AI Agent needs Crypto, not Crypto needs AI

Author: Scarlett Zhang

I increasingly feel that the crypto space is a bit too eager to be seen by the AI space.

In the past six months, you will find that the entire crypto space is making a significant effort to get closer to AI. Talking about AI, pivoting to AI, organizing AI events, creating AI demos, changing AI narratives, and trying to prove that every project has some connection to AI.

That feeling is very much like:

A child desperately trying to squeeze onto the adult table.

But what about the other side?

Many people who are genuinely working on AI have a very nuanced attitude towards crypto. They are neither openly criticizing you nor outright rejecting you, but rather there is a very dignified sense of detachment:

"We are not opposed to blockchain, but we do not wish to be too closely tied to crypto for now."

"Technically interesting, but our clients and investors might mind."

Translating this means:

You are fine, but I don't really want to associate with your circle.

Currently, there indeed exists a subtle hierarchy of disdain between these two circles.

And this is not without reason.

In the minds of many AI builders, AI is a genuine productivity revolution, a technological advancement that is changing the way we work, the forms of products, and the flow of information.

And what about crypto? In their eyes, it resembles an overly financialized industry, driven by narratives, always looking for the next story to prove its relevance, while conveniently issuing tokens to profit from the unsuspecting.

So when the crypto space suddenly starts talking about AI on a large scale, many people in the AI space's first reaction is actually:

Are you seriously building products, or are you just riding on a narrative?

Honestly, I completely understand this reaction.

Because over the past few years, crypto has indeed been too good at packaging the "next round."

DeFi / NFT / GameFi / SocialFi / DePIN / inscriptions, and now it's AI x Crypto.

Every round has someone stitching the latest buzzword onto themselves and then telling you that the future has arrived.

Over time, the outside world has formed a difficult-to-reverse impression of crypto:

You are always talking about the future, but it often raises doubts about whether you are creating value or just creating an atmosphere.

This is also why many people in the AI space today naturally feel they are in a higher position.

They feel:

AI is solving real problems.

Crypto is still searching for its new legitimacy.

This bias is very real. This hierarchy of disdain does indeed exist.

But the more I think about it recently, the more I feel that the interesting question is not why crypto wants to get closer to AI.

But rather another more counterintuitive question:

Could it be that in the end, the one that truly needs the other is actually AI?

To be more precise:

It’s not that Crypto needs AI.

It’s that AI Agents need Crypto.

This is not a question of "Is AI smarter?" but rather "Can AI move money?"

I am increasingly convinced of this because many Agent demos end up getting stuck in the same place.

During this time, many of you should have seen quite a few demos:

Some can write code, some can call tools, some can browse the web automatically, some can execute multi-step tasks, and some have even started to appear that can conduct transactions, make payments, and perform automated operations on the blockchain.

The first time you see it, of course, it seems cool.

But after seeing more, I increasingly care about one question:

Does it just "know how," or can it really "do"?

Because the difference between "knowing how" and "doing" is not just a matter of product details.

What lies in between is:

Permissions, funds, responsibilities, boundaries.

Having an agent summarize a report for you is completely different from having an agent complete a real transaction for you.

If the former goes wrong, you might just think it’s a bit silly.

If the latter goes wrong, money is lost.

So I increasingly feel that AI demos easily create an illusion:

It seems like everything is smoothly connected.

But what is truly not connected is often the most difficult layer.

That is:

The execution layer.

The real bottleneck for Agents is not in thinking, but in execution related to money.

If an AI agent really starts doing work for you, it will quickly need to buy APIs, rent computing power, call paid services, execute transactions, manage budgets, transfer assets, and complete payments between different systems.

In other words, it not only needs to "understand your intentions."

It needs to start participating in economic activities.

And once it enters this layer, the questions change.

Traditional finance can support automation, but it is not designed for the "Agent world"

At this point, many friends might want to ask,

Traditional finance can also do these things.

I have certainly thought about this, and to be honest, traditional finance is indeed more mature than crypto in many dimensions.

Risk control, auditing, permission management, responsibility chains, traceability—traditional finance is simply stronger in these aspects today.

