AI Is Here — Play the Infinite Game of Entrepreneurship | A Conversation with Cong Guangle, Founding Member of Toutiao and Founder of Piaoquan Video

十字路口Crossing十字路口Crossing··0·0

A Successful Entrepreneur and His Three New AI Bets

A serial entrepreneur and his three new AI "bets."

👦🏻 Podcast interview: Koji, Ronghui

🥷 Edited by: Starry

🧑‍🎨 Layout: NCon

This week on Crossing, we invited Cong Guangle — a seasoned yet remarkably low-profile serial entrepreneur — to discuss the three wildly ambitious new AI directions he's currently exploring.

He joined 99fang, founded by Yiming Zhang, in 2011, witnessing firsthand the early embryonic form of ByteDance's product methodology. But on the eve of Toutiao's launch, he chose to pivot into the mobile internet wave, hoping to "blaze his own trail." He went on to co-found Red Live and the community app Zuiyou ("The Right"). His most recent venture was Piaoquan Video, which achieved revenue scale "that would make everyone envious or even jealous" within the WeChat ecosystem.

Today, facing the AI wave, he sees mobile internet as a five-to-six-year "Age of Exploration" (a finite game), while AI is a "productivity revolution" like electricity (an infinite game). And he is now placing bets on three AI directions that look like "infinite games."

In this episode, you'll hear Guangle share the product philosophy he learned at ByteDance, his reflections on community products, and his thinking on the three AI directions he's currently exploring.

Listen on WeChat:

Listen on Xiaoyuzhou:

As the full interview is quite long (12,230 Chinese characters), here's a table of contents for reference:

🟢 Lightning Round: Age, entrepreneurial experience, MBTI, one-sentence company description

🟢 Debriefing 99fang: Yiming Zhang's "App Factory" and ByteDance Methodology

  • In 2011, Yiming Zhang was already using an "exhaustive" app factory logic to explore mobile at 99fang.
  • The embryonic ByteDance product philosophy: individual intuition, logic, imagination.
  • ByteDance's core early capability: thoroughly understanding and leveraging data, migrating PC internet information to mobile.
  • In retrospect, was ByteDance's success inevitable?

🟢 Crossroads: Why Didn't You Join Yiming Zhang to Build Toutiao?

"I remember Yiming once came to talk to me about joining him to work on the Toutiao direction."

  • Yiming Zhang at the time: calm, extraordinarily courageous, extremely hardworking — someone with a pure pursuit of creating value.
  • Why leave? Partly academic conflicts, partly wanting to learn "business-level" methodology.

🟢 From Red Live to Zuiyou: Could China's Discord Be Inside WeChat?

"Almost all mobile internet communities in China are either very vertical, or when they try to scale, they become content platforms. But overseas, it's not like that."

  • Core disagreement that led to leaving Zuiyou: build a high-efficiency "content platform," or persist in exploring "community" itself?
  • Why doesn't domestic community product logic work? First, WeChat groups already satisfy 70% of community needs.
  • Second, startups have business metrics and can't govern by "non-action" like WeChat, giving communities enough decentralized space to grow.

🟢 The Other Side of Piaoquan Video: From WeChat Ecosystem to AI Conviction

  • 2018 entrepreneurial judgment: "No more opportunities in mobile internet" — the three universal needs (people-information, people-people, people-services) were already solved.
  • Why do video in the WeChat ecosystem? At the time WeChat had no Channels; video posts were like "files" — we wanted to provide a content production and storage tool.
  • In 2020, when WeChat Channels launched, we realized this was no longer our grandest vision.
  • Why start looking at AI in the GPT-2 era? — "The video supply side can finally change."

🟢 Guangle's Three AI Bets: "I Want to Play Infinite Games"

"This is also a methodology for entrepreneurial choice... The problems I might like are more like WeChat — he only thinks about what communication is at the most fundamental level."

  • Bet One: Build a creator-replacing AIGC system, covering the full chain from inspiration, topic selection, scripting to production and even monetization.
  • Bet Two: A true "AI App Factory," where AI automates product initiation, PRD, code implementation, and release.
  • Bet Three: AI-empowered community. The mobile era was interaction defining products (e.g., Uber); the AI era is giving generative capabilities to organizers, achieving a "non-3D metaverse."

🟢 Ultimate Comparison: AI Wave vs. Mobile Internet Wave

  • Mobile internet is "Age of Exploration": discovering new continents, "quantitative change leading to qualitative change," a "finite opportunity" that ends in 5-6 years.
  • AI is a "productivity revolution": like the Industrial Revolution and Electrical Revolution, "qualitative change triggering quantitative change," with an infinite transformation cycle.
  • The biggest difference: mobile internet is a 60-to-90 problem; AI is a 0-to-1 problem.

