Author:Scarlett Zhang

I'm increasingly getting the strong feeling that the crypto space is a bit too eager to be seen by the AI space.

Over the past six months, you've probably noticed the entire crypto world desperately trying to cozy up to AI. Talking about AI, retweeting AI, hosting AI events, building AI demos, pivoting to an AI narrative—every project seems hell-bent on proving it has some connection to AI.

It feels a lot like:

A kid desperately trying to squeeze onto the adults' table.

But on the other side?

Many people genuinely building in AI have a very nuanced attitude towards crypto. It's not about publicly bashing you or outright rejecting you, but a kind of polite, respectable distance:

"We're not against blockchain, but we don't want to be too deeply associated with crypto just yet."

"It's technically interesting, but our clients and investors would have concerns."

In other words:

You're okay, but I'd rather not get too involved with your crowd.

Right now, there's a very subtle hierarchy of disdain between these two worlds.

And it's not without reason.

In the minds of many AI builders, AI represents a tangible productivity revolution, technological progress, something that's changing how we work, the shape of products, and the flow of information.

And crypto? In their eyes, it's more like an industry that's overly financialized, narrative-first, always searching for the next story to prove its relevance, and incidentally, a convenient way to launch tokens and cash out.

So when the crypto space suddenly starts talking about AI en masse, the first reaction from many in the AI world is essentially:

Are you actually building something serious, or are you just latching onto another narrative?

Frankly, I completely understand that reaction.

Because over the years, crypto has indeed become too adept at packaging the "next big thing."

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

Every cycle, someone stitches the latest buzzword onto themselves and tells you the future has arrived.

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

You're always talking about the future, but it always makes people wonder if you're actually creating value or just manufacturing hype.

That's also why many in the AI space today naturally feel they occupy a higher ground.

They think:

AI is solving real problems.

Crypto is still searching for its new legitimacy.

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

But the more I think about it lately, the interesting part isn't why crypto is so eager to get close to AI.

It's another, more counterintuitive question:

Could it be that, in the end, the one who 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 Isn't a Question of "AI Being Smarter," But of "Whether AI Can Move Money"

I'm becoming more convinced of this because many Agent demos ultimately get stuck at the same point.

You've probably seen plenty of demos lately:

Agents that can write code, call tools, browse the web automatically, execute multi-step tasks, and even ones that are starting to trade, make payments, and run automated operational chains.

The first time you see it, it's undeniably cool.

But after seeing many, I keep coming back to one question:

Is it just "capable of" doing something, or can it actually "do" it?

Because the gap between "capable of" and "doing" isn't just a matter of product details.

What lies in between is:

Permissions, funds, responsibility, boundaries.

Having an agent summarize a report for you and having an agent execute a real transaction for you are completely different problems.

If the former gets it wrong, you just think it's a bit dumb.

If the latter gets it wrong, the money is gone.

So I'm increasingly feeling that AI demos are great at creating one particular illusion:

It looks like everything is almost connected.

But what's often not connected is the hardest layer.

Namely:

The execution layer.

Where agents truly get stuck isn't in thinking, but in execution related to money.

If an AI agent truly starts acting on your behalf, it will quickly need to buy APIs, rent compute power, call paid services, execute trades, manage budgets, transfer assets, and complete payments between different systems.

In other words, it doesn't just need to "understand your intent."

It needs to start participating in economic activity.

And once you enter this layer, the problem changes.

Traditional Finance Can Handle Automation, But It Wasn't Designed for an "Agent World"

At this point, many friends might ask,

Traditional finance can clearly do these things too.

I've certainly thought about that, and honestly, traditional finance is indeed more mature than crypto in many dimensions.

Risk control, auditing, permission management, accountability chains, reversibility—in these areas, traditional finance is simply stronger today.

So the real point of this article isn't:

Crypto is better than traditional finance.

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

If we're talking about internal enterprise agents or platform-internal agents, many things can certainly continue to run using banking APIs, corporate payment systems, virtual cards, approval workflows, sub-account systems, platform credits, and centralized custodial accounts.

These can all work, and in the short term, they will likely remain the mainstream.

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

An agent is not a native execution entity.

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

That's fine for many scenarios.

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

So the real question isn't:

Whether traditional finance can handle it.

But rather:

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

What the agent world needs isn't just accounts, but an execution structure.

These are actually two completely different problems.

The Key to AI Agents Isn't Whether They Are "Legal Entities," But That They Are Increasingly Like "Execution Units"

At this point, you might want to say:

"But agents aren't a third type of entity either. They're not people, they're not companies, they're just software agents."

That's true.

Strictly speaking, AI agents may not necessarily become independent legal entities. Most of the time, they are more likely just proxies for users, businesses, or platforms.

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

That's the key.

This issue hasn't fully erupted yet because there aren't enough agents, and many things are still in the "human-supervised" stage.

But if there truly are agents at scale in the future:

Trading for you,

Procuring for you,

Handling operations for you,

Managing budgets for you,

Automatically calling resources across systems for you,

You'll encounter a very awkward problem:

How should these things actually have permissions?

Whose account do they belong to?

Who does their payment authorization tie back to?

How much can they spend?

Who is responsible if they exceed their permissions?

When they call services globally, how does the underlying settlement work?

Traditional finance isn't incapable of handling it.

But it will become increasingly awkward.

Because it was never designed with the premise that "software execution units would participate in economic activity at scale."

Traditional finance isn't incapable, but it becomes less natural the further you go.

Once You Swap the Protagonist for an Agent, Those Crypto Concepts That Used to Sound Like "Talking to Themselves" Start to Become Concrete

In the past, many looking at crypto felt it was always talking about very abstract terms:

Programmable money

Programmable identity

Permissionless

Global settlement

Trustless execution

A lot of the time, it did sound like gibberish.

But if you swap the protagonist for an AI agent, these concepts suddenly become less abstract.

Because what agents truly need might precisely be these things:

They need a form of money that is natively callable.

They need an execution identity that doesn't require first becoming a "corporate account."

They need budgets and permissions that can be programmatically constrained.

They need to complete low-friction settlement globally.

They need to establish a native connection between calling behavior and asset behavior.

At this point, when you look at a wallet, your perspective shifts completely.

A wallet isn't a "place to hold coins."

What is it more like?

More like an execution container with permission boundaries.

A Wallet Isn't a "Place to Hold Coins," But an Execution Container for Agents

It doesn't just hold assets.

It can also hold rules:

What it's allowed to do

How much it's allowed to spend

Which actions can be automated

What threshold requires human confirmation

Which scenarios are read-only, which are writable

Which policies are enforced on-chain, which must stop

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

AI is responsible for understanding.