What to Know

  • AI agents are emerging as a new class of customers capable of making autonomous payments for online services.
  • Early use cases center on small, frequent purchases such as data, computing power, API access, inference and online tools.
  • Stablecoins, particularly USDC, have taken an early lead in measured machine-to-machine payments designed for software.
  • Coinbase said x402 processed more than 165 million payments worth a combined $50 million earlier in the year.
  • Coinbase’s AI product lead estimated that about 99% of x402 payments use USDC.
  • x402 moved about $24 million over 30 days in July, while disclosed activity remains small compared with major card networks.
  • Coinbase, Cloudflare, MoonPay, Visa and Mastercard are developing systems that let agents spend within preset limits.
  • Key unresolved issues include security, funding, liability, refunds, dispute rights and responsibility when an agent makes the wrong purchase.

AI Agents Move From Software Tools to Paying Customers

AI agents are starting to reshape the conversation around crypto adoption. For years, the industry debated how the next large wave of users might arrive, often focusing on consumers seeking cheaper payments, broader financial access or protection from unstable local currencies. A different possibility is now gaining attention: the next major class of crypto users may not be human at all. It may be software that can search, decide and pay on behalf of a person or business.

These agents are artificial intelligence systems capable of carrying out multi-step tasks. When they perform those tasks, they may need to buy something along the way. That purchase could be access to a dataset, a one-time API call, extra computing power, an inference request or an online tool that helps complete a broader workflow. In that environment, payment is not a separate consumer checkout experience. It becomes a machine-readable step in a larger automated process.

The market remains early, but major crypto, fintech and payments companies are already building infrastructure for this shift. Coinbase has developed x402, a payment protocol designed for AI agents. MoonPay has introduced PayBox, which can connect agents with cards and crypto wallets. Cloudflare has announced Cloudflare Wallets and cloudflare.pay, with infrastructure aimed at allowing agents to make online purchases within preset limits while giving sellers visibility into the human identity behind an agent.

Stablecoins Take an Early Lead in Machine Payments

Stablecoins currently appear best positioned in the part of agentic payments built from the ground up for software. The appeal is straightforward: AI agents may need to make many very small payments, often across borders, at high frequency and without the friction of traditional account creation or manual card entry. In that setting, a dollar-denominated stablecoin such as USDC can function as programmable money for automated internet transactions.

Coinbase said earlier in the year that x402 had processed more than 165 million payments worth a combined $50 million. Under the x402 process, an agent receives a price, sends payment and receives the requested service without opening an account or typing card details. Human involvement may be limited to the initial instruction given to the agent, while the software handles the transactional step needed to complete the task.

Coinbase’s head of AI product, Lincoln Murr, estimated that about 99% of x402 payments use USDC. If accurate, that gives dollar-denominated stablecoins a strong early position in the measurable segment of the agentic payments market. The dominant use case is not yet agents buying consumer goods or booking complex trips. Instead, many transactions involve agents paying small amounts for APIs, data, computing resources or access to digital tools.

The disclosed dollar value remains modest. x402 moved about $24 million over 30 days in July, an amount described as roughly what Visa handles in one minute. That comparison underlines both the promise and the limitation of the current market. Activity is real enough to attract major companies, but it is still far from mainstream payment scale.

How x402 Turns Web Payments Into Software Logic

x402 is essentially a paywall for software. When an agent requests data or a service, the seller returns a price and payment information. The agent then pays, and after the payment is verified, the seller releases the result. The name refers to the “402 Payment Required” status code, a long-standing part of the web that had rarely been used in ordinary internet commerce.

This model could change how digital services are priced. Today, many users pay subscriptions even when they need only occasional access. Agents could instead pay on demand, with tiny purchases replacing bundled monthly plans in some contexts. A person may not want to manually approve a payment of a few cents for a single piece of data, but an agent can handle that friction as part of a broader automated workflow.

Coinbase’s April update of the x402 ecosystem counted more than 480,000 agents. The disclosed payment volume and transaction count imply an average transaction of roughly 30 cents. Activity has been concentrated heavily on Base, Coinbase’s layer two blockchain, according to dashboard data referenced in market discussions. Murr also estimated that 25% to 30% of transactions may have been driven by users trying to climb public leaderboards rather than purchasing services they genuinely needed, which is an important caveat for adoption claims.

Why Stablecoins Fit Small, High-Frequency Purchases

The strongest case for stablecoins is found in small transactions where conventional payment costs can become awkward. Murr put card acceptance costs at 2% to 4%. That cost structure can be difficult when the product being sold costs less than a dollar, especially when fixed processing costs are involved. Stablecoins can also operate around the clock, and payment companies or merchants may cover blockchain fees so an agent does not necessarily need to hold a separate cryptocurrency for transaction costs.

Cloudflare’s strategy points in a similar direction. Its planned Wallets product is designed more like a controlled corporate card than an unrestricted crypto wallet. An operator could deposit funds into a main wallet, give an agent a smaller allowance and set rules around budget, approved sellers and maximum purchase size. Within those guardrails, the agent could act without requiring human approval for every small transaction.

Cloudflare has also discussed batching, where multiple small purchases are grouped before their combined value is recorded on a blockchain. This can reduce fees and may be paired with escrow, keeping funds locked until the seller provides what the agent purchased. The company’s planned Monetization Gateway would let websites charge for an individual page, dataset or online tool rather than requiring a subscription, though the system has not been broadly released.

