When AI spends money for you, socializes for you, and earns money for you, are you still you?
Let’s start with some harsh truths.
Before writing this, I want to lay out three of the harshest statements about this industry. Not for theatrics, but because if a direction cannot pass these three hurdles, it is not worth anyone's serious consideration.
First statement: If agents can spend money on their own, who is responsible when things go wrong?
In recent weeks, Cloudflare launched a stablecoin wallet for AI agents, Coinbase integrated x402 and stablecoin wallets into AWS's agent runtime environment, allowing businesses to set budgets and governance rules for agents, and Google also released an agent payment protocol. Several tech giants are doing the same thing at almost the same time: enabling machines to pay for themselves.
It sounds cool. But a very basic question arises: if a machine places an order, signs a contract, and spends money, whose account does that charge fall on? It has no ID, cannot go to jail, and will not feel remorse.
Second statement: In a place like Suzhou, what is the actual utility of your products?
Suzhou is one of the cities in China with the highest total industrial output. It has precision manufacturing, biomedicine, and nanotechnology, with real production lines, real yield rates, and real delivery cycles. Do the assets, scores, and activity records of agents on the blockchain have any actual impact on a machine tool, a supply chain, or an order?
The phrase "digital-physical symbiosis" has been hanging around from March to August. Where exactly is the symbiosis?
Third statement: If influence is calculated as a score, and the agent's score increases as it runs, isn’t this just a numbers game?
A person's network, reputation, and weight in a circle are inherently complex, subtle, and difficult to quantify. Now, an algorithm claims it can score this and even distribute money based on the score. The more troublesome part is that agents can tirelessly establish connections for you 24/7. So, what is ultimately measured by this score: you as a person, or your machine working harder than others?
Three questions, each more difficult than the last. Let’s tackle them one by one.
First question: Where does responsibility lie?
To answer "who is responsible when an agent spends money and something goes wrong," we must first clarify a more fundamental issue: our entire economic and legal system is built on an unspoken assumption—actors can bear consequences.
A person signs a contract because they can fulfill it and can be penalized for breaching it. A company can conduct business because it has legal personhood, assets, and can be sued. This "ability to bear" is the foundation of the entire credit system.
Agents break this assumption. They can act but cannot bear responsibility. They can execute, judge, and optimize, but they cannot "be responsible"—because being responsible requires something that can be taken away, and they have nothing.
Once you understand this, looking at the solutions from the three tech giants reveals they are doing the same thing: anchoring responsibility back to humans.
Google's authorization credentials record "what the user allows the agent to purchase, the maximum amount, and who initiated the operation"—note the last part, it answers "who is it?" Coinbase's budget and governance rules are also set by businesses. In other words, the more autonomous the agent is, the clearer the traces of human authorization must be recorded.
Here lies a counterintuitive conclusion: autonomy and traceability do not diminish each other; they grow in tandem. The more an agent can do, the more rigorous the record that backs it must be. Whoever can clearly understand this and create mechanisms in advance will secure a position in this round.
The position of ATM's agent product on this line is: users set the rules, and the agent executes within those rules, with the source of will always on the human side. Logically, this makes sense.
However, there is a boundary that is becoming blurred, and no one can clearly define where it is—the line between "execution" and "decision-making." If your agent autonomously decides whom to contact, when to act, and where to allocate resources, is that execution or decision-making? If it counts as decision-making, does that loosen the anchor of responsibility?
This question currently has no answer. Whoever claims to have an answer is merely telling a story.
Second question: What lies between the digital and the physical?
The second doubt is the most solid because it is backed by data, and the data is self-evident.
First, let’s look at the positive side: as of August 7, 2026, the total market value of RWA on the blockchain reached $37.94 billion, with the total number of asset holders soaring to 1.6294 million, a month-on-month increase of 55.34%, with a net increase of over 560,000 in a single month, setting a historical record for the largest increase.
The numbers are impressive. But pay attention to what is increasing—government bonds, money market funds, private credit, gold, fund shares, stocks. All are financial assets.
Machine tools have not been put on the chain. Supply chains have not been put on the chain. The orders in Suzhou Industrial Park have not been put on the chain either.
Why?
The answer is quite simple: financial assets are inherently information, while manufacturing assets are inherently physical.
A government bond or a fund share is fundamentally a string of rights records that already exist in the ledger. Moving it onto the blockchain is merely changing ledgers, with minimal friction.
However, a machine tool, a batch of materials, or a production line is fundamentally a physical reality. To put it on the chain, you must first equip it with sensors, establish traceability, create standards, confirm rights, and have someone responsible for the authenticity of the data—each of these steps involves money, time, and coordination costs. The essence of putting something on the chain is transforming physical objects into trustworthy information, and that conversion process is the real threshold.
