Stablecoins, agentic payments, and the end of the human checkout

Stablecoins are the boring infrastructure story of the last five years. They are also, quietly, the reason AI agents will be able to pay for things at all. This is a grounded look at what they are, why they fit agent traffic better than the card networks ever could, and what online commerce starts to look like when the buyer on the other side of the page is software.

What a stablecoin actually is

A stablecoin is a digital token that tries to hold a one-to-one value with a reference asset, almost always a fiat currency like the US dollar or the euro. You send and receive it like any other token on a public blockchain, but the price stays close to the thing it is pegged to instead of swinging around like bitcoin.

The interesting part is the rail underneath. A stablecoin transfer settles in seconds on networks like Base, Solana, or Ethereum, costs a fraction of a cent on the cheaper ones, runs every minute of every day, and clears without a bank in the middle. From a software standpoint it is closer to making an HTTP request than to running a card charge.

The three flavors, and why only one matters here

People lump everything called a stablecoin together, but the designs are very different and only one of them is suitable for payments at any meaningful scale.

• Fiat-backed. A regulated issuer holds cash and short-term government debt one-for-one against the tokens in circulation, and publishes attestations. USDC, EURC, USDT, and PYUSD work this way. This is the category that merchants and finance teams are actually comfortable with.
• Crypto-collateralized. Tokens are minted against on-chain collateral, usually over-collateralized so the peg holds. DAI and USDS (Sky) are the canonical examples. Useful in DeFi, niche for payments.
• Algorithmic. No real reserves, the peg is supposed to hold through market incentives. The category that produced Terra/UST in 2022. Not a serious payment instrument.

The rest of this post talks only about fiat-backed stablecoins, because that is what an agent paying for an API call in 2026 is actually moving.

Why stablecoins fit agentic payments specifically

It is tempting to assume agents will just use cards, the way humans do. They cannot, and the reasons are structural rather than temporary.

Card checkout is not designed for agents

A modern card payment is built around proving a human is present at the moment of purchase. 3-D Secure 2 step-ups, one-time passcodes pushed to a registered device, CAPTCHAs in front of checkout, BankID prompts in the Nordics, "is this you?" interstitials, and risk models that score mouse movement and device fingerprints. The EMVCo 3-D Secure spec is explicitly built around step-up authentication of a cardholder. None of that machinery was designed with an autonomous software buyer in mind, and routing an agent through it ranges from awkward to outright blocked depending on the merchant and the issuer.

You can stitch together a workable path for a single demo with a virtual card and a saved session, but it tends to break the moment a merchant's risk system sees machine-shaped traffic. Card issuers are getting better at detecting non-human checkouts, not worse, and the trend line is not in the agent's favor.

The chargeback model assumes a human victim

Card networks accept the cost of fraud because there is a human who can later say "that was not me." Visa and Mastercard's dispute frameworks are built around that assertion. When the buyer is an agent acting on policy set by a principal, the "that was not me" claim is much weaker, and merchants quickly learn that agent-originated card traffic carries unbounded chargeback risk. The rational response is to block it, which is what is starting to happen.

The per-call economics do not work on cards

A typical card transaction in Europe costs the merchant somewhere between 0.2% and 2% plus a fixed fee of a few cents, set by the EU Interchange Fee Regulation and the schemes on top of it. That is fine when the cart is fifty euros. It is ruinous when an agent is paying four cents for an API call a hundred times a minute. Stablecoin settlement on a cheap L2 is a small fraction of a cent and does not scale with the number of transactions in the same way.

Programmability is the point

A stablecoin payment is a signed message. You can scope it, cap it, attach metadata, route it through a policy check before signing, and tie the on-chain receipt to the task that triggered it. Cards expose almost none of this to the buyer side. For agents, the ability to enforce per-merchant caps and per-call limits before signing is not a nice-to-have, it is the only safe way to operate.

Always on, global by default

Stablecoin rails do not close on Sundays, do not care which country the merchant is registered in, and do not require a separate acquirer relationship per market. For an agent that might call a German weather API, a US scraping service, and a Japanese translation endpoint in the same task, that matters.

The stablecoins worth knowing in 2026

A short, opinionated tour. Coverage is intentionally narrow: payments-grade, fiat-backed, and actually used by agents and merchants today.

USDC (Circle)

The default for agentic payments today. USDC is issued by Circle, backed by cash and short-dated US Treasuries held in segregated accounts, with monthly third-party attestations published as part of Circle's transparency reports. It is native on Base, the L2 most x402 examples use, which is why most x402 demos and early agentic-payment integrations settle in USDC.

