Recently, I've been hearing the claim that code is a liability, not an asset. After years of AI coding advancements, we still haven't seen the kind of explosive, large-scale AI-native applications emerge—outside of a few niche verticals. I believe AI represents a true phase-transition leap in the industrial dimension. In the early stages of every major phase shift in history, massive societal wealth flows toward the new technology itself, while broad, transformative applications tend to lag. This dynamic naturally creates extreme wealth polarization in the beginning. What we currently call 'demand' is mostly software demand—an artifact of the last few decades of the internet and software industry. In reality, these needs have existed for a long time, and so have their suppliers. Most of these demands aren't scarce at all; in many cases, supply already exceeds demand. A lot of AI coding efforts are built on the mistaken belief that one's own idea is uniquely brilliant. From a macro perspective, nearly all demand categories that emerged in the last few decades within software and internet are now saturated. In the coming AI era, the next wave of demand won't be more of the same human software and web needs we've grown used to. Instead, it will likely be entirely new categories—most notably, demand from AI agents themselves, rather than from humans. This explains why, even if Anthropic reaches a trillion-dollar valuation and enormous amounts of societal capital plus developer tokens are poured into AI coding, we're still not seeing a meaningful surge in high-value applications at the societal level. Unlike the Web 2.0 era that produced Facebook, Twitter, and countless other breakout social and consumer apps, the AI value chain currently stops short. It hasn't closed the loop. The greatest near-term value of AI may ironically remain AI coding itself. We haven't yet reached the final ring of the chain: using AI coding to generate a massive ecosystem of valuable applications whose total value far exceeds that of the underlying LLMs. My view is that this new class of demand—and the resulting explosion of high-value applications—will likely emerge only after a major AI capital market correction, similar to how the real Web 2.0 boom followed the 2000 dot-com bust. And the future demand won't come from traditional human software and internet use cases. It will come from AI agents' own needs—or even from the AI entities themselves (for example, LLM-to-LLM demand)—rather than from humans.