So the real meaning of this article is not:

Crypto is better than traditional finance.

Nor is it that without crypto, AI agents cannot work at all.

If it’s just an internal agent for a company or a platform, many things can certainly continue to run using bank APIs, enterprise payment systems, virtual cards, approval flows, sub-account systems, platform credits, and centralized custodial accounts.

These can work, and in the short term, they are likely to remain mainstream.

But the problem is that these systems are essentially built on the same premise:

An Agent is not a native execution entity.

It is merely an automated extension of a user, a company, or a platform.

This is not a problem in many scenarios.

But as agents become more autonomous, more cross-platform, more cross-border, and increasingly need to natively call resources and funds between different systems, traditional systems will start to feel increasingly awkward.

So the real question is not:

Can traditional finance support this?

But rather:

Is it the most natural, scalable, and natively adaptable structure for agents?

What the Agent world needs is not just accounts, but a set of execution structures.

These are actually two completely different questions.

The key to AI Agents is not whether they are "legal entities," but rather that they increasingly resemble "execution units"

At this point, you might want to say:

"But agents are not a third type of entity. They are neither people nor companies; they are just software agents."

This statement is correct.

Strictly speaking, AI agents may not necessarily become independent legal entities. Most of the time, they are more likely to be agents of users, companies, or platforms.

But even so, they will increasingly resemble execution units that can be endowed with budgets, permissions, tasks, and boundaries.

That is the key.

The reason this issue has not fully erupted yet is that agents have not reached that level; many things are still at the stage of "humans supervising them."

But if there are indeed large-scale agents in the future:

Helping you make transactions,

Helping you with procurement,

Helping you with operations,

Helping you manage budgets,

Helping you automatically call resources between systems,

You will discover a very awkward question:

How should these things have permissions?

Whose account is it?

Whose payment authorization is it tied to?

How much can it spend?

If it exceeds its permissions, who is responsible?

When it calls services globally, how is the underlying settlement handled?

Traditional finance can support this.

But it will become increasingly awkward.

Because it was never designed with the premise that "software execution units will participate in economic activities on a large scale."

Traditional finance is not incapable of supporting this; it just becomes less natural as time goes on.

Once the protagonist is changed to Agents, those concepts in crypto that previously seemed like "self-talk" start to become concrete

In the past, many people looked at crypto and felt that it was always talking about some very abstract terms:

Programmable money

Programmable identity

Permissionless

Global settlement

Trustless execution

Many times it indeed sounds like a monk chanting scriptures.

But if the protagonist is changed to AI agents, these concepts suddenly become much less abstract.

Because what agents truly need might just be these things:

They need a natively callable form of funds.

They need an execution identity that does not have to first become a "company account."

They need budgets and permissions that can be programmatically constrained.

They need to complete low-friction settlements globally.

They need to establish native connections between behavioral actions and asset actions.

At this point, when you look at wallets, the perspective will be completely different.

A wallet is not a "place to store coins."

What is it more like?

It is more like an execution container with permission boundaries.

Wallet is not a "place to store coins," but an execution container for Agents.

It stores not just assets.

It can also store rules:

What is allowed to be done

How much is allowed to be spent

Which actions can be executed automatically

What thresholds must be manually confirmed

Which scenarios are read-only, and which are writable

Which strategies take effect on-chain, and which must be paused

From this perspective, the relationship between AI and wallets becomes very interesting:

AI is responsible for understanding.

Wallet is responsible for constraints.

Agent is responsible for action.

This is what a complete system looks like.

The real irony is: The question posed by AI is trust, and what crypto lacks the most is also trust

If I were to stand from the perspective of an opponent, I would also say:

You just said that what AI truly lacks is trust, so why does the answer point to crypto?

This criticism is very reasonable.

After all, in the eyes of most ordinary people, crypto is precisely not that "naturally trustworthy" system.

Managing private keys is complex.

On-chain transactions are irreversible.

Phishing and theft are rampant.

Contract risks are high.

Responsibility boundaries are often blurred.

After incidents, there may not be anyone to take responsibility.

So what I really want to express is not:

Crypto has already solved trust.

On the contrary.

My judgment is:

AI will force crypto to confront trust head-on.