🟢 AI-Era Organization Management and Category-B Opportunities

  • When talented people want to build "one-person companies," how do you attract them? — Find people interested in "invention" and "fundamental transformation."
  • How to manage exploratory AI teams? Provide "non-deterministic goal orientation," even measuring by "speed of learning and trial-and-error."
  • Where are the remaining $1 billion opportunities? Large model infrastructure, tools, industry-level "supply-side replacement solutions," and new platforms from new interactions.

🟢 AI Entrepreneurship Traps and Moats

"If you can't clearly think through your moat, it's probably meaningless."

  • Cold water: which directions may be meaningless? "Patching" large models; and simple tools based on individual intuition.
  • "Speed" is not a moat — it's "too linear."
  • Real moats are: data, user stickiness, complexity from deep domain integration, and user feedback flywheels.
  • Today's AI era may resemble the "software era" more than the "internet era."

👦🏻 Koji

This week's Crossing guest is Cong Guangle. Two weeks ago, Guangle and I had seafood hot pot in Beijing. After that meal, one sentence kept floating in my mind: "China's internet scene really is full of hidden dragons and crouching tigers."

The last time I saw Guangle was seven or eight years ago, at a restaurant near Beihang University. Back then he had just started his venture "Piaoquan Long Video," having only raised funding from Capital Today and Atypical — his first time as CEO. I remember Guangle was rubbing his fists together, eager to go. So many years have flashed by. I'm not a Piaoquan Long Video user myself; his target users are aunties and uncles.

Piaoquan Long Video and Guangle have always kept a low profile, but I've kept hearing about them — their user numbers, their revenue, all very strong. It wasn't until that seafood hot pot two weeks ago that I learned: not just strong, but "stronger than strong." Guangle made me promise to keep confidential exactly how good the numbers are, and of course I will. All I can say is, their annual profit is a figure that would make everyone envious or even jealous.

Very happy to have Guangle with us today. In 2011, he joined 99fang, founded by Yiming Zhang — at a time when Yiming Zhang was still one of many entrepreneurs, and Guangle worked closely with him. He didn't join ByteDance, instead co-founding a live-streaming product called "Red Live." His second venture was as co-founder of the community product Zuiyou. Then came Piaoquan Long Video, where as CEO he built a mini-program product in the WeChat ecosystem and achieved excellent results.

Today we'll chat with Guangle about his product insights from the mobile internet era, what he observed and learned from Yiming Zhang and ByteDance's product logic, how he analyzed and decided at each crossroads. And most importantly — his thinking and practice on AI today.

👦🏻 Cong Guangle

Great, thanks Koji.

Lightning Round

👩🏻 Ronghui

Let's start with a lightning round with Guangle. Age?

👦🏻 Cong Guangle

👩🏻 Ronghui

What year of entrepreneurship is this for you?

👦🏻 Cong Guangle

Year 12 of entrepreneurship, year 7 of Piaoquan.

👩🏻 Ronghui

What did you do before Piaoquan?

👦🏻 Cong Guangle

My entrepreneurial experience is quite diverse. Beyond what Koji mentioned, in the earliest days of mobile internet I also built many personal utility products. Some performed well but couldn't achieve scale or sustainable monetization. After that I did live streaming, community, and video platforms within WeChat, and in recent years also explored Web3 and metaverse-related directions.

👩🏻 Ronghui

One sentence describing your current company and product.

👦🏻 Cong Guangle

The company's core business has two parts: video platform within WeChat, and innovation exploration in AI directions.

👩🏻 Ronghui

Your MBTI and zodiac sign?

👦🏻 Cong Guangle

MBTI is INTP, sometimes INTJ.

👦🏻 Koji

Zodiac sign?

👦🏻 Cong Guangle

Sagittarius.

👩🏻 Ronghui

Your company's revenue and profit?

👦🏻 Cong Guangle

Not convenient to disclose for now. Team size is 50–100 people.

👩🏻 Ronghui

That's a pretty wide range.

👦🏻 Cong Guangle

It's manageable because the structure is fairly flexible. The core team is split between Beijing and Changsha — roughly 50–60 people in Beijing, 30–40 in Changsha.

Debriefing 99Fang: Yiming Zhang's "App Factory" and the ByteDance Methodology

👩🏻 Ronghui

Today let's start with an early mobile internet experience that was important for Guang Le before his ten-year entrepreneurship cycle — his time at 99Fang. I remember you said something on a podcast that really stuck with me: "I hope to walk a path of my own." Looking back at your journey, that certainly rings true. I recall you mentioned joining 99Fang in your junior year. Do you still remember why you joined?

