Quick wins for a faster PC:
Repair Windows errors before they cause bigger problemsFix Now →Scan for outdated or missing drivers - takes under a minuteDriver Scan →Choosing where an AI agent sends a payment is only one part of making agent commerce work. A complete transaction also needs evidence of what the user delegated, credentials that limit what the agent can do, authorization at the point of purchase, movement of value, and a way to reconcile or dispute the result. Those jobs are being addressed by overlapping protocols and company services—not by one settled, universal system.
Why isn’t routing enough for agent commerce settlement?
Routing answers a narrow question: which available path should carry a transaction? It does not, on its own, establish that an agent had permission to buy a particular item, protect the user’s payment credentials, confirm that the merchant will accept the payment, or resolve what happens if the transaction goes wrong.
As an Amazon Associate I earn from qualifying purchases.
It also helps to distinguish authorization from settlement. Authorization is approval to initiate a payment under particular credentials and permissions. Settlement is the movement and finalization of value between the relevant parties. A token or mandate may let an agent initiate a payment without exposing the underlying account details, but the token is not itself the transfer of funds. Stripe describes network tokens as scoped to customer intent and passed to an agent; Mastercard describes Agent Pay as spanning credentialing, permissioning, transacting, and settlement.
Free tools Windows power users keep installed
One-click scans. No signup required.
In practice, agent commerce therefore depends on several connected functions: recording delegated intent, identifying the agent and counterparty, enforcing limits, authorizing a payment, selecting a rail, and handling reconciliation and exceptions. The International Monetary Fund describes orchestration, routing, compliance, settlement, and post-settlement monitoring as related elements in emerging cross-border designs. It presents these as design patterns, not a standardized architecture.
#1 Best Overall
How do AI agents pay for things?
A useful way to understand an agent payment is as a sequence of decisions and controls, rather than as a single protocol call:
- Establish what the user meant. The system needs a record of the user’s intent or mandate: what may be bought, for whom, under what limits, and potentially for how long.
- Provide bounded credentials. A credential, token, or other payment authority should let the agent act within the delegated scope without handing it unrestricted access to the user’s payment details.
- Check the agent, merchant, and transaction. The payment system must determine whether the agent and merchant are recognized and whether the attempted purchase fits the applicable permissions and fraud controls.
- Authorize and choose a payment path. The system obtains approval using an accepted payment method and routes the transaction to an available rail or processor. Routing can help choose a path, but it cannot create a missing permission or merchant acceptance.
- Move value and reconcile the outcome. The relevant rail completes or records the transfer. The parties still need to reconcile what was ordered, paid, delivered, refunded, or left unresolved.
These functions may be provided by different layers or bundled into a vendor’s service. Product descriptions that use words such as “agentic,” “orchestration,” or “multi-rail” do not by themselves establish which implementations, merchants, payment methods, or regions are supported in a particular transaction.
Which agent-payment approaches are competing?
The current approaches overlap, but they are not interchangeable protocols. Some emphasize intent evidence or agent identity; others focus on credentials, network authorization, or machine-oriented payment flows. The descriptions below reflect what the named organizations say their initiatives cover, rather than independent confirmation that every capability is broadly available.
Rank #2
| Approach | What its source says it covers | Questions to check for a real deployment |
|---|---|---|
| Google AP2 | Google announced the Agent Payments Protocol with more than 60 participating organizations and described it as an effort to shape agent payments, with partner statements emphasizing security, trust, and interoperability. | What intent or mandate evidence does the implementation carry? Which rails and implementations are actually available for the relevant geography and use case? |
| Visa Trusted Agent Protocol and Intelligent Commerce | Visa describes network trust, fraud management, and authorization capabilities for agent-initiated transactions. Its 2026 materials also describe agent directories, credentialing, and token signals. | How are agents and merchants recognized? Which controls follow the credential into authorization, and which merchants and issuers support the relevant features? |
| Mastercard Agent Pay and Agent Pay for Machines | Mastercard describes agent credentialing, permissioning and spend limits, programmatic transactions, and multi-rail settlement across cards, accounts, and stablecoins. Its June 2026 announcement presents Agent Pay for Machines as supporting those functions. | What is available to the specific participant? How are limits enforced, and which rails, counterparties, and dispute rules apply? |
| Stripe Shared Payment Tokens and network tokens | Stripe says permitted agents can initiate payments without receiving underlying credentials, with network tokens scoped to customer intent. In March 2026, Stripe said Shared Payment Token support was expanding to network-led agent payments and buy now, pay later methods, with capabilities rolling out. | Who holds the token? How is its scope represented and revoked? What does “supported” mean for this merchant, agent, and payment method at the time of use? |
| x402 and Machine Payments Protocol (MPP) | Visa and Artemis describe x402 as an open protocol with activity since May 2025 and MPP as a newer design with more than one settlement rail. Their report describes small average payment values. | Are the transaction amounts low enough that card economics become a problem? Which asset, chain, fiat conversion path, fees, liquidity assumptions, and finality conditions apply? |
Comparing these options means looking beyond the protocol or routing feature. Six useful axes are delegated intent and authorization; credential custody and scope; merchant or counterparty reach; settlement rail, currency, and finality; per-transaction economics; and the treatment of disputes, refunds, reversals, and accountability. A partnership announcement or a common vocabulary does not establish universal interoperability across those axes.
Which settlement rail fits an agent transaction?
There is no single best rail for every agent purchase. Visa and Artemis argue that card networks can suit proxy purchases and larger merchant transactions, while stablecoins may suit machine micropayments. That is the report authors’ analysis, not a universal rule: the fit depends on transaction size, counterparties, fees, currency conversion, settlement expectations, and the remedies available if something goes wrong.