Thus, the difficulty of "digital-physical symbiosis" does not lie in "digital" or "physical," but in whether the cost of that conversion process can be reduced. This is the first time a slogan has gained concrete weight.
ATM's actions on this path involve applying LUCA to real travel expenses—booking hotels and flights, generating real order records. This is indeed a real cut between the digital and the physical, but to be honest: it is on the consumption side, not the production side. There are still several layers to go from "being able to buy things" to "being able to enter the production site."
So why is this stop placed in Suzhou?
Precisely because this is the most challenging place. In a city that speaks through yield rates and delivery cycles, discussing digital-physical integration, any pretty rhetoric will be quickly disproven by production line data. Only things that can be seriously discussed here have a chance of being real.
Third question: Is the score still yours?
The third doubt is the hardest of the three because it attacks the foundation of this system.
First, let’s discuss the origins of this design. It is derived from Google’s PageRank—the algorithm that determines the ranking of search results and supports a trillion-dollar company. The core idea of PageRank is: the importance of a webpage is not determined by what it says about itself, but by how many important webpages link to it.
Translating this to individuals means: a person's influence does not depend on how loudly they shout or how much money they spend, but on how many influential people they establish genuine two-way connections with. "Two-way" is key—unilateral attention does not count; both parties must lock in assets and incur costs for the connection to be valid. The intent of this design is clear: to prevent influence from being purchased unilaterally.
Using cost constraints to combat score manipulation is the right approach.
But PageRank itself serves as a cautionary tale. It gave rise to the entire SEO industry. Once a metric can be exchanged for benefits, there will always be someone who lives for that metric. This is not a failure of the algorithm; it is the fate of all quantitative systems—schools measure students by scores, so students learn for scores; platforms measure content by interaction volume, so creators produce content that is easy to interact with.
Now the question arises: will agents render cost constraints ineffective?
The reason the two-way locking cost constraint is effective is that it assumes establishing connections requires human time and judgment—you must know the other party, assess them, and decide whether it is worth the investment. But if agents can tirelessly filter, contact, and maintain relationships 24/7, the human time cost is circumvented. The remaining asset locking cost can be piled up with money.
This loophole theoretically exists. Whether ATM's design has hedges and how large a scale it can hedge against requires real data for verification.
Personally, I believe this question has not been fully answered yet. It is worth noting that this is not an issue with a single project. Any system attempting to measure "human value" in the AI era will eventually face it. The difference lies in whether some projects avoid it or bring it to the forefront. Bringing the issue to light is, in itself, a form of honesty.
A Common Thread
Having written this, I want to shift perspectives and talk about Suzhou embroidery.
Double-sided embroidery is the most famous craft in Suzhou—a single silk thread is split into sixteenth parts, weaving back and forth on a fabric as thin as a cicada's wing, with a cat on one side and another cat on the reverse, each with different expressions, colors, and postures.
Most people use this metaphor to say that "both sides are real." This is true, but it only tells half the story.
The reason double-sided embroidery is true on both sides is that they share the same thread. It is not two separate works; it is two sides of one work. Remove that thread, and neither side exists.
With this shift in perspective, the criteria for judgment change.
The real question is not whether "the virtual side counts as real"—this way of asking sets up a dichotomy between the virtual and the real from the start. The real question is: is it the same thread as the real side?
If the connections established by the agent are indeed an extension of your will, set by you, serving you, and for which you are responsible, then it and you are the same thread; both sides are you. But if the agent begins to have its own thread—making its own judgments, decisions, and accumulations—then no matter how beautiful the front side is, the back side is already empty.
This standard of judgment is much clearer and more challenging than the "virtual-real debate" because it requires you to trace back to the source of that thread.
And this, in turn, circles back to the answer to the first question: why is the trace of authorization important? Because it is the proof of that thread.
This content is provided for general informational purposes only and doesn't constitute financial, investment, legal, or tax advice. Any events, rewards, online promotions, or related information mentioned herein should not be considered a recommendation, solicitation, or invitation to purchase, sell, trade, or otherwise deal in any crypto assets. Crypto assets are highly volatile and may result in loss. The availability of WEEX services, products, and related events may vary by region. You are responsible for ensuring that your participation is in accordance with applicable local laws and regulations.






![[Kang Ryun-ho's Crypto Zoom-In] Key Points and Implications of Strengthened Reporting and Regulation for Virtual Asset Businesses](/public-static/11_6579de73a6.png?format=avif)






















