Why is cryptocurrency said to be the bank for AI agents?

Written by: @0xfishylosopher

Translated by: Blockchain Knight

By 2026, AI agents will begin to become economic entities.

They will call SaaS APIs, execute transactions, purchase cloud computing resources, and connect workflows—all autonomously.

Just as humans need credit cards as a “banking channel” in the real world for transactions, AI agents will also need a bank, which I believe will exist in the form of stablecoins.

This paper is divided into two parts. The first part is “Why”—why cryptocurrencies (rather than credit cards) are particularly suitable to become the banking layer for agents. The second part is “How”—assuming we accept cryptocurrencies as the banking layer for agents, what exactly do we need to build to achieve this?

Why do agents prefer cryptocurrencies over bank cards?

Crypto Twitter users often mock credit cards, believing they don’t work for agents at all. This view is superficial and may not be accurate. Visa and other companies have already made significant progress in the agent business sector.

For example, Visa’s smart commerce creates a payment gateway similar to Apple Pay for agents:

Like Apple Pay, these “agent cards” initially assume you personally own a credit card.

Visa then issues a “tokenized credential” with limits, authorization, and validity conditions. Like Apple Pay, these credentials have a virtual card number that you can safely provide to your agent.

When your agent (e.g., OpenClaw) uses the tokenized credential for a transaction, the credential is decrypted on Visa’s servers and linked to your actual bank card, with Visa handling the payment process. No cryptocurrency is involved in this entire process.

Some products demonstrate this process well in practice. In short, agent cards are effective and sometimes even more popular and secure than cryptocurrencies.

So, why choose cryptocurrencies at all? There are three reasons:

(1) Extended trust structures;

(2) Internet-native currency for a global audience;

(3) New payment methods.

1 - Extended Trust Structures

Credit cards and the derived virtual agent cards rely on rigid and fixed “trust structures.”

This structure assumes that, for payments, users always need a KYC-verified bank account as a trust guarantee.

Then, users “delegate” trust and authorization to the agent, similar to parents issuing subsidiary cards for their children.

In contrast, cryptocurrency and stablecoin payments are not limited by such trust assumptions. While you can (and often should) link a stablecoin wallet to a KYC-verified bank account (e.g., on centralized exchanges), you can also make payments without doing so.

You can link a stablecoin wallet to almost anything—government-issued IDs, social media accounts (e.g., Google, TikTok, Instagram OAuth), domain name servers, or headless smart contracts.

Many agents may choose fiat-pegged trust structures, but there will also emerge from other corners of the internet outside traditional fiat trust models. Stablecoins are essentially the best (or even the only) way to handle large-scale transaction funds.

There’s an old saying: “On the internet, nobody knows you’re a dog.” Similarly, in crypto, nobody knows if you’re a bot.

2 - Internet-Native Currency for a Global Audience

Stablecoins are a globally accessible, internet-native currency. The integration of Alipay and Qwen, offering pearl milk tea vouchers, exemplifies the future development of agent-based commerce.

If you’ve lived or visited China in the past decade, you’ve experienced the convenience brought by “internet-native currency,” which permeates daily applications like food delivery, ride-hailing, and salary payments.

However, this system is geographically limited and operates within a closed technical ecosystem maintained by fiat currency authorities.

In contrast, stablecoins have been inherently global from the start, providing the world with an internet-native currency experience.

This is crucial for agents because their initial automation workflows will involve chain SaaS and API calls across multiple jurisdictions and service providers.

If an agent needs to coordinate a workflow that accesses U.S. LLM endpoints, European data providers, and Southeast Asian computing clusters, it doesn’t need three different payment channels—just one.

3 - New Payment Primitives

In the world of stablecoin payments, anything can become a payment endpoint. The reality is paradoxical: for the internet economy, you both increase the number of users who can transact (by giving anyone a wallet) and increase the transaction volume per new user (by enabling more purchases).