👦🏻 Cong Guangle

I was fascinated by new things at the time. Mobile internet was just emerging, and my lab work and thesis projects at school were all mobile-related. We were taking photos on campus phones, using image recognition algorithms to identify buildings, then doing navigation and introductions. That direction was quite innovative back then — combining algorithmic technology with mobile device interaction. Also, the internship location was close to school, which was convenient.

👩🏻 Ronghui

Who did you meet during interviews? I think you said you only remembered Yiming Zhang much later.

👦🏻 Cong Guangle

Right. What impressed me most about the interviews were Li Fei and Huang He. Huang He came from Baidu as an algorithms lead and was in charge of the mobile team at 99Fang. Since I was interviewing for an engineering role, the questions were partly about mathematical logic and modeling, and partly about understanding mobile platforms — whether you were curious about new technologies and willing to go deep.

👩🏻 Ronghui

So what did you mainly do at 99Fang? Did it really do a lot around real estate rentals and sales?

👦🏻 Cong Guangle

Yes. From what I heard later, 99Fang's business originated from Kuxun's real estate channel in the PC era. Kuxun did vertical search but its overall development was mediocre, while real estate demand was surging and industry informatization was lacking — so Yiming started 99Fang. He went very deep into mobile exploration, building out the full transaction product suite:掌上租房 (Palm Renting), 掌上新房 (Palm New Homes), 掌上买房 (Palm Home Buying), covering second-hand homes, new homes, and rentals — these products were organized in highly similar ways.

At the same time, he explored many non-transactional product forms: real estate news, property viewing diaries, agent lookups — one content-oriented, one community-oriented, one tool-oriented. Very strong innovation capability. He didn't just deepen an existing business line, but combined temporal and technological variables to explore different product forms and the value they could create.

👩🏻 Ronghui

Like an app factory?

👦🏻 Cong Guangle

Exactly, kind of an "exhaustive" logic. He would build out every opportunity he saw, then use experimentation and feedback to judge which ones to deepen. ByteDance was very similar in its early days.

👦🏻 Koji

Looking back now, how do you think that 99Fang experience influenced your later entrepreneurship?

👦🏻 Cong Guangle

The most crucial lesson: different people should do what they're best at. For example, if Yiming had continued in real estate, he couldn't have fundamentally solved the industry's deep problems — intermediaries, service experience, offline new home supply chains, etc. What he could do was mostly the improvement from informatization. But when he moved into content, he could accomplish so much more. Content formats, models, interaction methods, supply-side transformation, and content entering various commercial scenarios — all offered massive room. So people with domain expertise are suited to solve deep problems in that domain, while tech talent is better at capturing large opportunities and universal problems, solving them through innovation.

Second takeaway: even working in a vertical direction, this team aimed extremely high. They would think about transformations overseas, look squarely at successful products, and extract abilities that could be emulated and trained.

At the implementation level, ByteDance's product philosophy was also important. Product defines the problem; engineering solves it. ByteDance's product methodology can be summarized in three layers: intuition, logic, and imagination.

Intuition is personal insight — what problems in life are worth solving, which ones haven't been solved well yet.

Logic is the analytical layer — like a ladder, using competitive analysis, old-vs-new experience differences, migration costs, and other methods to judge whether a direction is worth pursuing and how.

Imagination is the ceiling — when you can think of many extreme cases, you can build disruptive new capabilities that solve old problems.

At the same time, ByteDance was very different from most mobile internet companies of that era, including 99Fang. Many mobile product innovations focused on interaction, but ByteDance fully understood and leveraged data. For example, it had macro data from crawling the entire web, and could harness enormous power under centralized big data volumes — something many mobile teams back then wouldn't even consider.

👩🏻 Ronghui

What do you mean by fully understanding data? Can you be specific?

👦🏻 Cong Guangle

For example, when doing mobile content products, others' first reaction would be how to get users to publish and consume. But ByteDance would first crawl all PC internet information, leveraging the momentum of PC-to-mobile migration. It would optimize many PC web pages into mobile-browsable formats, doing supply-side migration first to release that potential energy, solving the chicken-and-egg problem of supply and consumption. Later, the exploration of recommendation algorithms was also based on the matching problem between production and consumption. In the past, matching relied on search, subscriptions, and categorization — ByteDance wondered whether there could be breakthrough solutions under macro data, namely personalized recommendation.

👦🏻 Koji

Looking back, do you think ByteDance's development to today was inevitable? Or besides methodology, were there external environmental factors?

