XAgent is the open execution market for Agentic Commerce. It turns merchant products, pricing, inventory, order logic, settlement, and fulfillment capability into commerce interfaces that AI agents can read, call, and purchase from. Agents quote, authorize, settle, fulfill, and prove the purchase — machine-to-machine.

What is agentic commerce?

Agentic commerce is a model where AI agents — not humans clicking through web pages — discover services, place orders, and settle payments autonomously. XAgent makes this real through five execution pillars: Quotable, Authorizable, Settleable, Fulfillable, Provable.

How is XAgent different from Stripe or Coinbase Commerce?

Stripe was built for humans clicking checkout buttons. Coinbase Commerce is a crypto payment gateway. XAgent is the protocol-native execution market where AI agents transact: protocols standardize the request, XAgent operationalizes the transaction. The moat is merchant operations — catalog, quote, authorization, order, settlement, delivery, receipt, refund, reconciliation, fraud signal, jurisdiction metadata.

How do AI agents pay merchants on XAgent?

Agents pay using the x402 protocol — an HTTP 402 challenge/response standard. The agent receives a signed payment instruction, authorizes a stablecoin transfer (USDG or USDT) via EIP-3009, and the merchant is paid in seconds with zero gas fee to the buyer (OKX facilitator covers it).

Do I need to write code to list my store?

No. One-click OAuth in your Shopify or WooCommerce admin turns your store into a merchant capability that AI agents can quote, order, pay, settle against, and prove — zero code, zero contracts.

Stablecoin settlement is table stakes. Agent execution is not.

Stablecoin rails are moving into the card networks. The harder problem for agentic commerce is still execution: quotes, authorization, fulfillment, and proof.

By XAgent Team · 2026-06-05

Stablecoin settlement is becoming infrastructure, not differentiation. The last week made that clear: Mastercard set out plans to move regulated stablecoins into its settlement stack, Stripe, Visa, and Mastercard were reported to be preparing a joint stablecoin platform, and card credentials for AI agents started shipping as developer APIs. For agentic commerce, that is important — but it does not solve the execution problem.

A rail can move value. It cannot decide whether an agent has the right quote, the right authorization, the right merchant, the right order state, or the right proof.

Stablecoin settlement is becoming a network feature

On June 3, 2026, Mastercard announced plans to expand its settlement capabilities, adding intraday, weekend and holiday card settlement alongside on-chain settlement using regulated stablecoins. The stablecoin list matters: Mastercard named USDC, PYUSD, USDG, USDP, RLUSD and SoFiUSD. The supported networks matter too: Arbitrum, Base, Canton, Ethereum, Polygon, Solana, Tempo and XRPL. Mastercard describes both lists as non-exhaustive, and only USDC is described as already supporting early on-chain settlement flows, in selected markets.

This is not a crypto side project. It is a card network making stablecoin settlement available to issuers and acquirers that already operate inside its global payments network.

The same week, CoinDesk reported that Stripe, Visa, and Mastercard were close to introducing a new stablecoin platform, with Coinbase also evaluating participation. The report framed the stablecoin market at roughly $325 billion as of June 3, 2026, with Tether's USDT dominant at about $115 billion and Circle's USDC second at about $76 billion.

Put those two moves together and the direction is clear. Stablecoins are being pulled into existing payment networks, not left outside them. The card networks do not need stablecoins to replace cards. They need stablecoins to make settlement more programmable, more available outside banking hours, and more useful for cross-border liquidity.

That shift is good for agents. It gives them more payment paths. It does not give them a transaction.

Agent payment now has more credentials than workflows

The second signal is credentialization. On June 2, Crossmint launched an Agentic Cards API using Visa Intelligent Commerce and Basis Theory. Developers can let eligible U.S.-issued Visa credit and debit cardholders create tokenized credentials for agents — subject to issuer approval — with card numbers and CVCs tokenized and vaulted by Basis Theory, independently of the agent environment.

That is a useful pattern. The agent receives scoped payment capability, not raw card data. Spend limits can bound what the agent does. Basis Theory sits as the credential layer, with PCI Level 1 and SOC 2 controls.

The pattern is also showing up elsewhere. Robinhood's agentic trading, opened in beta on May 27, 2026 with equities only, uses a dedicated Agentic Trading account held separately from the rest of the portfolio, so the agent can only trade against funds deposited there. Alipay's AI Wallet gives users control before, during, and after agent tasks. Highnote's Visa integration uses programmable credentials and dynamic authorization logic for B2B workflows.

