Written by: Zuo Ye

Looking back over 500 years, the labor-capital conflict under the capitalist system has always been marked by the continuous victory of capital.

On the production side, the level of labor participation is gradually shrinking to the level of operating machines; on the consumption side, the value of users lies in producing and using data for the platform.

Together, these two factors support the company's valuation in the capital market.

However, organizational models for people cannot be fully quantified in the long term. White-collar KPIs/OKRs are still bureaucratic systems, and annual salaries of millions and piece-rate wages are variations of Taylorism.

Without a clear formula, capital cannot value it, which in turn affects capital efficiency. Whether algorithmic stablecoins are the holy grail of DeFi remains to be seen, but the computability of organizations is indeed a measuring cup for financial leverage.

The large model decided to use brute force with token quantity. The collapse of security SaaS is just the tip of the iceberg. Product design is also underway. The key is to replace niche professional capabilities and scale them up . Innovation is venturing into uncharted territory.

This offers us endless insights, especially now that the DAO model in DeFi is gradually collapsing and token economics is becoming increasingly bankrupt.

In short, why are AI's organizational and token models more efficient than DeFi?

How did all of this begin?

" Tokens become cheaper, agents become more practical."

For a 300% profit, capitalists are willing to sell their noose.

To keep their current jobs, workers can write skills for agents.

At the capital level, agents backed by skills hold a status as sacred as profits.

Agent represents the refinement of "human ability" into Skill. Moreover, human organizations become interactive ritual chains centered around Agent.

From the so-called Prompt and Context to the current Harness project, the goal is to transform the human organizational model into uncharted territory, or at least reduce the number of people involved.

Your next colleague may not be a robot, but rather someone with "ability" and instinct.

This is not wishful thinking. The scaling law at the data level is gradually becoming ineffective. However, data collection and production are no longer important. Before AGI succeeds, a new valuation target is needed.

Image caption: Content is no longer valuable.

General Information: @ARKInvest

Starting with Claude's first step in implementing AGI in the programming field, AI is moving beyond the entertainment mode of chat boxes and into existing markets in real-world fields such as programming, security, and newly released design.

Whether this disruptive innovation will ultimately create new economic growth or drag the economy into a permanent low-employment model where tokens are used for jobs and people are laid off, we are witnessing this process.

However, the current devaluation of tokens, which has delegated capabilities previously monopolized by a few large enterprises to small and micro enterprises, thereby creating super individuals, is not a fantasy.

Taking China as an example, the daily token usage increased from 100 billion in 2024 to 100 trillion by the end of 2025, and is now 140 trillion. The production of content and data is about to enter an era of zero cost.

It is important to note that the shortage of computing power is a relative state. Large enterprises no longer monopolize the "capabilities," but they still want to maintain their existing advantages by monopolizing the "computing power." However, they cannot stop the inevitable trend of the overall cheapening of tokens.

There are many different paradigms for evaluating the foundational models, but the evolution of "how AI helps people" has not received much attention for a long time.

In my view, Harness is a spatial form that allows the agent to focus on the task within the boundary for the first time, using a depth-first strategy, which is different from the breadth-first approach of question-answering.

Image caption: The Evolution of Agents

Image source: @zuoyeweb3

From the moment the Tab key was first used to complete code, it was only a matter of time before humans became the input layer for AI.

The cost of trial and error is reduced exponentially, allowing for more interesting experiments with human collaboration models:

• Software: SaaS, where the source of human capabilities is no longer people, but the emergence of agents.
• Hardware: Computing power cards + HBM, data centers directly serve AI needs for the first time.
• Space: Harness is not a physical space for human collaboration, but a digital space for agent interaction.
• Interaction: The Doubao phone is dead; Google supports GUI Agents at the Android system level.

AI's ability to say things doesn't have much commercial value. The cost of generating text is very low for humans. However, "what to do" will cause token consumption to surpass that of image and video generation. It's similar to AWS selling usage time, not servers.

AI isn't selling tokens, but rather "work capabilities." This is the root of fear in the SaaS industry. Unfortunately, DeFi has become SaaS, not a big model.

SaaSification of DeFi Protocols

" DeFi is not outdated, but premature."

AI is reinventing software engineering, and it's not just SaaS that's being replaced, but SaaS is undoubtedly the most typical example.