👦🏻 Cong Guangle

I think it was quite inevitable. Of course it needed the "right timing" — like the PC-to-mobile internet transformation. The mobile era brought more users online, more intimate device form factors, and massive sensor data collection capabilities. ByteDance leveraged these transformations extremely well — achieving excellence in product definition, interaction implementation, supply-demand matching efficiency, and more.

Its success was both inevitable and solid. First solving content supply through crawling, then solving distribution through multiple products. Funny囧途, Neihan Duanzi, Morning & Evening Must-Reads, Pithy Quotes... first breaking down demand into different products to serve individually. At a certain stage, then building unified products like Toutiao.

At the same time, it didn't overlook mobile internet's ultimate content transformation — mobile-native short video. Before Douyin, ByteDance had already tried many short video experiments: Volcano Engine, Xigua, Toutiao Video, and even earlier in 2012 with "Tonight's Must-Watch Videos." It always knew video was a more native content form for mobile, and that UGC would eventually become widespread.

The Crossroads: Why Didn't You Follow Yiming Zhang to Build Toutiao?

👦🏻 Koji

I remember around 2015–16, I met Yiming at Guomao once. He mentioned that month Toutiao's VV (Video View) had exceeded PV, so he felt the next two to three years needed to be all-in on short video. VV is video play count, PV is news page view count. That timing left a deep impression on me.

I'm curious — standing at that junction from 99Fang to Toutiao, what considerations led you not to continue with Yiming on Toutiao, but to start your own business instead? What were your thoughts and considerations at the time? Can you walk us through it?

👦🏻 Cong Guangle

I remember Yiming talked to me once, saying he hoped we could build Toutiao together. Toutiao's core was the mobile team and the algorithms/backend team, because these two groups could be reused. 99Fang's operations and marketing people in the business domain essentially couldn't participate.

I actually participated for a few months, doing some different products from zero during the incubation phase, and ultimately didn't join. Looking back now, my memory may not be completely accurate, but I can summarize a few points.

One aspect was my graduation thesis. Toutiao's early pace was extremely intense — half a day off per week — and my school schedule conflicted with this. Another aspect was that ByteDance's way of thinking was still quite similar to my classmates around me. At that time, 99Fang hadn't completely shut down either; they'd brought in some Anjuke executives to take over. After interacting with them, I found their business thinking was more substantial, and I wanted to learn about the business side and how to operate vertical domains.

👩🏻 Ronghui

What was Yiming Zhang's state at that time? When he talked about Toutiao, did he show the ByteDance temperament he later became known for?

👦🏻 Cong Guangle

I think truly great ideas or relatively deep thinking don't necessarily manifest outwardly very obviously. He was calm, but had tremendous courage and very deep accumulation.

For example, he would think about overseas products like Pinterest and Instagram, or PC-era PUGC products, and mobile-era native content or social products, judging that similar directions might emerge in China too. But his solution path wouldn't be completely copied. He bought the domain feifei.com very early, wanting to do Pinterest.

At that time, he clearly saw the direction as "content," but the means would combine with team DNA, doing different scenario apps through crawling. It was hard to see the complete vision in that era's context, but in retrospect — including what he wrote on a napkin for Wang Qiong about content modality analysis, core content problems, matching problems, content production problems — he probably defined a direction clearly with a few horizontal and vertical strokes. Strategically concise and grand, fundamentally strong; tactically implementable, matching team DNA and conditions at the time.

Actually, I didn't have much contact with Yiming — just occasional chats during meetings or meals. I had the most contact with mobile team classmates. But I think he was extremely hardworking. I mentioned in a previous podcast that during our team-building trip to Zhangbei, his back pain was so severe he couldn't sleep. He wasn't very old at the time, but he invested so much.

At the same time, I think he's someone with a pure pursuit of creating value. He could do many domains, reach what ordinary people consider goals and returns at a certain stage, but he would always pursue the next stage, the next problem.

From Redpoint to Zuiyou: Could China's Discord Be Inside WeChat?

👩🏻 Ronghui

Let's talk about the opportunity that led to your first choice to do Redpoint Live. There weren't many platforms in the live streaming industry at the time.

👦🏻 Koji

Redpoint was among the very earliest live streaming startups.

👦🏻 Cong Guangle

Right. We started in August 2013. At that time, some overseas game live streaming platforms were exiting, while domestically there were still somewhat known products like Fengyun Live. Later platforms like HUYA Inc. rose, which all proved one thing: live streaming is essentially a modality, a real-time information transmission method.

This form can extend into many application scenarios — it can be tool-based SaaS, used for classes, meetings, sharing, or vertical platforms like showrooms, game streamers, e-commerce live selling, etc.

Live streaming as a form truly matured in the mobile internet era, because user devices first gained audio and video capture capabilities. China's PC era didn't have this environment: most people didn't have webcams or microphones, so live streaming was novel then, and also a genuine interaction innovation.