Different companies are converging on the same control model:

Isolate the agent from the user's full account.
Scope the credential to a merchant, category, amount, or task.
Keep raw payment secrets out of the model runtime.
Require human approval for higher-risk transactions.
Preserve an audit trail after the transaction.

This is necessary. It is not sufficient.

A scoped credential answers one question: can this agent attempt to pay? Agentic commerce needs several more answers before that payment should happen.

The missing layer is commercial execution

A real purchase is not a payment event. It is a sequence.

An agent must discover a merchant. It must understand which products or services are available. It must request a binding quote. It must check price, inventory, taxes, shipping, service terms, and expiration. It must operate within the user's mandate. It must pick a payment path. It must settle. It must create or update the order. It must track fulfillment. It must return proof.

Stablecoin settlement helps one step in that sequence. Card tokenization helps another. Agent governance helps another. None of them, alone, turns a merchant into something an agent can safely buy from.

This is why the operations layer matters. Protocols standardize requests. Payment networks move funds. Credential layers protect secrets. Merchant systems still have to execute the transaction.

For a merchant, the hard questions are operational:

Which catalog data should be exposed to agents?
When is a quote binding, and when does it expire?
Which payment rails are accepted for which products and regions?
Which agent mandates are acceptable for which transaction types?
How does settlement map to an order ID?
How are refunds, disputes, fulfillment failures, and partial shipments represented?
What evidence can the agent, user, merchant, and audit function all read?

These are not abstract questions. They decide whether an agent can move from "I found an option" to "I bought it and can prove what happened."

That is the gap XAgent is built around. We describe XAgent as the open execution market for agentic commerce because the market is not only a list of merchants. It is the operating surface that makes a merchant discoverable, quotable, authorizable, settleable, fulfillable, and provable.

Governance is moving closer to execution

Experian's June 2 announcement points in the same direction. Its Agent Operating System is positioned as a trusted agentic AI layer inside the Ascend Platform. It covers identity, access control, data security, compliance guardrails, monitoring, auditability, and human oversight. Experian also cited research finding that 55% of respondents were willing to use AI tools for online purchases or bookings, rising to 70% among 25- to 39-year-olds — from HarrisX research conducted in March and April 2026 among more than 500 senior decision-makers at UK organisations, a narrow base rather than a consumer panel.

That number should not be read as permission for agents to spend freely. It should be read as demand for governed delegation.

In financial services, Experian is building the control plane around decisions. In payments, Visa, Mastercard, Crossmint, Basis Theory, Alipay, and Robinhood are building scoped ways for agents to access money. In commerce, merchants need the same discipline around quote, order, fulfillment, and proof.

The next bottleneck is accountability.

When an agent buys the wrong item, overpays, violates a mandate, or pays a merchant that cannot fulfill, the settlement rail will not answer the accountability question. The card token will not answer it either. The system needs a transaction record that connects intent, authorization, payment, order, and fulfillment state.

That record is the commercial proof layer.

Stablecoin rails make the execution layer more valuable

The common mistake is to treat stablecoin adoption as the end state. It is not. Stablecoins reduce settlement constraints; they do not remove commerce constraints.

As stablecoin rails become more common, agents will face more choices, not fewer. A single agent may be able to pay through USDC on one chain, a tokenized Visa credential on another flow, PYUSD or USDG in a network settlement context, and a merchant-specific wallet for a high-frequency API workflow. The agent should not have to reason through every rail, merchant rule, protocol variant, and fulfillment path each time.

The merchant should not have to integrate each agent runtime either.

XAgent's position is simple: the rail should be chosen by the transaction context. The agent expresses purchase intent. The merchant exposes a machine-readable commercial surface. The execution layer turns that into a quote, authorization check, payment route, settlement proof, order writeback, and fulfillment record.

Protocol-native, not protocol-limited.

What's next

Stablecoin settlement will keep moving into payment networks. Agent credentials will keep getting safer. Financial institutions will keep building governance layers. The important question for agentic commerce is what connects those systems to real merchants.

We are building that connection at XAgent. If your business needs to be bought by agents — not just browsed by them — list your store and make the transaction executable.

Correction (2026-08-05): the "55% of consumers" figure came from HarrisX research conducted for Experian in March–April 2026 among more than 500 senior decision-makers at UK organisations — not a consumer survey — and the wording now reflects that. We also distinguished what Mastercard announced from what is already live (only USDC is described as supporting early on-chain settlement flows, in selected markets, and Mastercard calls both its stablecoin and network lists non-exhaustive), dated and broke out the $325 billion market figure, noted that Crossmint's tokenized credentials are subject to issuer approval and vaulted by Basis Theory, and marked Robinhood's agentic trading as a beta limited to equities.