In the past, crypto could still linger at the level of "can transfer, can use, can run."

But if it truly wants to become the execution layer for AI agents, it must address the most challenging lessons:

Permission models

Security boundaries

Responsibility attribution

Risk control systems

Recoverability

Human-machine collaboration confirmation mechanisms

In other words, AI will not automatically make crypto successful.

Instead, AI will bring to light all the vague, lazy, and narrative-covered aspects of crypto.

So I am not saying that crypto is already the answer.

I am saying:

If there truly exists agent-native execution infrastructure in the future, it will likely resemble crypto more than today’s traditional account systems.

So the question may never have been "How does Crypto leverage AI to turn around?"

This is also the perspective that has been bothering me the most recently.

Many people, when talking about AI x Crypto, automatically interpret it as:

Crypto is trying to ride on AI again.

Crypto wants to tell a new story with AI.

Crypto needs AI to extend its life.

I do not deny that there are indeed many projects in the market that fit this description, and quite a few.

But if we only stop at this level, we will miss a more fundamental layer:

Once AI truly moves towards execution, it will inevitably encounter issues of funds, permissions, responsibilities, identities, and settlements.

And these issues cannot be solved simply by "making the model a bit stronger."

They are essentially another layer of infrastructure issues.

In other words, as AI develops further, it will increasingly approach the domain of problems that crypto excels at handling.

Not because crypto is more advanced than AI.

But because when AI reaches out to the real world, it must confront:

How does money move?

How are permissions granted?

How is responsibility accounted for?

And this is precisely not something that prompts can solve.

What AI truly lacks may not be more intelligence, but more trustworthiness

I am increasingly convinced that the most challenging part of AI x Crypto has never been intelligence.

But rather trust.

You can create a stunning demo:

Complete a swap with one sentence

Complete a bridge with one sentence

Automatically configure assets with one sentence

Automatically execute on-chain actions with one sentence

It certainly looks very futuristic.

But do users really dare to use it?

Even if they dare to try it once, do they dare to use it long-term?

Even if they dare to use it long-term, how will responsibility be defined if something goes wrong?

Will the product dare to make promises?

Will the platform dare to take responsibility?

Will developers dare to grant higher permissions?

So in the end, you will find that what truly limits AI agents from entering the financial and asset world is not whether they are smart enough.

But rather:

Are they constrained enough?

Who can define their boundaries?

Who can verify their actions?

Who can stop them before risks occur?

Who can clarify responsibilities after risks occur?

So what will truly be scarce in the future may not be the strongest models or the most articulate agents.

But rather:

The most trustworthy execution layer.

This is also why I increasingly believe in this statement

AI Agents need Crypto, rather than Crypto needing AI.

To be more precise:

Not all AI needs crypto.

Not all agent scenarios require crypto.

Nor has crypto provided a mature answer yet.

But I increasingly believe:

As AI agents move towards real execution, real assets, real permissions, and real responsibilities, they will increasingly need a foundational infrastructure that resembles crypto more.

What they need is not more concepts.

What they need is:

Programmable money

Programmable permissions

Programmable identities

Native global settlement

Verifiable execution boundaries

These are precisely the few areas where crypto is genuinely not just talking in vain.

So in a sense, I do not think the current disdain chain from the AI space towards crypto is entirely unreasonable.

But I also increasingly doubt:

This is more about both sides standing on different timelines.

Today, the AI space is most concerned with models, products, distribution, and efficiency.

While the crypto space has been living longer in issues of assets, permissions, custody, settlements, and responsibilities.

Everyone seems to be discussing the future.

But in fact, they are not discussing the same layer of the future.

The AI space feels that crypto is too narrative-driven, too financial, too speculative.

The crypto space feels that the AI space has not yet truly encountered the most challenging execution problems.

In a sense, neither side is completely wrong.

I just increasingly feel that when AI agents truly begin to participate in economic activities on a large scale, the seemingly stable hierarchy of disdain today may slowly reverse.

At that time, the question may no longer be:

Why does crypto always want to get closer to AI?

But rather:

How can AI agents truly enter the real world without a more suitable execution infrastructure for agents?