pioneer by the World Economic Forum (one of just 15 companies from mainland China selected). He was also named to Forbes Asia's 30 Under 30 list. Now he's building in AI, and in 2025 released the world's first Agentic Browser — an action-oriented browser called Fellou.
At the same age Xie Yang was writing his first lines of code, Qin Tian and Deng Jie were embarking on their own "social survival challenge." To escape domestic violence, the 14-year-old Qin Tian ran away from home with his childhood friend Deng Jie. They worked as game power-levelers, toiled in illegal factories, even welded batteries. Deng Jie eventually returned to school; Qin Tian kept grinding. Operating entirely outside mainstream view, this pair grew up wild, developing a sharp nose for traffic. Over eleven years, they built multiple hit products in succession, eventually earning $6 million annually from a single VR game. Now they've turned to AI, launching Proactor AI, a proactive AI product.
On one side, a technically precocious elite who dominated his field. On the other, grassroots strivers who clawed their way up from the mud. Two radically different life trajectories, converging at the wave of AI entrepreneurship. And this proves something — in the raw early days of AI startups, opportunity doesn't ask where you came from. It only rewards those bold enough to set out first into the fog.
Scan the QR code to listen to the audio
Guests:
Xie Yang, Founder of Fellou
Qin Tian, Co-founder & CEO of Proactor AI
Deng Jie, Co-founder & CTO of Proactor AI
Host:
Xing Meng, Partner at 5Y Capital
"Every major technological transformation, especially a platform-level shift like AI, opens a window of time. It belongs to founders who don't look standard on paper — the misfits who set out before the world is ready for them."
"Productivity shouldn't be about rationality. It should be about sensibility. Productivity shouldn't just be about saving time. It should be about people — helping people gain the ability to understand the world."
"From day one, we have to consider whether the project can sustain itself. The upside of this model is resilience. The downside is strategic conservatism. We're learning to shift from quick payback to long-term investment."
Selected excerpts from the conversation:
Xing Meng: Today we're going back to the early days of entrepreneurship. I believe that in every major technological transformation, especially a platform-level shift like AI, there emerges a window of time that belongs to founders who look "not so standard" — the misfits. By misfits, I mean those who set out before the world is ready for them.
Today we've invited three founders who don't fit the typical profile of the perfectly credentialed entrepreneur. Each has their own unique story, starting point, and setbacks. Let's start with introductions.
Xie Yang: Glad to share some updates. I'm Xie Yang. We're building an AI browser that can think, act, and complete tasks autonomously within computer and internet environments. You could say we're building an autonomous driving system for computers and the internet. I'm currently based in the US, expanding our market there.
Qin Tian: Our product is called Proactor AI. Unlike most passive AI, it doesn't wait for your prompt or task. It proactively understands context from your meetings and daily work, captures latent needs from what you say and do, then plans and completes tasks automatically. Often you don't even need to articulate anything — it discovers problems and delivers results for you.
Deng Jie: Qin Tian and I started this together, so I'll skip the product intro and talk about our background. We ran away from home together in 2009, when we were just 14 or 15, beginning our life adventure early. We entered the software and internet industry in 2013, and founded our own company in 2014. Over the past decade, we've tried our hand at software, going global, VR games, and more, accumulating substantial experience. When the AI wave arrived last year, we decided to meet the challenges and opportunities of this new cycle.
From "Nine-Year Education Dropout" to AI
Entrepreneurship and Survival With No Safety Net
Xing Meng: I think you two had a particularly unusual childhood and entrepreneurial path. Qin Tian stopped schooling at 14 or 15; Deng Jie later went back to study software. By traditional standards, you're neither "typical entrepreneurs" nor "typical students." You've done genuinely hard things, grinding your way up from the very bottom of society to where you are today. What's even more remarkable is that your Proactor launched very early in the proactive AI space, when the direction was still extremely niche — only one or two American companies were doing it at the time. How do you think this background has created advantages and disadvantages in your entrepreneurship? What impact has it had on you?
Deng Jie: The advantage is that perhaps because we don't have deep technical backgrounds, we also don't carry heavy technical path dependencies. Often it's not "can we do this?" but "we think we can, so let's try." Our previous entrepreneurial experiences also spanned different tracks — from software outsourcing, to going global, to VR games, to AI. For us, there were no real barriers to entry. We were more like charging in and learning as we went.
But this background also brings some psychological challenges. Because we came from relatively grassroots origins and never had many resources, we never had the luxury of time to polish products or invest in long-term R&D. We were constantly thinking about where next month's money would come from, how to survive. So we tend to be eager for quick results — we want to see outcomes fast, rather than waiting six months or a year.
