What Are AI Agents and How Will They Use Stablecoins?

The Rise of AI Agents

AI is no longer just a chatbot you ask questions. A new generation of autonomous software — known as AI agents — can act, decide, and transact on your behalf. They can browse the web, compare prices, book travel, manage subscriptions, negotiate deals, and execute purchases without a human clicking a single button.

The speed of development has been remarkable. In crypto and AI circles, the intersection of these two technologies has generated enormous energy.

“If you ask people, where does crypto fit within AI? The velocity and speed of development and excitement in this space of crypto and AI together was incredible.”

— Cuy Sheffield, Head of Crypto at Visa

But here's the problem. These agents can do almost anything a human can do online — except pay for things. They can't sign checks. They can't walk into a bank. They can't authenticate with a credit card the way a human would. The payment layer for autonomous software doesn't really exist yet.

That gap is now driving one of the most consequential debates in fintech: how should AI agents move money?

Why Agents Need Their Own Payment Rails

Today's payment infrastructure was built for humans. It assumes someone is sitting at a screen, entering card details, confirming a purchase, and passing identity checks. None of that works when software is the buyer.

AI agents operate 24/7, across borders, at machine speed. They might need to pay for cloud compute in one country, purchase API access in another, and settle a microtransaction for data in a third — all within the same second. Traditional payment systems weren't designed for this kind of velocity or granularity.

The core challenge is authorization without identity. How does a merchant know an agent is authorized to spend? How do you set spending limits for software that makes thousands of decisions per hour? How do you handle disputes when there's no human in the loop?

Two competing models are emerging to solve this problem, and they represent fundamentally different visions of how machine commerce will work.

Model 1: AI-Exclusive Virtual Cards

The first approach extends existing payment infrastructure to agents. Companies are building AI-exclusive virtual cards — prepaid or pre-authorized card numbers that an agent can use to transact on traditional card rails.

The appeal is obvious: merchants already accept cards. There's no integration work. The agent gets a card number, uses it to buy something, and the transaction flows through Visa or Mastercard like any other purchase. From the merchant's perspective, nothing changes.

Visa has observed this trend firsthand. Stablecoin-linked cards — which let users hold stablecoins and spend them anywhere cards are accepted — are seeing explosive demand.

“Stablecoin-linked cards are in hyper growth mode. It's hard to keep up with all of the companies coming to us that want to issue stablecoin-linked cards.”

— Cuy Sheffield, Head of Crypto at Visa

The virtual card model works well for structured, high-value transactions where existing merchant acceptance matters — booking flights, paying for SaaS subscriptions, purchasing enterprise software. It's the pragmatic, near-term solution.

But it has limitations. Card transactions carry fees, settlement delays, and intermediary risk. They require human-designed authorization frameworks that may not scale to millions of agent-to-agent transactions. And they don't work well for micropayments — the kind of sub-cent transactions that agents will need for data access, API calls, and computational resources.

Model 2: Stablecoin Wallets for Agents

The second model skips card rails entirely. Instead, agents hold stablecoins in their own wallets and pay other agents (or merchants) directly onchain. No card network. No bank. No intermediary.

This is where companies like Squads, Crossmint, and Stable Sea come in. Squads has been building stablecoin-first accounts on Solana — its Grid API and Altitude business account let developers build fintech applications on stablecoin rails. Crossmint launched AI agent exclusive virtual cards powered by Visa Intelligent Commerce. Stable Sea provisions stablecoin treasury platforms for enterprise companies. The idea is that agents need their own payment tools — whether programmable onchain accounts, virtual cards backed by existing credit lines, or stablecoin wallets with spending rules.

“Because we're stablecoin first, our job is to create these authentic pathways for a traditional business or a consumer who came originally to just store stablecoins and make payments to now interact with these novel primitives.”

— Stepan Simkin, CEO of Squads (Episode 32)

“We started serving the agent tech finance ecosystem a year ago. At the beginning we started thinking it's just stablecoins for everything. And then we hit the market with it, had people, and realized that was not what people were looking for.”

— Alfonso Gómez-Jordana, Founder of Crossmint (Episode 73)

The stablecoin model has several advantages for agent commerce. Transactions settle in seconds, not days. There are no chargebacks — settlement is final. Micropayments are practical because onchain fees can be fractions of a cent. And the entire system is programmable: spending limits, approved counterparties, and transaction rules can all be encoded in smart contracts.