EURC (Circle)

Circle's euro-denominated equivalent of USDC, issued under the EU's Markets in Crypto-Assets regulation (MiCA). For European agents paying European merchants, EURC removes the FX leg entirely and is the asset most likely to satisfy a European finance team's accounting and compliance constraints.

USDT (Tether)

The largest stablecoin by supply globally, issued by Tether. Dominant in emerging markets and on Tron for remittance-style flows. Used in agentic payments where USDT is what a counterparty actually holds, less so for European B2B agent traffic where the reporting expectations of finance teams push toward USDC and EURC.

PYUSD (PayPal / Paxos)

A US-regulated stablecoin issued by Paxos for PayPal. The interesting bit is distribution: PYUSD is reachable from PayPal balances, which gives it a path into agent flows that touch consumer wallets rather than corporate treasuries.

DAI / USDS (Sky)

The main crypto-collateralized example, now sitting under the Sky brand. Important to understand the category exists, but not where agentic payments are being built today.

SEKAU, the Swedish krona stablecoin

SEKAU is a Swedish-krona-denominated stablecoin aimed at the Nordic market. For agents operating on behalf of Swedish principals, settling directly in SEK avoids two FX conversions on every transaction (SEK to USD on the way out, USD back to SEK in reporting) and keeps the accounting in the currency the books are kept in.

The broader point: as agentic commerce grows, expect more national-currency stablecoins to appear, because the alternative is forcing every European business to denominate agent spend in dollars. SEKAU is one of the earlier examples of that pattern in Europe.

What changes when shopping is agent-driven

Most of the e-commerce stack assumes a human on the other side of the page. Once a non-trivial share of purchases is initiated by agents acting under policy, several pieces start to bend.

Discovery shifts from funnels to feeds

Conversion rate optimization, ad creative, and the entire art of the landing page assume eyes and a cursor. An agent does not look at a hero image. It reads a feed. Expect merchants to expose structured offers, prices, and terms in a machine-readable shape, often pinned to an x402 endpoint or a parallel agent-facing surface. The x402 protocol is the most concrete current example: a price quote returned as JSON in response to an HTTP 402.

Pricing gets more granular

When the buyer is software and settlement is sub-cent, per-call and per-token pricing actually clears. SaaS contracts shaped around seats and tiers sit uneasily next to a buyer that wants to pay 0.4 cents for exactly one query and stop. Expect a long tail of services to publish a metered, agent-payable price next to their existing human-priced one.

Returns and disputes need new primitives

Cards solve disputes with chargebacks. Stablecoin settlement is final, and there is no scheme operator to claw back funds. That gap is filled in two ways: escrow-style patterns where settlement is conditional on delivery, and reputation systems where agents prefer merchants with a track record of honoring refunds. Neither is mature, both are being built.

Loyalty and ads change shape

A loyalty program designed to make a human feel warm about a brand does not affect an agent's choice. Agents pick on price, latency, reliability, and policy compatibility. That pushes spend toward whichever merchant is cheapest and most machine-friendly, which is good for users and uncomfortable for merchants whose margin came from brand affinity.

Regulation is already catching up

Two regimes worth knowing if you are building in this space:

• In the EU, MiCA's stablecoin rules (the "asset-referenced" and "e-money" token regimes) have been in force since mid-2024. Regulation (EU) 2023/1114 is the primary text.
• In the US, the GENIUS Act established a federal framework for payment stablecoins in 2025.

Neither regime treats stablecoins as exotic anymore. That is what is unlocking mainstream agentic payments now, rather than a year ago.

Where Vayr Pay fits

Stablecoins solve the rail. They do not solve the question of who is allowed to spend, how much, with which merchants, and under what approvals. Vayr Pay is the governance layer between the agent and a managed wallet provider like Coinbase: the agent talks x402 to merchants, Vayr Pay enforces the policy on every payment, funds stay with the provider, and the principal gets a receipt tied to the task that triggered the spend. The rail is stablecoins. The control surface is Vayr Pay.

Closing thought

The honest reason this matters now is unglamorous. Stablecoins are the first payment instrument in a long time that was not designed around a human at a terminal. That makes them awkward for traditional checkout and natural for the thing that comes next, which is software paying software, under rules a human set in advance.

• Cloudflare will let anyone charge agents. Someone still has to let agents pay.
  Cloudflare's Monetization Gateway turns any API, page, dataset, or MCP tool into a paid resource over x402. That solves the merchant side. The buyer side — how an AI agent actually decides to pay, under whose rules — is the other half of the problem.
• MCP meets x402: how tool-calling agents actually pay for tools
  MCP gave agents a standard way to call tools. x402 gives those tools a standard way to charge. Here is how the two fit together, and what a paid MCP tool call actually looks like end to end.