This mindset has indeed hurt us at certain stages. Even now that we have some cushion, we still make decisions with unusual caution and conservatism. Looking back, we've missed opportunities that later proved important.
Qin Tian: I feel very much the same as Deng Jie. One of our advantages is "the fearlessness of ignorance." You don't know where the ceiling is, so you dare to do anything. And through trying, you naturally discover new capabilities and understanding. This reminds me of Musk doing SpaceX — he wasn't a rocket expert either, but he dared to do it.
This fearlessness of ignorance can become an advantage in entrepreneurship, though it's a double-edged sword. Because your视野 is limited, you may have a harder time seeing the globally optimal solution from the start, making it easier to make short-sighted decisions. Another factor is that for over a decade, we've almost never taken funding — everything has been self-funded through our own earnings to support the team. This means from day one, we've had to consider whether a project can sustain itself. We couldn't lay groundwork for too long or play very long games. We could only earn money back from customers through products and services. The upside of this model is tenacious vitality and strong profitability. The downside is obvious — it breeds strategic conservatism and short-sightedness.
Xing Meng: Have you found the balance today?
Qin Tian: I think we're moving toward a new kind of balance. Our past projects, whether games or early global tools, were basically "small investment, quick returns" types — ideally launch in half a year, profitable within a year.
But in the AI direction, we've chosen a more challenging project with heavier investment and a longer payoff cycle. Proactor AI's R&D investment far exceeds anything we've done before, and the timeline is longer. In this process, we've also started actively engaging with capital markets, trying to bring in more external resources and leverage. I see this as our exploration of learning to shift from "quick payback projects" to "long-term investment innovation" — a way of finding balance.
After 35 Rounds of Deep Conversation With Claude
The Person Who Iterates Himself Like a Product
Xing Meng: Xie Yang is a serial entrepreneur born in 1996, and his first venture wasn't a "quick trial and failure" — it was a relatively successful company in the B2B space. You're an engineer by background, and building enterprise services in China itself isn't easy. Now you've chosen to restart in AI at this moment. From your first company, you've carried many labels and identity markers. What experiences did these past chapters bring you? And how have they influenced your AI entrepreneurship today?
Xie Yang: I started engaging with computers very early — Windows programming, network security, competitions as a kid, then knowledge graphs, then serverless systems at ByteDance. My path has always been inseparable from the internet. These experiences, interests, and reflections on the Web itself have accumulated certain insights through continuous practice, and allowed me to seize opportunities at different stages. I've always felt that much understanding must come from "learning by doing, doing by learning" — the experience itself is crucial.
Being CEO also brings considerable baggage. Or rather, a company's growth ceiling is often closely tied to the founder's personality, and the strengths and limitations of personality are double-edged swords. When talking with investors in the first half of this year, they often asked what my biggest takeaway from these years has been. I often answered: I've learned how to build connection with people.
Recently I had a conversation with Claude and asked three questions: First, why do I feel deep resistance at the thought of talking with people? Second, why does self-isolation feel like my comfort zone? Third, I feel that being CEO for so many years has caused some damage to my brain. Because being CEO basically requires constant permutation and combination, making many decisions daily, and most of these decisions are relatively macro. When I was a programmer, I could focus attention down to every logic segment, every button, every database table — completely different dimensions of thinking.
After 35 rounds of deep conversation, Claude's assessment was that I may have some ASD characteristics (neurodiversity spectrum, often called Asperger's). The ASD brain lacks an automatic social system. While others intuitively know others' emotions or intentions, people with related traits often have to rely on rational reasoning to analyze "why they said this, what it means, how I should respond." It's like most people drive automatic, while we drive manual. But at the same time, in deep thinking and pattern recognition, this type of brain often has more prominent advantages.
So all these years of entrepreneurship, for me, have also been attempts to break through certain personal limitations. How to cross these personality shortcomings and cognitive bottlenecks is itself a long road, and something entrepreneurs need to constantly think about and face throughout the process.
Xing Meng: I think you're very good at treating yourself as a product to iterate, maintaining long-term introspection. Next I want to ask — everyone knows the AI browser is an extremely fierce battlefield. When we talked in March, your product was scheduled to launch in April, basically among the earliest in the industry. But within just half a year, almost all major tech companies have entered this space. The browser itself has been the main battlefield of the internet for the past 20–30 years. Setting out here requires tremendous conviction. Why did you choose to enter from this track? From that day to now, what industry changes have you observed, and how have they affected you? Were they within your expectations?
Xie Yang: Actually all my entrepreneurship has revolved around productivity. I want to help people save time, improve efficiency, hand tedious work to AI, and let people return to nature, return to life itself.
