Cryptocurrency is a bank assisted by AI

Cryptocurrencies (especially stablecoins) will become the foundational banking layer for AI agents to conduct economic activities. This article details how cryptocurrencies outperform traditional credit cards by expanding trust structures, providing a global, internet-native currency, and supporting new payment primitives. It also outlines the key components needed to build an AI agent banking stack, including identity, liquidity, security, and app stores. 2026 will be the year AI agents begin to become economic participants. They will autonomously invoke SaaS APIs, execute transactions, purchase cloud computing resources, and connect workflows. Just as humans need credit cards as a “banking track” for transactions in the real world, agents also need a bank—and I argue that this bank will be based on stablecoins. This argument has two parts. The first is “why”—why is cryptocurrency (rather than credit cards) particularly well-suited to be the banking layer for agents? The second part is “how”—assuming we accept that cryptocurrency will be the banking layer for agents; what exactly do we need to build to achieve this? Why cryptocurrency, and not credit cards? The cryptocurrency community often ridicules credit cards, arguing that they are simply not suitable for agents. This is a superficial view and may not be accurate. Companies like Visa have made significant strides in agent commerce—for example, Visa Smart Commerce created a payment gateway for agents, similar to how Apple Pay works: Like Apple Pay, these “agent cards” first assume that you, as a human, possess a credit card. Visa then issues a “tokenized credential” with set limits, authorization, and expiration conditions. Like Apple Pay, these credentials have a unique virtual card number that you can securely provide to your agent. When your agent (such as OpenClaw) uses the tokenized credential to make a transaction, it is decrypted on Visa’s servers, linked to your physical card, and then Visa runs the payment process. Cryptocurrency does not have to be involved in the process. Lobster.cash has a good demonstration of this process. In short, agent cards are viable and sometimes even more popular than cryptocurrencies. So why use cryptocurrencies? There are three reasons: (1) an expanded trust structure, (2) an internet-native currency with a global audience, and (3) a new payment primitive. Expanded Trust Structure Credit cards—and the resulting agent virtual cards—have a rigid and fixed “trust structure.” It assumes that to make a payment, you always need a KYC-verified human bank account as a backing of trust. Humans then delegate trust and authorization to agents in a manner similar to parents opening supplementary cards for their children. Cryptocurrency and stablecoin payments, on the other hand, are not bound by such trust assumptions. While you can (and in many cases should) link a stablecoin wallet to a KYC-verified bank account (as is the case with centralized exchanges), you don't necessarily have to do so to gain payment authorization. You can link a stablecoin wallet to almost anything—whether it's a government ID, a social media account (Google, TikTok, Instagram OAuth), a domain name server, or a headless smart contract. Many agents may emerge from Lobster.cash and other fiat-linked trust structures. But even more agents—such as Conway—will emerge from other corners of the internet beyond fiat-linked trust structures, and stablecoins are essentially (arguably the only) optimal way to transact funds on a large scale. There's an old saying—"On the internet, nobody knows you're a dog." So, on the cryptocurrency track, nobody knows you're a bot. Stablecoins are a type of internet-native currency with a global audience. The integration of Alipay with Qwen, offering milk tea coupons, demonstrates what the future of agency commerce might look like. If you've lived in or visited China in the past decade, you've experienced the convenience of integrating an "internet-native currency" into your everyday ecosystem of applications (food delivery, ride-hailing, payroll). But that system is geographically limited, operating within a closed technological ecosystem enforced by fiat currency authority. Stablecoins, on the other hand, are global from day one, offering this internet-native currency experience to the rest of the world. This is crucial for agencies, as their first automated workflow will be a concatenation of SaaS and API calls across multiple jurisdictions and service providers. An agency coordinating workflows shouldn't need three different payment tracks if it needs to call an LLM endpoint in the US, a data provider in Europe, and a computing cluster in Southeast Asia. It only needs one.

New Payment Primitives

In this world of stablecoin payments, everything becomes a payable endpoint. What is actually happening is the double Jevons paradox of the internet economy—you increase both the number of users who can transact (by giving anyone or anything a wallet) and the transaction volume per user (you can check out in more places).