“The agent can handle the complexity of on-ramping, switching from one to another, as long as it has the tools like virtual cards. It depends on what the agent wants to do — we see the vast majority want to buy a domain, hire Twilio to send SMS. Those things naturally, the agent will find a credit card form.”

— Alfonso Gómez-Jordana, Founder of Crossmint

For agent-to-agent commerce — where both the buyer and the seller are software — stablecoins are arguably the natural medium. There's no human to swipe a card. There's no reason to involve a card network. Two pieces of software can simply exchange value directly.

The Eliza Framework and Open-Source Agents

Much of the momentum in crypto-native AI agents has been driven by open-source development. The Eliza framework, originally built by ai16z in partnership with Stanford, became one of the most-starred projects on GitHub. It provides a modular architecture for building AI agents that can interact with blockchain protocols — including holding wallets, executing trades, and managing onchain assets.

Eliza demonstrated something important: the building blocks for autonomous onchain agents already exist. Developers can spin up an agent that holds a Solana wallet, monitors DeFi protocols, and executes transactions based on programmatic rules — all without human intervention.

The open-source approach matters because it means agent infrastructure isn't controlled by any single company. Anyone can build on it, extend it, and deploy agents that participate in onchain commerce. This creates a permissionless ecosystem where innovation can happen at the edges.

x402: A Protocol for Agentic Commerce

One of the most interesting developments is x402 — a protocol designed specifically for machine-to-machine payments. The name references HTTP status code 402 (“Payment Required”), which was reserved in the original HTTP specification but never implemented.

x402 enables a simple interaction: an agent requests a resource (data, compute, API access), the server responds with a payment requirement, and the agent pays with stablecoins. The entire cycle — request, payment, delivery — happens programmatically, with no human in the loop.

This is the kind of infrastructure that makes agent-to-agent commerce practical at scale. It's not about replacing human transactions. It's about enabling an entirely new category of transactions that humans were never going to make in the first place — millions of sub-cent payments between software services, negotiated and settled in real time.

Programmable Commerce: What Comes Next

The convergence of AI agents and stablecoins points toward something bigger than either technology alone: programmable commerce.

Imagine a business that deploys a fleet of AI agents to manage its supply chain. One agent monitors inventory levels and automatically reorders supplies when stock runs low. Another negotiates prices with supplier agents in real time, comparing quotes across vendors in milliseconds. A third manages payments — releasing stablecoins from a multisig wallet when delivery is confirmed onchain.

None of this requires a human to approve a purchase order, authorize a wire transfer, or reconcile an invoice. The entire flow — from demand signal to payment settlement — is automated, auditable, and instant.

Cross-border transactions become particularly interesting. An agent in Singapore can pay a supplier agent in Brazil using USDC, settling in seconds with no correspondent banking, no FX conversion delays, and no intermediary fees. Companies like Stable Sea are already building stablecoin infrastructure for exactly these kinds of cross-border use cases.

Stablecoins are a natural fit for this world because they're programmable, instant, global, and require no human authentication. They can be held by software, spent by software, and verified by software. They don't need a signature, a PIN, or a face scan.

What This Means for Businesses

For businesses watching this space, the practical implications are significant:

• New commerce flows are emerging. Agent-to-agent transactions will create markets that don't exist today — micropayments for data, real-time negotiation between procurement agents, automated treasury management.
• Payment infrastructure is fragmenting. The future won't be cards or stablecoins. It will be both. Virtual cards for high-value, human-initiated transactions. Stablecoins for machine-speed, autonomous ones. Businesses will need to support multiple rails.
• Programmable controls matter more than ever. When software is spending money, the rules need to be bulletproof. Smart contract-based spending limits, multisig approvals for large transactions, and real-time monitoring become essential.
• Speed becomes a competitive advantage. Companies that can deploy agents with stablecoin wallets will be able to operate faster, across more markets, with less friction than those relying on traditional payment infrastructure.

The race to build payment rails for AI agents is not a theoretical exercise. Visa is seeing demand from card issuers. Squads is building programmable onchain accounts. Crossmint is building AI agent virtual cards. x402 is creating open protocols. The ecosystem is moving fast.

The question isn't whether AI agents will need to move money. It's which infrastructure they'll use to do it — and who will build it first.