But my experience in the US over the past month has given me new thoughts on productivity. Beijing is like a massive energy field, while the Bay Area is relatively more peaceful, relaxed, non-anxious, full of opportunity. In this context, I think my previous understanding of productivity was wrong. I think productivity shouldn't be about rationality, but about sensibility. Productivity shouldn't be about saving time or improving efficiency. It should be about people — helping more people gain energy, gain the ability to understand the world.
The browser is a natural environment for doing this. People browse webpages, search for information, make decisions in it daily, leaving massive behavioral data. We're building memory and profiles for users based on this data, and further providing proactive AI services based on these understandings.
For example, when a user shops for furniture at IKEA, they may not be sure what they want. They spend a few minutes on product A, check reviews on product B, add product C to cart. We can generate real-time price comparisons of the three products based on these behaviors, and provide capabilities like "one-click check lowest price across the web" or "quickly view reviews across the web." We're not trying to take over the user's choices, but providing appropriately automated support while the user maintains autonomous browsing and judgment. So far, from the overall automation capability score, tools including ChatGPT and Comate haven't surpassed us.
The thinking behind all this is that AI products shouldn't pursue comprehensive automation, but should respect human rhythm, human choice, and human ability to understand. When products are human-centered, what you build isn't a cold efficiency tool, but a system that understands needs, boosts energy, and delivers positive experiences. We believe in human growth, and hope to help people gain more through the browser — an entry point everyone uses every day.
Xing Meng: Can you give more specific examples of how the browser "understands people" and thereby helps users improve productivity? At this point, what's the biggest difference in philosophy between you and OpenAI's browser?
Xie Yang: The main difference is that we believe in infinitely amplifying the model's capabilities. Amplifying the model's capabilities is essentially amplifying human capabilities. Many people want to become stronger, and acquiring and absorbing more information is the prerequisite for improving productivity and execution.
So the question becomes: how do we inspire people? On this point, we've built more multi-dimensional intent understanding mechanisms and result presentation systems for our Agent system. When people think and make decisions, they often need discussions and challenges from different perspectives to reach more effective paths. Now I even feel that dialogue with AI is often more efficient than discussion between people, because you can have it give you pros and cons, then synthesize these perspectives into new insights.
Based on this thinking, we perform multi-dimensional analysis of the memory, knowledge base, behavioral trajectories, and profiles users form while browsing, and present these understandings to users in more complete, more inspiring ways — letting them have genuine aha moments, genuine insights when facing information, assisting their decision-making. This is one of our core philosophies.
The Secret Behind Billions of Views
Xing Meng: I want to come back to Qin Tian and Deng Jie. Your AI product has very high requirements for real-timeliness and proactivity, which actually places high demands on AI pipeline, real-time performance, memory organization, and workflow orchestration. But your team itself isn't specialized in this technical direction, yet you were able to launch such a product in a very short time. Why is that? What were the key challenges you overcame in this process?
Deng Jie: Regarding real-timeliness, we've actually done some instant messaging products in the past, so it's not completely unfamiliar — there's some connection to current needs. The real real-time challenge lies more in the model's immediate response, not the communication link itself. Currently we don't train our own models; that's not our strength. We mainly rely on third-party models.
As for real-timeliness, we may not achieve completely absolute real-time, but we can reduce users' perception of latency through workflow optimization, prompt design optimization, caching techniques, and streaming responses. So from an engineering perspective, although we're not a native AI foundation model team, our past software development experience, including having built networked multiplayer games, gives us certain understanding and accumulation regarding real-timeliness.
Xing Meng: I remember when we first met, Qin Tian showed me your content marketing playbook in the office. You explained your video production and distribution strategy to me from the beginning, like interviewing me as a new employee. You mentioned each video basically had to reach at least one million views on YouTube, with cumulative views approaching two billion.
Qin Tian: Currently it should be three to four billion views per year.
Xing Meng: For a startup this is very rare. The first time I visited your office, I found that in your roughly forty-person team, only about a dozen were in R&D, but you had a fairly complete content production team — uncommon in typical tech companies. Usually tech companies might have 90% in R&D, but you clearly have a completely different product promotion approach.
Can you talk about how you think about building an in-house content marketing team? And how did you develop a methodology that lets your videos achieve such massive reach?
Qin Tian: This methodology originated from habits we developed and transferred over from the gaming industry. The gaming industry has a common problem: most profits get swallowed by user acquisition platforms. When we did VR games, we faced a special difficulty — VR products are nearly impossible to buy traffic for, because VR games must be installed on Quest headsets, but users typically see ad videos on mobile phones. The link in between can't be attributed, so there's basically no ad space to buy.