Fixed card fees can outweigh the value of a sub-dollar payment, which is one economic reason cited for exploring blockchain-based settlement. Lower costs in some designs do not make a payment risk-free or guarantee low total costs. A buyer or operator still needs to account for the applicable network or protocol fees, liquidity, conversion into the currency a merchant accepts, and the meaning of finality on the chosen path.
A single agent task can also span more than one rail. For example, an agent might make a small machine-to-machine payment as one part of a broader purchase that ultimately uses a conventional merchant payment method. In that case, the system needs a way to connect the small transfer to the larger order and to decide what happens if one leg succeeds while another fails. Mastercard’s description of multi-rail settlement is a vendor claim about its service; it should not be read as proof that all rails or counterparties already work together everywhere.
What do reported x402 and MPP figures show?
Visa and Artemis reported roughly $15.0 million in adjusted volume across 109.6 million x402 transactions since its May 2025 launch in their 2026 report snapshot. The same report put MPP at about $25,000 across roughly 115,000 transactions in its first few weeks after a mid-March 2026 launch. Those figures are source-reported snapshots, not audited or independently verified measures of adoption.
The MPP total covers only its early launch period, while the x402 figure covers activity since May 2025, so the two are not equivalent observation windows. Neither figure alone establishes how many transactions were completed purchases, how much value ultimately settled, or whether the systems are broadly accepted by merchants. Their reported scale is consistent with experimentation in low-value machine payments, but it does not settle questions about reliability, reach, or commercial maturity.
Visa’s June 2026 materials also cite an approximately $7 billion annualized stablecoin settlement run rate across VisaNet as of March 2026. That is a run rate, not completed annual volume. Visa separately says more than 160 stablecoin-linked card programs were live or in development globally; that combined count must not be read as 160 active launches.
Independent reader supportYour contribution helps us test, update, and keep practical guides available for everyone.Can AI agents make payments safely?
They can be designed with controls, but safe payment execution is not guaranteed by routing, tokenization, or a protocol label alone. The central issue is delegated authority: if an agent buys the wrong thing or a prompt redirects its spending, responsibility may be contested among the user, agent platform, model provider, and merchant. Visa and Artemis note that existing legal and regulatory frameworks were not written for this delegation model and that clear precedents may not be available.
PC Slower Than It Used to Be?
A free scan shows the junk files, broken settings and background clutter dragging Windows down - then fixes them in one click.Free scan · Windows 10 & 11Crashes, No Sound, or Screen Glitches?
Random freezes, missing sound and display glitches usually trace back to one bad driver. Find and replace yours safely.Free scan · under a minuteBefore enabling an agent to spend, a buyer or service operator should be able to answer practical control questions:
Best Value
- Brand: Generic
- NEW-Who Not How: The Formula to Achieve Bigger Goals Through Accelerating Teamwork
- What exact purchases, merchants, purposes, or amounts may the agent authorize?
- Can the permission expire, be revoked, or require confirmation above a threshold?
- Can the agent access a scoped credential rather than the underlying payment credentials?
- Which party records the user’s intent and the agent’s actions in a form that can be examined later?
- What evidence is retained if the user disputes the purchase, the merchant disputes delivery, or part of a multi-rail transaction fails?
These questions are not merely about fraud detection. A system may correctly identify an agent and still execute a transaction that exceeds what the user intended; conversely, a valid authorization does not prove that goods or services were delivered as expected.
Why are disputes and reversals especially difficult?
Traditional payment disputes assume a recognizable purchase, a human-speed decision, and evidence that can be gathered within familiar windows. Agent commerce may involve rapid transactions between services, or a chain in which one agent’s payment enables another agent’s purchase. Visa and Artemis identify the resulting evidence and unwind problem as unresolved: there is not yet a settled method for reversing some chains of agent-to-agent payments.
That uncertainty matters when selecting a rail. A fast or final transfer may be operationally useful, but it can make recovery harder if the underlying purchase is unauthorized, duplicated, or not fulfilled. A conventional dispute mechanism may offer different remedies from a stablecoin or machine-payment flow. The applicable refund, reversal, and accountability rules therefore need to be understood for the actual payment path, not inferred from an orchestration layer’s ability to select it.
How should a business evaluate an agent-settlement system?
Start with the transaction the business actually needs to support, then test each layer against it. The IMF’s treatment of payment orchestration and settlement as emerging design patterns is a useful caution: current initiatives are developing, and announcements do not establish a settled architecture or a winner.
- Define the delegation. Specify what the agent may buy, its spending ceiling, permitted counterparties, and whether a person must approve exceptions.
- Trace credential custody. Identify who issues and holds the credential, how its scope is enforced, and how it can be revoked.
- Verify reach. Confirm support with the specific agent, merchant, issuer, processor, and payment method—not just a vendor’s general statement about compatibility.
- Model the entire payment cost. Compare the transaction’s size and frequency with fees, conversion, liquidity, and settlement conditions on the available rails.
- Write the failure path. Establish who handles duplicate charges, failed fulfillment, partial completion, refunds, reversals, and evidence retention.
- Check availability claims. Distinguish a protocol announcement, pilot, or rollout from production support for the particular region and transaction.
A system is not ready for a use case merely because it can route an agent’s request. The implementation also has to show that the request was authorized, the credential remained within scope, value reached the intended counterparty, and a workable path exists when the transaction does not go as planned.
Quick Recap
Product prices and availability are accurate as of the date/time indicated and are subject to change. Any price and availability information displayed on Amazon at the time of purchase will apply.