Because you can have (1) an extended trust structure for payments and (2) internet-native currency flowing through SaaS workflows and global supply chains, new payment methods will emerge endlessly. For example, anyone creating a Dune dashboard can use stablecoins to collect viewing requests.

For instance, a few weeks ago, I coded for Tokker at the Mistral Hackathon, showcasing a brand management AI agent for TikTok creators.

We equipped it with a stablecoin wallet (via Privy platform) to receive payments from brands it contacts, avoiding banking issues associated with TikTok creators.

A simple extension is to use the same stablecoin wallet to pay for various computing services, online advertising, and more, creating a positive economic cycle: increasing bank account usage and boosting online consumption.

How to build an AI banking tech stack?

Having grasped the “meaning” of cryptocurrencies, the next step is to figure out how to build this AI banking system based on crypto.

In the human world, banks serve multiple roles: storing, spending, increasing value, and lending funds. They are also where we register identities, serve as neutral dispute resolution bodies, and act as security barriers against malicious activities through AML.

To build a banking system for AI agents, we need more than just a wallet—we need a comprehensive set of security mechanisms to regulate how they transact with funds.

I believe it will include several interconnected components: (1) identity and authorization, (2) liquidity procurement, (3) security safeguards, and (4) application storefronts for AI agents. Blockchain outperforms traditional payment methods in all four areas.

1 - Identity and Authorization

The first aspect is identity and authorization—who is the transaction agent, and who do they represent? There are various design options.

You can emulate Visa’s approach by creating an ID linked to your fiat card. Or you can associate wallets with email or social media accounts. For example, I created a prototype for a hackathon:

How to create a ZKID for agent payments using email domain.

You can also directly embed agent identity info onto public blockchains like Ethereum. Standards like ERC 8004 enable this by creating “agent registries.”

2 - Liquidity Procurement

The second aspect is ensuring agents can actually pay the required costs. Funds don’t appear out of thin air just because you create a stablecoin wallet.

Currently, most agent platforms offer “sponsorship” points, but this isn’t sustainable at scale. Fiat-to-crypto gateways, prepayments, buy-now-pay-later (BNPL), and other authorization mechanisms will become vital components of the agent economy.

Additionally, blockchain infrastructure must operate smoothly. Today, agents mainly perform microtransactions on-chain (average size around $0.09). As scale and transaction volume grow, designs like batch processing, payment channels, and pre-authorization are necessary to prevent small payments from clogging the public blockchain.

3 - Security Safeguards

Just as banks need to prevent malicious activities like money laundering, agent banks must guard against rapid injections, API fee surges, and credential leaks.

Coincidentally, the blockchain space has long addressed private key theft, developing robust cryptographic immunity systems, including trusted execution environments, multi-party computation, multi-signatures, zero-knowledge proofs, and other security measures.

We should implement these protections directly around agent payment systems and credential storage—since wallet private keys are just more sensitive API keys, they should be protected with comprehensive “private key” security mechanisms to ensure AI agents can interact safely with the online economy.

4 - Application Storefronts

Finally, at the application layer, we arrive at the “app store” era—where intelligent trading has entered the “agent commerce” age.

From Merit Systems to ATXP, Sponge, and Sapiom, many service providers have developed “app store”-like skill management systems, enabling agents to perform a variety of operations—from scraping LinkedIn info to sending emails and trading on Hyperliquid.

Whether paying for real-world services via e-commerce, accessing on-demand SaaS, or automatically trading tokens, agents need a “discovery tool” to decide which services to invoke, which wallets to use, and how much to pay for each.

Protocols like Coinbase’s x402 provide a permissionless, universal way for agents to access real-world services, ultimately allowing agents to actively participate as independent financial actors.

Conclusion

The era of agent-based internet economy has just begun; the rise of Claude Code and OpenClaw is less than six months old.

Over the past decade, blockchain infrastructure has proven capable of supporting billions of dollars in on-chain economic activity. I believe these two factors will rapidly converge, with blockchain and stablecoins becoming the foundational banking layer for the agent economy.

AI agent banks will look less like traditional banks and more like blockchain.