Even for Bloomberg Terminal, its most important commercial value is not its technological advancement, but its authoritative information, which is built upon decades of non-standard data such as industry connections and personal networks.

The agent offers an option to predict the future from the data, and even if it's a risky next step, it may still outperform its competitors and earn a small profit.

Image caption: SaaS is crashing.

Image source: @zuoyeweb3

You can think of it as the agent cleverly taking advantage of the profit-seeking nature of capital. Of course, it can wait for complete Bloomberg terminal information, or it can use pieced-together, inaccurate data to try and make a profit.

This is not a new thing. Thomas Peterfee, the founder of IBKR, was the first to "invent" or assemble a physical trading terminal in the financial field, and it all started with an idle P101.

If a certain way of using data can generate more profit, then you can obtain more data, and the flywheel starts.

SaaS monopolies are a thing of the past; AI-driven sales are the future.

Unfortunately, we're about to delve into DeFi from here. Remember the API paywall of Dune/DeFiLlama, where they were begging for money with Jinshuju (a financial data platform), or the eventual closure of Arkham Exchange?

Data in the crypto industry has never been worth much.

However, the crypto industry is a direct and open financial system, and the data it generates can be repeatedly learned. Even before AI, the speed of forking projects had been reduced to months, and PumpFun's clone meme could be compressed to the second level.

There is a counterintuitive conclusion here: DeFi is a testbed for the financial system, and the AI+DeFi we are trying today will become the template for the future evolution of finance.

• For example, before the 2008 financial crisis, LIBOR, an unsecured trading instrument, "triggered" the financial tsunami. It was later replaced by SOFR, an indicator generated by US Treasury bond trading. However, the over-collateralization mechanism ensured the finality of DeFi liquidation.

• For example, large-scale manufacturers don't want to sell tokens based on consumption; they insist on tiered marketing, customized capabilities, and professional modifications. Token economics has twisted "use value" into a tangled mess.

Crypto tokens focus on use value, while AI tokens focus on economic value.

From this perspective, DeFi hacks are merely a routine stress test, representing the external entropy that open systems cannot fix on their own.

Similar to the dark humor of Catch-22, without the stimulation of external signal systems, encryption assumes the current environment is secure. However, in the event of a security crisis, it collapses into a centralized processing system.

For example, in the Drift incident, the target of people's blame became the frozen and sluggish Circle.

Image caption: Code cannot solve security problems.

Image source: @zuoyeweb3

It can be said that before the leap in AI capabilities, DeFi had already completed its SaaS transformation and could only charge based on the number of transactions, without being able to directly migrate "finance" onto the blockchain.

RWA lacks liquidity on-chain, and DeFi has no good solution to this problem.

However, the evolution of Agent capabilities seems to offer a glimmer of hope for rewriting the rules of DeFi, though the future remains uncertain.

1. Token economics: Deployment based on usage across different channels and according to "capital efficiency";

2. Rule setting: Mythos provides security finality, AI defense battle zero-day crisis;

3. Human organization: Great, DeFi has long been managed by a few people managing tens of billions.

The Revival of Engineering Narratives

" Where does safety come from? The certainty of the Turing machine. Where does danger come from? Infinite possibilities."

YC Garry Tan's saying "Fat Skill, Thin Harness" resonates deeply with me. Essentially, it's about establishing fundamental rules, a kind of "freedom based on order."

Turing machines can be combined infinitely, the von Neumann architecture always has a time lag in its in-memory computation, and large models cannot generate truly random numbers.

In a future where data is worthless, only human behavior can create value through the flow of money.

However, it will take time for human behavior to be fully learned by AI and then internalized into an engineered and coded expression.

The pursuit of the infinite with the finite will ultimately be futile. LLM cannot completely eliminate the illusion. It must approach the point where "this is beyond the reach of AI and beyond the reach of human beings" before the market mechanism can price it, and only then can we truly believe in smart contracts.

Current smart contracts are far from successful. The DAO fork, the Curve programming language bug, and even Drift multisignature all prove that "humans have ultimate control over the code."

Moral judgment has no economic value. The reason why the collaborative model in the DeFi field has collapsed from DAO to foundations and "teams" is ultimately due to the real need for contract upgrades and business cooperation.

