What Is the Core of AI Software?

https://x.com/ZenoRho/status/2051291866881310760

Everyone is essentially doing the same thing: imagining and racing to capture the entry point of the AI era.

They aim to replicate the success of the mobile internet era—snatching the search entry point like Google, or capturing the attention entry point like Douyin (TikTok).

Thus, we have AI glasses, AI voice recorders, AI recording beans, AI rings, and various other devices that can record text and store language data.

For a long time, I also believed that investors were visionary, thinking they were right and agreeing deeply.

Yet, I always felt something was missing. Now, I finally understand.

The core of AI software is not the "entry point," nor anything else.

It is, and only is, Token.

Everything revolves around tokens; everything unfolds around tokens. Understanding tokens is the first step to truly grasping AI.

In the AI era, there are only two problems: creating tokens and selling tokens.

To consume tokens faster and sell them quicker, we have agents executing long-duration tasks.

To sell tokens at a higher price, we build software offering various services, thereby adding more value to tokens, allowing us—the token distributors—to sell them at a premium.

This is because others cannot achieve their goals using tokens from elsewhere; only your tokens can fulfill their objectives.

If you view AI from the perspectives of "entry points," "application scenarios," or "vertical industries," you'll find a tangled mess, incredibly complex, yet never touching the essence.

But when you shift your perspective to tokens, the industry becomes remarkably simple. The entire value chain is nothing more than creating tokens and selling tokens.

Once you think from an economic and profit standpoint, along the token production and distribution chain, everything becomes clear.

From token production to distribution, who makes money and what kind of money they make—the entire process and all industries become incredibly straightforward. You can instantly see which startup directions are doomed to fail.

This is also the most brilliant method from Mao Zedong's Selected Works. I only recently discovered the similarity between the two.

To explain it fully, please take a moment to follow my line of thought and understand how this formidable man from the Qing Dynasty era analyzed problems and found the key leverage point.

Now, put yourself in 1920s China.

1920s China was just like today's AI industry—all the smart people were analyzing China's way out, each with their own theory, but the more they analyzed, the more chaotic it became, and no one could convince anyone else.

Lu Xun said the fundamental problem was culture—the national character was flawed, servile, and the people needed to be awakened.

Kang Youwei said the fundamental problem was morality—the Three Bonds and Five Constants were lost, and Confucian rites needed to be restored.

Liang Qichao said the fundamental problem was education—the people were unenlightened, and newspapers and enlightenment were needed.

Zhang Jian said the fundamental problem was industry—there was no manufacturing, and factories needed to be built.

Sun Yat-sen said the fundamental problem was the political system—the monarchy was corrupt, and a republic was needed.

There were far more than these six perspectives and six theories back then. Each could justify itself, but each only illuminated a part, never the whole picture.

What made Mao Zedong brilliant was that his analytical dimension was on an entirely different level.

The things these six people analyzed—culture, morality, system, education—were all "superstructure." Were they important? Yes. But they all share a characteristic: remove them, and people still live.

The things these six people analyzed—culture, morality, system, education—were all "superstructure."

They share a common trait: remove them, and people still live.

Without a good system, people still need to eat. Without a new culture, people still need to eat.

Systems can collapse, Culture can fracture, Morality can decay,

But as long as people are alive, they need to eat and fight for the resources to survive.

What Mao Zedong found was the one thing that remains after stripping everything else away: economic interests.

What Lu Xun saw as "national inferiority," dig one layer deeper, and it's poverty. When people are so poor they can only focus on their next meal, they naturally become numb.

What Kang Youwei saw as "moral decay," dig one layer deeper, and it's a shift in the interest structure. The old order could no longer sustain people, so they naturally abandoned old rules.

What Sun Yat-sen saw as "system failure," dig one layer deeper, and it's still interests—warlords didn't listen to parliament not because the system was poorly designed, but because their economic interests didn't require a parliament.

These people saw symptoms; Mao Zedong saw the deepest driving force. Today, there's a trendy term called "first principles"—peeling away surface layers to find the indivisible core underneath.

Economic interest relations were that indivisible core in society at the time.

Starting from there, landlords lived off rent, so they would inevitably oppose land reform, regardless of their morality. Poor peasants owned nothing, so they would inevitably support change, regardless of their education.

Today's AI industry is exactly the same.

Some say their project is great because it captures the AI entry point—I call this the "Entry Point Theory."

Dividing AI by "where users first encounter AI," Those making glasses say glasses are the entry point, Those making earphones say earphones are the entry point.

Some say their project is great because it deeply integrates with a scenario—I call this the "Scenario Theory."

Dividing by "where it's used," Those in law say law + AI is a goldmine, Those in healthcare say healthcare + AI is a necessity.

Some say their project is great because AI empowers an industry—I call this the "Industry Theory."

The Industry Theory divides by "which industry is being transformed," with each claiming to be the leader in AI adoption for their sector.

Similarly: The "Technology Theory" divides by "whose model is stronger," with whoever has larger parameters and faster inference winning.

The "Human Nature Theory" divides by "what needs are being met," with those making AI girlfriends claiming to tap into human nature's core.

The "Platform Theory" divides by "who can become an ecosystem," with everyone wanting to be the iOS of the AI era.

The "Data Theory" divides by "who has more data," with everyone building data flywheels and data moats.

Countless ways to slice it, countless theories. Each has merit, yet none captures the whole picture. It's exactly like a hundred years ago.

The things these theories analyze—entry points, scenarios, industries, technology—are all "superstructure." Remove them, and AI still runs.

Without glasses as an entry point, AI still runs inference. Without the "education" scenario label, the computing power consumed by AI to answer a math problem remains the same.

Strip away all entry points, scenarios, and industry classifications. What's the most basic fact left?

A piece of text goes in, a piece of text comes out. Each in-and-out consumes tokens, incurs cost, and creates value. This is the indivisible core of the AI industry.

Dig one layer deeper into the Entry Point Theory, and it's about competing for token consumption channels. Dig one layer deeper into the Scenario Theory, and it's about competing for token added value—the same token is worth a penny for casual chat, a dollar for legal advice, and ten dollars for medical diagnosis. Dig one layer deeper into the Technology Theory, and it's about competing for token production efficiency. Dig one layer deeper into the Platform Theory, and it's about competing for monopoly over token distribution. These theories all see the surface-level competition; tokens are the fundamental thing underneath.

Faced with a China of hundreds of millions of people and hundreds of contradictions, Mao Zedong found a key: land.

Who owned land and who didn't determined how a person ate, which side they took, and how they acted. Following this thread, hundreds of millions of people were divided into clear, distinct camps.

Today's AI industry, with thousands of companies and dozens of tracks, also has a key: tokens.

Who creates tokens and who sells tokens determines how a company makes money, its position, and its fate.

All AI-related companies are either token producers or token distributors, nothing more.