👦🏻 Koji

From Red to Zuiyou to Piaoquan, you've actually been doing PUGC the whole time. After finishing Red, why did you choose to build Zuiyou?

👦🏻 Cong Guangle

Red worked as a tool-based product, but I hadn't figured out which domains SaaS should really dig into: serving education? Pure communication? Each direction required massive iteration. So I left Red and started exploring UGC in a broader sense.

When building Zuiyou, I was thinking: how do you promote UGC? There are two approaches — providing stronger production tools, or providing culturally affiliated groups where content emerges naturally through interaction.

The first approach already had short video apps, Meipai, Kuaishou, but they weren't doing enough to promote UGC. So I thought of the second approach: bringing together groups with unique culture and identity, providing rich interactive components — bullet comments, voice messages, image editing, polls, chain stories, and mechanisms like "god comments" that carried social incentives.

Zuiyou's innovations focused on three things:

  1. Shifting from "search to find groups" to "recommendation-driven distribution and connection";
  2. More systematic organization and management of community groups;
  3. Richer multimodal structures for content publishing and interaction.

So at that time, the cultural atmosphere and group creativity were both very vibrant.

👩🏻 Ronghui

Then why did you later leave Zuiyou to build Piaoquan?

👦🏻 Cong Guangle

This involves a fundamental divergence: in China's mobile internet, communities either become extremely vertical, or get flattened into content platforms. But overseas isn't like this. Reddit, Discord — they can maintain the state of "community as community" long-term without needing to become information distribution platforms.

Some partners wanted Zuiyou to be lighter, thinner, more like a content platform, but community organizational structures are complex and can't run at the efficiency of content platforms. Content platforms rely on information distribution efficiency; communities rely on relationship and interaction density. These two directions aren't compatible. What I always wanted to explore was "communication and production," not "content consumption."

👩🏻 Ronghui

So communities need patience to let them grow naturally?

👦🏻 Cong Guangle

Yes. And one more thing: digital-world communities don't equal real-world communities. The United States is a multi-ethnic, multi-cultural society — they have natural needs for information partitioning, cultural partitioning. Recommendation systems can't solve these problems. Only by bringing similar people together can real discussion happen.

And to make communities produce more customized interactions in the digital world, companies need to invest heavily in "scenario-level components" — check-ins, reminders, logging... these are all highly customized. Without customization, communities slide toward an "information production — information interaction" structure, becoming similar to content platforms.

👩🏻 Ronghui

Then why have Reddit and Discord maintained vitality long-term? What did they do right?

👦🏻 Cong Guangle

First, it's determined by social structure — Western cultural and group differences create needs for partitioned community structures rather than unified recommendation feeds. Second, they've consistently used decentralized approaches, letting communities grow naturally. Look at the various groups in WeChat — it's actually similar logic. It's not WeChat officially operating them, but users operating them themselves, which instead creates an unimaginably rich ecosystem.

👩🏻 Ronghui

So WeChat in this sense also learned from this model?

👦🏻 Cong Guangle

Right, it respects decentralized group growth rather than using centralized product will to guide things. If you drive communities with "retention" and "scale," it's hard to let ecosystems grow naturally. Discord is the same — its growth doesn't come from company operations, but from countless independently operated groups.

👩🏻 Ronghui

But this logic doesn't seem to apply to independent communities domestically?

👦🏻 Cong Guangle

First, most community behavior in China already happens inside WeChat. School clubs, KOL circles, industry groups... WeChat already satisfies 70% of community needs. Second, domestic companies don't have that much patience to build an ecosystem-type community. WeChat can accommodate this growth because it didn't carry commercial metrics early on. But startup companies wanting to build communities can't "govern by non-action" — yet once commercialization enters, it conflicts with group ecology, and there's also no profit distribution mechanism.

So truly decentralized, multi-ecosystem communities are hard to accommodate early on under "agile iteration" and "metrics pressure."

The B-Side of Piaoquan Video: From WeChat Ecosystem to AI Conviction

👦🏻 Koji

Actually, we just mentioned WeChat "Piaoquan long video" — this is a company that grew within the WeChat ecosystem. When you decided to start this business, what opportunity did you see? Can you look back on it?

👦🏻 Cong Guangle

More because I believed there weren't many opportunities left in mobile internet. Mobile internet's problems had been solved over 95%. If we don't talk about underlying operating systems, infrastructure directions, or some specific industry-level directions, and just look at universal consumer demand, it can be divided into several categories: people and information/content, people and social, people and goods/services.

