Welcome to Deep Dive Panda! We've spent the week decoding the most important conversations from over 27 hours of Chinese tech, business, and AI podcasts.

The AI boom isn't really about tech. It's about the rising value of everything AI can't do. The most important chats in Chinese tech right now aren't about building bigger models. They're about building defenses in the messy, human world that code just can't touch.

The Unifying Theme

A weird thing is happening across the Chinese tech world. While everyone's talking about AI, the real strategic conversation has already moved on. The big idea is this: as AI models become a cheap utility, like electricity, just being good at the tech stuff doesn't matter as much. The moats that will last are being built with sweat and relationships, not code.

Even AI founders are the first to admit it. They're saying things like "user loyalty is zero" for AI models and "code is actually cheaper than talk." Trying to build a slightly better model is a fool's errand for a startup. Your edge will just get erased by the next big release from OpenAI. The real game has shifted from how to build AI to what problems we should even be solving. The bottleneck isn't execution anymore, it's insight.

This is where the less flashy businesses show us the way.

The founder of a liquor brand that sells 400 million bottles a year isn't worried about his tech. His moat is a network of one million offline stores and the hard-won respect of local distributors. These distributors are like gods in their own towns. The guy exporting cars to Africa doesn't win with a better app. He wins by building a team of expert mechanics in Shanghai and handling the logistical nightmare of getting things done in Nigeria. That's the kind of hard stuff competitors won't touch. These moats are built on complexity and human trust, and AI can't do a thing about them.

Why It Matters

If you're a founder in the West, this is a wake-up call. If your only edge is a clever prompt or a thin layer on top of a big AI model, you're building on land you don't own.

You need to ask yourself a new question. It's not "What's my tech advantage?" anymore. It's "What part of my business is hardest for a solo developer with GPT-5 to copy over a weekend?" The answer probably isn't in your code. It's in your unique sales process, your loyal community, or your willingness to do the unscalable, "dirty work" that algorithms can't.

For investors, this changes the game. A team's tech skills are now just the entry fee, not the winning ticket. The better questions are about their grit and their plan to reach customers. Can they build a brand people actually trust? Can they manage a messy supply chain? Do they get the unwritten rules of their industry? In a world of infinite, cheap code, the big returns will go to people who master the expensive, messy, human parts of building a company.

Key Patterns & Strategic Takeaways

  • Pattern: AI is cheap. Context is everything. Getting a smart answer from an AI is easy now. The hard part is giving it the right context, like project files or unspoken team rules. "Context engineering" is the new skill. It makes humans the curators and strategists, not just the producers.

  • Pattern: Your best defense is offline and hard to scale. The businesses that last have moats you can't just growth-hack. It's a liquor brand's network of small-town distributors you have to win over face-to-face. It’s a car company’s willingness to operate in tough markets. This old-school "dirty work" is a surprisingly strong defense.

  • Debate: Is it better to be fast or to be untouchable? One VC argues that with AI, "speed is the only thing that counts" because any tech lead disappears overnight. But others say it's better to run a slow marathon, building a brand and community over a decade. It's a real tension: Do you try to outrun the wave, or do you build something that isn't in its path?

  • Pattern: Your brand is who you hang out with. Sam's Club in China learned a tough lesson. By partnering with cheap, low-status local brands, they weakened their own. It showed that "who you stand with" defines you. Even for giants, your reputation is fragile and often more valuable than your supply chain.

  • Takeaway: The 'free' business model is dying. A tech billionaire, Zhou Hongyi, pointed out that every AI query costs real money. So the classic internet model of offering a free service to everyone is breaking down. This forces us back to basics. You have to build something so good that people will actually pay you for it.

Pivotal Quotes

❝

"In the AI era, user loyalty is zero. People can switch at any time; there's no loyalty at all."

Zi, Host of 枫言枫语
❝

"What kind of brand you are in the consumer's mind basically depends on who you stand with, and who is willing to stand with you."

Huang Hai, Fengtouquan (Crazy Capital Circle)
❝

"Talk is cheap, and code is actually cheaper."

Ren Chuan, Co-founder of Palona AI
❝

"It's impossible for a large model to enslave humanity. It's like a brain in a vat—it can reason and talk, but it has no hands or feet."

Zhou Hongyi, Founder & CEO of Qihoo 360

This week's Deep Dive Panda covers 19 specific podcast episodes, including:

Here are the key lessons & mental models from each episode:

1. Zhang Xiaojun Jùn | Business Interviews - 115. A 3-hour interview with Yao Shunyú of OpenAI: 6 years of Agent research, humans and systems, the devouring boundary, and a world that is both unipolar and pluralistic

Original: 张小珺Jùn｜商业访谈录 — 115. 对OpenAI姚顺雨3小时访谈：6年Agent研究、人与系统、吞噬的边界、既单极又多元的世界

  • The Game Isn't 'How,' It's 'What': The AI race isn't about building bigger models anymore. That was the first half. Now, it's all about finding the right problems to solve. We've got these incredible tools, so the big question is: what should we aim them at?

  • Don't Build a Better Chatbot, Build a New Interface: Big companies get stuck on what's already working, like trying to build the perfect chatbot. That focus creates huge blind spots. The real opportunity for startups is to create completely new ways to interact with AI, even if they seem a little weird at first.

  • Intelligence is Useless Without Context: An AI's intelligence doesn't matter much without context. The real roadblock isn't brainpower; it's that agents can't pick up on the unwritten rules of how your team works. Whoever solves this "context problem" is going to unlock a ton of value.

  • Language is the OS for Thinking: Language models are so powerful because language itself is our original operating system for thinking. It’s how we take an idea from one place and apply it somewhere else. These models are just tapping into the way we've reasoned for thousands of years.

  • Reward the Result, Not the Process: If you want to build agents that actually get things done, reward the final outcome, not the effort. Define success with a simple, clear goal, like "the code has to run and pass the test." Otherwise, you'll just end up with AIs that are great at looking busy but don't solve the real problem.

❝

"[The bet is on] having different super app product forms, with different interaction methods. If you don't believe this, then the world becomes very bleak—only OpenAI or Anthropic have a chance. But if you believe this, there will be many new opportunities."

Yao Shunyue, Researcher at OpenAI
logo

Subscribe to Deep Dive Panda to unlock the rest.

Become a paying subscriber of Deep Dive Panda to get access to this post and other subscriber-only content.

Upgrade to paid

Keep Reading

No posts found

© 2025 Deep Dive Panda.

Privacy policy

Terms of use

Powered by beehiiv