Because you can have (1) an expanded trust structure for payment methods, and (2) an internet-native currency connected through SaaS workflows and global supplier chains, you are likely to see a surge in new payment primitives. For example, anyone who creates a Dune dashboard can use stablecoins to charge for view requests.

For example, a few weeks ago at the Mistral Hackathon, I "improvised" Tokker—an AI brand management agency for TikTok creators. We gave it a stablecoin wallet through Privy to collect payments from the brands it contacts, making payments independent of the banking methods used by TikTok creators. A simple extension could be using the same stablecoin wallet to pay for various computing services, online advertising services, and so on, creating a flywheel of economic output. You increase both the number of people accessing banking services and the amount they spend online. How to Build an AI Bank Stack? Now that we have a basic understanding of why cryptocurrencies are meaningful, we need to figure out how to build this AI banking stack on the cryptocurrency track. In the human world, banks play multiple roles. They are where we store, spend, grow, and lend money. But they are also where we register our identities, act as neutral arbitrators in dispute resolution, and provide security against malicious activity (via AML). To build a bank for AI agents, we need more than just a wallet—we need a whole suite of safeguards to mediate how they transact with money. I believe there will be several interconnected components: (1) identity and authorization, (2) purchasing liquidity, (3) security, and (4) the application "storefront" for AI agents to purchase. Blockchain outperforms the traditional track in all four of these aspects. You can also link your wallet to email or social media. For example, I created a hackathon prototype demonstrating how to create a ZKID for agent payments using an email domain. You can also simply stub your agent identity on a public blockchain like Ethereum. This is the fundamental role standards like ERC 8004 play in creating an “agent registry.” The second aspect is ensuring agents can actually make payments. Money doesn't magically appear just because you create a stablecoin wallet. Most agent platforms today “sponsor” credits, but this is unsustainable at scale. Fiat-to-crypto gateways, pre-funding, Buy-Now-Pay-Later (BNPL), and other authorization schemes will be a crucial part of the agent economy. Furthermore, we need to ensure the blockchain infrastructure is actually operational. Currently, agents primarily conduct tiered microtransactions on-chain (average size $0.09). As the size and volume of these transactions increase, we need to design—e.g., batch processing, payment channels, pre-authorization—to ensure these micropayments don't clog the public blockchain. Security Assurance Just as banks need to prevent money laundering by malicious actors, correspondent banks also need to prevent malicious activities such as injection vulnerabilities, runaway API costs, and credential leaks. Coincidentally, the blockchain space has been grappling with private key theft for years and has developed a robust cryptographic immune system—including Trusted Execution Environments (TEEs), Multi-Party Computation (MPC) and multi-signatures, Zero-Knowledge Proofs (ZKP), and other safeguards. We should apply these safeguards directly to correspondent payment systems and credential storage—in many ways, a wallet private key is simply a more sensitive API key, so the entire "private key" protection stack should be adopted for correspondent skills and API credentials to ensure AI agents can securely interact with the online economy. Finally, at the application layer, we are in the era of the "app store wars" of proxy commerce. From Merit Systems to ATXP, and providers like Sponge and Sapiom, many have developed similar "app store"-like skill curation mechanisms—allowing proxies to perform tasks ranging from scraping LinkedIn profiles to sending emails and trading on Hyperliquid. Whether paying for real-world services through e-commerce, calling pay-as-you-go SaaS endpoints, or automatically trading crypto tokens, proxies need a "discovery tool" to decide which endpoints to call, which wallets to use, and how much to pay for each service. Underlying protocols like Coinbase's x402 provide proxies with a generic, permissionless way to access real-world services, ultimately enabling them to actively participate in the economy as independent financial participants. Conclusion: The era of agency in the internet economy has only just begun—Claude Code and OpenClaw only gained popularity less than six months ago. Over the past decade, the blockchain track has proven capable of supporting a multi-billion dollar on-chain economy. I believe these two factors will rapidly converge, with blockchain and stablecoins becoming the banking cornerstone of the agency economy. An AI-assisted bank won't resemble a bank at all. It will resemble a blockchain.