You'll find: for people and social, there's WeChat; for people and goods/services, beyond Alibaba and JD.com, service-type industries like Meituan and DiDi emerged; for people and content, ByteDance made Douyin, innovating from supply to consumption to content formats. Overall, there were no obvious new opportunities in mobile internet's major directions.

When we were doing Red, we noticed many people used Red within WeChat scenarios to fulfill listening needs. WeChat's network is very diverse, with lots of internal space. Every capability WeChat adds brings industry changes. For example, after adding red packets and monetary capabilities, massive transactions emerged inside WeChat; without providing this capability, many things couldn't happen. We thought at the time: could we, as a startup, provide some capabilities for WeChat, capabilities that could create new scenarios and new value satisfaction inside WeChat?

At that time WeChat didn't have WeChat Channels yet — video existed in a file-like way, unable to achieve centralized storage, comments, datafication, creator revenue, and other functions. We believed we could provide content production and content storage tools for it, and found demand was strong. Later we further developed platformization and distribution capabilities.

👦🏻 Koji

I know you've done a lot of AI-related exploration, and we're very much looking forward to discussing that today. But before that, I want to ask one more question about "Piaoquan long video." It was actually very successful, but just now you only talked about the opportunity you saw at the time. Can you talk about what key things you feel you "did right" during the process? Were there some key milestones?

👦🏻 Cong Guangle

I think first, choosing the right network scenario. WeChat is a social network — it's different from mobile networks, but it fits the underlying definition of networks, while not being a very concrete application-layer product. Second, timing. Our mini program launched in 2017, and we quickly made related products in 2018. The entire practice process was relatively smooth: start with tools, then radiate to some scenarios, then build platform, then build recommendation algorithms oriented toward distribution, plus commercialization.

Compared to early mobile internet community platforms, short video platforms, and other major tracks, this direction wasn't that complex, so the effect and accuracy of each step would also be higher.

👩🏻 Ronghui

You mentioned earlier that you started paying attention when GPT-2 came out. Can you talk about why you were following GPT-2 at that time? And after ChatGPT came out, how did your judgment change?

👦🏻 Cong Guangle

Piaoquan started in 2018, and around 2020 WeChat Channels appeared. We realized we probably couldn't carry WeChat Channels' mission within WeChat, and felt it no longer matched the maximum vision we were pursuing, so we started looking in different directions. We paid attention to AI because at that time GPT, and DALL·E concepts, had just emerged — around 2020.

We believed the supply side of video could finally change. The original supply side came from shooting and editing; if generative production became possible, it would be very different. Past generation leaned more toward modeling, CG, film-level technology, while language intelligence improvement let models process logic and information organization; generative capabilities from text to image, image to video, also had development potential.

So we made a product偏向 video creation: input text, use NLP to search and match images or video materials, then generate complete video, plus TTS. Whether explanatory, expressive, or content-display types, it was completely different from shooting and editing. But limited by technology at the time — the direction was right, the trend was right, but productivity wasn't at the level of the post-GPT era, and there were no products like Midjourney, so what could be done was limited. Although the product attracted tens of thousands of creators, it didn't reach the "variable-level effect" we hoped for.

Guangle's Three AI Bets: "I Want to Play Infinite Games"

👩🏻 Ronghui

What attempts have you made in AI? I understand there are two aspects. One is adding AI capabilities to your existing products; the other is starting completely from scratch to make something fully AI.

👦🏻 Cong Guangle

Actually Piaoquan's explorations weren't that many. The new exploration is to fully understand content creators themselves, including creative methods, creative elements, and scheduling tool capabilities, then build an AIGC system that replaces creators. We'd rather evaluate it as "a system." This system's core steps include decomposing creative processes, creative elements, and creative modes, then giving different weights to these means based on user preference, and establishing combinatorial relationships between means and elements.

This way you can define an account: what kind of character is it? What are its unchanging elements? What are its core creative methods and core creative domains? Based on this definition, it can generate content, and the generated content will likely conform to a point-line-plane-body expansion: from the point of inspiration, to the line of topic selection, to the plane of script, to the丰满 entity after production realization.

👩🏻 Ronghui

So video generation models are just a starting point?

👦🏻 Cong Guangle

I think they're one环节, like CG technology's position in the film and television industry. The film and television industry divides into creation and production — creation goes from screenwriter to director, with the director responsible for connecting creation and production; production includes lighting, cinematography, costume and makeup, executive director, actors, etc. Production is an implementation problem; creation is a creativity and definition problem. Production requires many steps, while creativity is essentially about finding special characteristics, finding key elements that humans like, and combining and recombining these elements.

👩🏻 Ronghui

Sounds like a very grand system, because you're doing both content creation generation, and also having videos be automatically edited after generation, calling tools, packaged into content that can be directly presented. At the same time it also needs to consider a video's entire chain from production to publishing to even promotion.