We realized then that if we wanted our VR project to acquire users, we had to rely on social media distribution. So we spent massive amounts of time studying social media mechanisms, thinking about how content could "go viral."
The first year we didn't really have much clue, but we persisted in one thing: for every video we published, we tracked 2-second drop-off rate, 6-second drop-off rate, and completion rate, then analyzed commonalities and differences between viral videos and duds. We gradually discovered two key things. One is that regardless of platform — TikTok, YouTube, or Instagram — all video views essentially follow certain patterns. These patterns can be mathematically summarized as "completion rate × engagement rate." If these two metrics are high enough, they form a relatively stable views benchmark.
Take completion rate for example: when a video's 2-second drop-off rate is below 25% and 6-second completion rate reaches 60–65% or above, its probability of reaching a million views is extremely high. Further analysis revealed that completion rate can actually be derived through function fitting from "6-second completion rate" and "video duration." When we later verified this, we found that whether gaming content or AI product content, as long as it ultimately became viral, their completion rate curves almost all fit this function fitting model.
So how to continuously obtain completion rate and engagement rate? We broke this down further and identified roughly twenty to thirty quantifiable characteristics. These characteristics aren't traditional content intuition, but metrics that can be quantitatively analyzed through mathematics and physics. Based on these metrics, we gradually formed a replicable methodology. Initially only one colleague could make high-view videos through personal ability. Later we hired over ten people, all of whom could master this method through training — most could produce content with hundreds of millions of views per year.
Xing Meng: You've broken down problems extremely finely and can polish something to the extreme. Whether it's AI itself or your distribution, every link goes very deep. My biggest feeling from talking with you is: when we look at AI applications, we often over-focus on engineering implementation while neglecting growth implementation. But in fact, the decomposition of growth can be just as complex and just as deep as engineering decomposition.
How to Find People Who Can "Run Together"
Xing Meng: Xie Yang, you've been living in the US for a while now, formally facing your real terminal market. What new feelings have you had in the past month or two?
Xie Yang: The most attractive aspects here are mainly two: the talent itself, and the overall entrepreneurial environment and ecosystem. If I had to add a third, the climate here is indeed very pleasant.
From a work perspective, the "intellectual density" here meets expectations. Because we're building a browser, I need to constantly think about why browsers still matter, what the next paradigm of human-computer interaction is, how people will consume information in the future, how they'll interact with Agents, and so on. Here I can meet large numbers of people researching these questions — professors, PhDs, technical staff from Google and Meta. They're not just thinking about these questions, but pushing standardized practices forward. You can feel very strong creativity. Many of your ideas can find people exploring the same things. This excites me greatly.
Of course there are also things different from what I originally imagined. Silicon Valley's product capabilities aren't as strong as imagined. But its real advantage lies in continuously attracting top talent from around the world. Silicon Valley isn't really America's Silicon Valley — it's the world's Silicon Valley. And here, I also feel Chinese people are extremely strong. You can find Chinese peers deeply researching almost any frontier direction, and in many fields, Chinese technical capabilities and research depth show very strong competitiveness.
Xing Meng: I want to ask one last question. The three of you have very different backgrounds and growth experiences. When building your respective teams, do you tend to look for "similar" people like yourselves, or do you prefer to find complementary people? If you look for "similar" people, how do you judge and find them?
Xie Yang: Overall, I prioritize finding people with similar values and thinking patterns, because execution becomes faster and strategic alignment costs are lower. But a team composed entirely of "similar" people also brings risks, so I deliberately add one or two voices that differ to break uniformity. In terms of selection criteria, my requirements are quite multi-dimensional, and I hope most standards are met simultaneously. In practice this works well, because strict standards genuinely improve team talent density.
Qin Tian: One of our biggest lessons over the years is discovering that regardless of school tier, if someone has ever gotten first place in a highly competitive domain, they tend to be excellent talent. This "first place" doesn't strongly correlate with whether the school is elite. Some people may have failed the gaokao, but maintained first place in their major throughout university, or reached the top 1% in some niche competition or research direction. From our actual experience, many such people perform comprehensively better at work than some with more privileged backgrounds. So when hiring, we deliberately look for this "first-place temperament." Practice has proven that this type of talent often brings us the biggest surprises.
Giveaway
We welcome you to share your thoughts or related perspectives in the comments. We'll select 2 featured comments to receive a gift prepared by 5Y Capital :)
5Y Capital seeks out, supports, and inspires lonely entrepreneurs, providing support from spirit to all operational matters. We believe that if the "crazy" you in others' eyes begins to be believed in, the world will become refreshingly different.
BEIJING · SHANGHAI · SHENZHEN · HONG KONG