But humans simply cannot write code that is always secure and dynamically upgradable. Remember, it is impossible.

If it is never upgraded, Curve's own experience tells us that the technology dependency stack can also have problems.

The present determines the past, and the past determines the future.

From the Simons Medallion Fund to Numerai's AI-driven strategies, AI is not uncommon in the financial field. In another counterintuitive case, trading signals can actually help AI evolve.

Image caption: AI and DeFi 10 years

Image source: @zuoyeweb3

AI models are still computer paradigms, state machines that process signals. Without external signals, they lack the ability to simulate the external world. The significance of Yang Lekun and Li Feifei's bet on world models lies in this.

However, from the perspective of DeFi, enabling AI to trade autonomously requires that human intentions be learned by the agent through behavior. This highlights the importance of humans to AI. Even if the agent replaces human labor, it is still imitating and summarizing human behavior.

Furthermore, even though humans cannot intentionally act randomly, slight deliberate actions can reveal statistical patterns. It is only human physiological characteristics that exhibit randomness. For example, the statement "I physiologically prefer Ethena's market-making strategy and dislike XX's arbitrage strategy" actually carries a vague preference.

It is quite certain that the attempt to make blockchain/DeFi the infrastructure for AI has suffered a dismal failure over the past decade, and deAI/deAgent/deOpenclaw will face a similar fate.

The latest large-scale models are used to modify various structures of DeFi. For example, contracts tested by Mythos have default security, and any changes will be detected in real time, thereby increasing the risk level.

In terms of human organization, AI's choice is "no people needed," only human "capabilities." DeFi is the most suitable industry for this, without exception. After the rules are designed, DeFi will only improve capital efficiency under the premise of security. Referring to the L1/2/3/4 levels of autonomous driving, it will inevitably go through the process of information authorization → limited fund usage rights → full fund usage rights.

If agents continuously learn engineered trader skills and Curator management capabilities, they will inevitably surpass humans in trading and profit. Unfortunately, the accumulated DeFi data has not yet been systematically learned and trained by AI, and current crypto AI is still in the stage of raising funds.

However, I am quite certain that the actual use of funds will be the next major wave in AI's transformation of DeFi, and that is inevitable.

So, what form will token economics take after security (contracts) and organization (humans) are upgraded?

• In the PoW era, tokens served as proof of computing power consumption, which is essentially the same as AI tokens at this stage.

• In the PoS era, tokens are discounted certificates of expected returns , and AI tokens are evolving in this direction (providing the ability to replace humans is the AI ​​expression of this economic value).

• Crypto tokens in the AI ​​era have transcended our engineering scope, and we can only make irresponsible predictions based on theory.

Similar to Sky using token allocation to control APY across channels, and Claude using token consumption to price model capabilities, future Crypto Tokens will most likely be a kind of " return on capital " certificate.

Note the distinction here: the expected return of tokens in the PoS era, such as ETH , is an economic assumption, a kind of empirical reasoning based on prior knowledge. However, with the engineering design of AI, the parameters of DeFi will be infinitely close to the real situation, and its rate of return and risk rate are highly reliable and can be verified in real time.

Furthermore, users can determine the current price of a token based on the large model and agent used by the DeFi protocol, as well as the score of the Harness optimization metric, buying if they are bullish and selling if they are bearish.

Conclusion

" Countless unspeakable sorrows and the unpredictable future of mankind."

The future of DeFi can be divided into economic and technological aspects. There is still no good solution for token economics, but security is showing a glimmer of hope. Claude Mythos can threaten the world, but conversely, it can help manage money well.

AlphaGo completely solved the Go problem, and Claude completely solved the programming problem. Such scenarios will only increase in the future. There is theoretical space for optimization in DeFi contracts, human organizations, and even economic units of account.

At least, people don't need to worry about being completely replaced. In an era where data is worthless, behavior still has its own meaning. At least for now, the takeover of people by agents is still in the details of "micro-tasks" and "micro-payments," and these are constantly repeated details. We need to make this repetitive and replicated behavior generate value. AI is causing the value of data and content to decrease infinitely, approaching zero cost. The unit economic value (cost) of AI Tokens and Crypto Tokens is also constantly decreasing, which is an inevitable trend.

This could even be argued as the first time money has truly opened its doors to individuals, whether for AI work or crypto for consumption.