Cards Remain Competitive for Larger Agent Purchases

Stablecoins are not the only contender. Card networks still have major advantages in conventional commerce. They already work at millions of merchants and bring familiar protections such as credit, refunds and dispute management. Those features become more important when an agent is buying travel, food, retail products or other larger consumer-facing goods.

Mastercard is developing a system based on a digital spending voucher. In that model, an owner defines what an agent may buy and how much it can spend. A seller checks those rules, provides the service and claims payment later. Mastercard calls the product Agent Pay for Machines and refers to the credentials as Verifiable Vouchers. Bundling vouchers could avoid routing every tiny transaction through the full card-payment process.

Mastercard’s approach reflects a broader expectation that agentic commerce will use multiple rails. The buyer and seller may not need to use the same form of money. An agent may see a price and spending limit, while the user pays through a card, bank account or stablecoin and the seller receives fiat currency or an accepted stablecoin. Mastercard’s product remains at an early stage through an Early Access Program, and the company has not released transaction or adoption figures.

Visa is also involved in agent-led payment experiments. In February, DBS and Visa demonstrated an agent buying food and drink with DBS/POSB credit and debit cards. DBS and Visa are exploring online shopping and travel bookings, categories that differ sharply from tiny API payments because they usually require stronger refund and dispute systems.

Security, Spending Limits and Liability Remain Unsettled

Allowing software to spend money introduces risks beyond the payment rail itself. An agent may misunderstand an instruction, choose the wrong service or follow a malicious prompt. Even if the payment works exactly as designed, the outcome may still be wrong from the user’s perspective. That makes permissioning and liability central to the development of agentic commerce.

Coinbase, Cloudflare, Mastercard, MoonPay and wallet infrastructure provider Turnkey are converging around a similar first line of defense: restrict the agent before it acts. Systems may use capped balances, approved sellers, transaction limits and human approval for sensitive actions. The idea is to define the agent’s operating boundaries in advance rather than approve every small purchase manually.

For micropayments, full fraud mitigation may cost more than the value at risk. If an agent wastes 10 cents, many users may accept the loss rather than pursue a dispute. But the trade-off changes for larger purchases. A mistake involving travel or expensive goods may require refunds, escrow, identity checks and dispute processes. Coinbase documentation describes checkouts where a buyer approves USDC before a seller collects it, and where completed payments can be partly or fully refunded.

Cloudflare points to escrow, identity and seller ratings as possible safeguards. Turnkey has argued that agents will eventually need technical proof that a digital service performed the work it was paid to do. These systems may establish who authorized an agent, how much it could spend and whether a seller delivered something. They still cannot fully answer whether the agent correctly understood the owner’s intent. Responsibility when an agent follows its rules but buys the wrong thing remains an open question.

An Early Market With Big Infrastructure Ambitions

The agentic payments market is still in an experimental phase. Cloudflare’s wallets are still in development. Mastercard’s solution is in early access. MoonPay has not disclosed adoption figures or the share of payments made with cards versus stablecoins. A meaningful portion of Coinbase’s transaction count may reflect testing, experimentation or leaderboard incentives. These limitations matter when assessing how quickly the market can scale.

Still, the strategic interest is clear. One agent could eventually complete a task by buying from dozens of services, replacing a bundle of subscriptions with a stream of tiny, purpose-specific purchases. That structure favors payment systems that are programmable, low-friction and suitable for machine-to-machine activity. Stablecoins do not need to replace cards everywhere to become a major payment method for agents. They only need to prove superior for the parts of the internet where software buys from software.

The hardest practical problem may be funding the agent in the first place. Wallet setup remains a major pain point, and companies are working on smoother onboarding through familiar tools such as fiat onramps. If users can easily create, fund and permission an agent wallet, the path toward broader adoption becomes more realistic. For now, stablecoins have measurable usage and an early lead, while cards retain deep advantages wherever trust, credit, refunds and merchant acceptance matter most.

Frequently Asked Questions (FAQs)

What are agentic payments?

Agentic payments are transactions made by AI agents on behalf of users or businesses. The agent can receive a price, send payment and access a service as part of completing a broader task.

Why are AI agents being discussed as crypto users?

AI agents may need to make many small online purchases for data, computing power, APIs and tools. Crypto rails, especially stablecoins, can support programmable payments that software can execute directly.

Which stablecoin is leading early agentic payment activity?

USDC appears to have an early lead in measured x402 activity. Coinbase’s AI product lead estimated that about 99% of x402 payments use USDC.

What is Coinbase x402?

x402 is a payment protocol for AI agents. It lets online services request payment in a machine-readable way, allowing an agent to pay and receive the requested data or service without a traditional checkout flow.

How large is the current x402 market?

Coinbase said x402 processed more than 165 million payments worth a combined $50 million earlier in the year. It also moved about $24 million over 30 days in July.

Why might stablecoins work better than cards for tiny payments?

Stablecoins can be useful for small, high-frequency payments because card acceptance costs can be difficult for purchases worth less than a dollar. They also operate continuously and can be integrated directly into software workflows.

Are card networks still relevant for AI agent payments?

Yes. Cards remain strong for larger purchases at conventional merchants because they offer broad acceptance, credit, refunds, dispute management and familiar user experiences.

What risks come with letting AI agents spend money?

An agent could misunderstand instructions, choose the wrong service or respond to a malicious prompt. Developers are using capped balances, approved sellers, transaction limits and human approvals to reduce potential damage.

Is the agentic payments market already mainstream?

No. Adoption remains early, and several products are still in development or early access. However, measurable stablecoin usage and growing infrastructure investment show that major payments companies are taking the category seriously.

Photo by RDNE Stock project on Pexels