👦🏻 Cong Guangle

That's right — you even have to think about monetization in advance, including commercial goals, monetization vehicles, and how to improve through reinforcement and feedback once it's running. No matter how many steps there are, it's still a finite system; no matter how difficult each step is, they can all be connected through goals, logic, knowledge, and means.

👦🏻 Koji

That reminds me of something: "If you aim for the moon, you'll at least land among the stars." I know Guangyue is currently exploring three directions. We just talked about the first one — using AI to replace human video creation. Could you share your second and third AI-related directions?

👦🏻 Cong Guangle

The reason I define the first direction this way, besides our long-term work on content platforms and UGC supply, is that we recognize the trajectory of AI development: from chat to reasoning, then to agent, instruction-following, innovation, and organization. I'm someone who enjoys innovation and embraces change, so we explore new directions based on demand across different domains on the internet, the current product implementation methods available, and needs that remain unmet.

The second direction is AI-automated product production: automated direction-setting and project initiation, automated PRD generation, automated coding and implementation, and automated release — essentially a true app factory or tool factory. ByteDance's app factory isn't actually that generalized in scope; it's concentrated in the content domain, just across different verticals. In our view, the productivity unleashed in the AI era has much greater room to expand. Combined with our understanding of entrepreneurship, product methodology, demand capture, and execution, we have the opportunity to build a complete AI app factory capability system.

👩🏻 Ronghui

Both of your plans sound quite ambitious.

👦🏻 Koji

Yes. Is the third one even more ambitious? Or more concrete?

👦🏻 Cong Guangle

The reason for these two directions is that I see AI as a long-distance race, a productivity revolution — quite different from mobile internet, which I can elaborate on later. The third direction is also quite interesting.

We just talked about some community products, so let me set up some context first. In the PC era, certain service and transaction-type products were all handled on a site called Craigslist; in the mobile era, they differentiated into vertical products like Uber and Airbnb. Why did they differentiate? Because on Craigslist, you could only post forms, display simple information, filter, and have comment-style discussions — but many functions needed further implementation, like LBS information publishing and dispatch matching. So each domain differentiated into its own standalone product. Essentially, the interaction determined how optimal the experience could be in a given domain.

What kind of interaction leads to greater optimality? When we think about directions like Zuoyou, we feel that at the information level, community interaction has already been done quite well, whether on domestic or international content platforms. But if we imbue community platforms with AI's generative capabilities, and give them to every vertical and every group organizer, what would be different? The key to community lies in group identity recognition, including DID identity verification. If through AI, everyone could access a Q&A and judgment system to determine whether they belong to a certain circle — for example, whether they're a Bilibili ACG user; or through interface programs, automatically verify via student ID and facial recognition against the CHSI database whether they belong to a certain school, ensuring they're from the same alumni group. This is identity-level implementation, and there would be far more possibilities at the functional, interaction, and content display levels.

Let me give an example. Say Koji likes birdwatching — have you ever thought: what if all birdwatchers had a product that wasn't on WeChat groups or any other platform. Every bird video and image posted would have LBS, and this map would only carry records from all birdwatchers. This is a special interaction, a special information model, a special publishing method. In the past, only when a centralized company deemed this need meaningful would they build a product to serve it; without a centralized company, massive amounts of demand simply couldn't be realized. There was no way. But with AI, organizers or operators could implement various components within a community product, achieving something like a non-3D metaverse effect.

This concept is similar to Metaverse, but doesn't rely on 3D physics engines — instead relying on interaction components, filtering systems, verification systems, etc. — allowing information exchange in many groups to generate behaviors and satisfy needs beyond just information exchange. This is something we're also thinking about seriously.

👩🏻 Ronghui

Where will you put your main resources?

👦🏻 Cong Guangle

The first direction will get seventy to eighty percent of resources; the remaining two directions will stay more at the research and thinking stage. Both directions also fit very well with the logic of the "infinite game," so we're quite interested, and they're connected to what we've done in the past.

👩🏻 Ronghui

I feel like what you're describing is quite different from many AI entrepreneurs — more like a one-stop approach.

👦🏻 Koji

I think there's a consensus in Chinese entrepreneurship: make money first. This relates to pragmatic culture and the social mood of recent years. But Guangyue offers a very different narrative today: I want to play the infinite game, I want to solve the most generalized, most universal problems with potentially infinite commercial value. This means the early stages may involve loneliness, high risk, even zero returns. But Piaoquan long-form video is currently the ticket warehouse, the granary, so you can indeed afford to be more willful in attacking bigger goals.

👦🏻 Cong Guangle

Yes, I think this is also a methodology for entrepreneurial choice. There are roughly three categories of entrepreneurial methodology:

The first is based on individual perception — discovering a problem in your own domain, profession, or environment, then切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入切入

Guangle, your thinking is really sharp. I'm curious — in this wave of generative AI entrepreneurship, which directions do you think could produce "billion-dollar" companies? Have you done any analysis on this?

👦🏻 Cong Guangle

We've thought about it, because if a direction doesn't pay off in the short term, it needs to have massive value in the "endgame." Here's how we see it:

First, the foundation models themselves and related infrastructure are extremely meaningful — they directly represent productivity gains and hold enormous long-term value.

Second, there's lots of room on the tools side: for consumers, for different professions, for different types of companies — good opportunities will emerge across all of these.

Third, industry-level AI substitution on the supply side — this didn't exist in the mobile internet era. AI could replace drivers, creators, developers, and so on. The leading opportunities in these sectors will be very large.

Finally, new platforms may emerge too, but these platforms will be built on new forms of intelligence and new interaction paradigms.

👩🏻 Ronghui

So what directions do you see right now, domestically and internationally, that seem especially valuable? And which ones feel off to you?

👦🏻 Cong Guangle

Early on I followed a lot of products, but later realized I didn't have the bandwidth — especially since our exploration scope was already quite broad. Let me use some mobile internet examples as analogies: early on there was consensus around OS-level patches like "Room," jailbreaks, and app stores, but they ultimately proved to have limited value. Today, some directions that "patch" large models or do small scene optimizations may similarly lack long-term significance.

Second, a word of caution for young entrepreneurs: in the mobile era, many "solo developer mini-tools" seemed to address needs but had very limited long-term value. Things like class schedules, subway queries, sticky notes. Many were just projections of the creator's own needs, not real value. I even built a notes tool myself, but eventually realized WeChat's file transfer assistant solved most use cases.

It wasn't until AI arrived — with summarization, information efficiency gains, and feedback capabilities — that tools truly underwent a qualitative leap. So a good direction needs to satisfy two conditions simultaneously: matching the fundamental variables of the era + matching the fundamental needs of the domain. Only by truly combining both can you have a good direction.

Traps and Moats in AI Entrepreneurship

👩🏻 Ronghui

You mentioned earlier that a lot of what Toutiao did in its early days was copying PC-era content onto mobile — first solving the supply problem, then letting native mobile product forms emerge. Now we're in a similar transition: from "what can you do on a phone" to "what AI capabilities can you add to a phone," and eventually to truly AI-native applications. Do you buy this analogy?

👦🏻 Cong Guangle

I buy the analogy, but I'm more focused on "what's the moat." If someone thinks AI has no moats and it's just a speed game, I disagree. Because speed is linear — it doesn't match how technology actually develops. Real moats come from either data, or irreplaceability created by user dependency, or complexity from deep domain integration that makes it hard for followers to catch up.

Another important moat is the flywheel effect of user feedback. Rather than discussing "what directions are meaningless," flip it around: if you can't clearly think through a direction's moat, it's probably meaningless. Of course, if your execution is #1 in the industry, everything's negotiable, because execution is most effective in same-dimension competition, while real differentiation comes from "dimension-lifting" — from building core competencies on a different dimension altogether.

I think this AI wave resembles the software era more than necessarily the internet era. Go back to the early days of Office, WPS — the core was "using human capability to program solutions to problems." But the internet era relied more on connectivity and data centralization, where AI help is limited. Conversely, in turning problems into solutions — whether through models, agents, or code — AI excels, so the software-era logic fits better.

Many of the innovation directions we're working on are also interconnected:

One is AIGC; two is AIGI (new interactions); three is AIGI-assisted UGI, letting communities generate new forms of interaction. They're not fully independent — when you discuss "the substitution and assistance relationship between humans and AI," they all converge on modes of interaction and information generation.

👩🏻 Ronghui

Why do you analogize it to the PC era specifically?

👦🏻 Cong Guangle

Mainly because PC-era software was the stage where humans programmed solutions to problems. The internet era relied more on connectivity and data centralization — where AI can help only so much. But in "turning problems into solutions" — whether through models, agents, or code — AI's capabilities are very strong, so the software-era analogy works better.

👦🏻 Koji

Thanks for your time today, Guangle. Thanks for sharing those early ByteDance experiences and for all the AI thinking. We rarely step back from frontline practice to discuss bigger visions and possibilities. Looking back in a year or three, some of these views may be validated, others overturned — but this is a voice and entrepreneurial sample worth recording. Thank you again.

👦🏻 Cong Guangle

Thank you both.