Visa and Artemis Release Agentic Payments Report, Highlighting Growth in Machine-Driven Commerce

Written by: Will A Wang

In July 2026, Visa and on-chain data firm Artemis jointly released "Agentic Payments from the Ground Up," commissioned by Visa. Three months earlier, Visa’s Chief Product and Strategy Officer, Jack Forestell, had publicly called Agentic Web the greatest growth opportunity he’d seen in over two decades of payments—immediately followed by the payment network funding its own data provider to tell the market: machine payments are real, here’s the data, and here’s where the gaps lie.

The nature of this document is closer to a strategic narrative’s data component than neutral research. The report’s greatest value lies in Artemis’s first systematic cleansing of on-chain data for x402 and MPP—revealing that approximately 90% of on-chain transaction volume consists of wash trading and testing. Its most restrained aspect is that, after listing issues such as responsibility allocation and regulatory classification—areas in which Visa itself is directly involved—it stops short of further analysis.

Previously, Dukey Fintech addressed three issues not expanded upon in this report across three separate studies: the distinction between substitution and creation in Agentic Commerce, the regulatory gap in assigning responsibility after Agent errors, and why machine economies are inherently immune to territorial regulation. This article takes the Visa report as its foundation and builds upon where it leaves off—everything introduced with “Dukey Fintech believes” represents our own perspective.

When Visa was founded, its goal was not merely to create a card network, but to build a global system for the secure exchange of value. Today, Agentic Commerce is emerging as the new frontier of this system—software programs are beginning to autonomously discover, evaluate, and purchase goods and services, transforming payments from a human-oriented settlement tool into a foundational capability embedded within machine workflows. In response to this demand, new protocols such as x402 and MPP have rapidly emerged, leading to the convergence of card networks, stablecoins, and machine payment protocols.

1.1 Two Types of Transactions

Agentic Commerce is a form of commerce in which software programs autonomously discover, evaluate, and execute transactions. It is not merely automating traditional e-commerce; rather, it enables software to become an independent economic agent capable of procuring, invoking resources, and making ongoing payments. Based on current applications, it can be divided into two types of transactions.

Large-scale agent transactions (Macro): Agents act on behalf of individuals or businesses to complete traditional procurement, essentially executing human commercial intent—e.g., an IT agent automatically adds or removes SaaS seats based on employee onboarding and offboarding, or a travel agent books flights and hotels according to budget and policy. These transactions involve higher amounts and resemble traditional e-commerce in terms of merchants and payment channels.

The real new challenge is: authorization occurs before the transaction, execution happens without supervision, and the system must prove who the agent represents, what they are allowed to purchase, and where the budget limits lie.

Micro Transactions: Agents autonomously purchase data, compute power, models, and APIs to complete tasks—for example, a research agent pays a few cents to multiple data providers to retrieve and cross-verify financial reports. These transactions are high-frequency, low-cost, and occur instantaneously, with no accounts, subscriptions, or long-term contracts required between buyer and seller. Upon discovering a service, the agent immediately evaluates, pays for, invokes, and departs. The demands for millisecond-level response times, extremely low fees, and machine identity are the direct catalysts for the emergence of new protocols such as x402 and MPP.

DuKe Fintech believes the distinction between these two types of transactions can be further refined: Macro represents a substitution relationship—taking share from existing human purchasing behaviors; Micro represents a generative relationship—these transactions did not previously exist, not due to lack of demand, but because human decision-making costs exceeded the transaction value. Agents reduce decision-making costs to near zero, enabling transactions to emerge spontaneously. This also means that all market predictions measuring the speed at which agents take over human shopping only capture the first half.

As for the size of the other half, Circle CEO Jeremy Allaire provided a reference point in The Agentic Economy: companies are essentially “organizations with logos,” where labor costs account for a quarter to half of revenue—and nearly all of it in knowledge-based firms. This cost is precisely what AI targets. When each function is refined into a clearly defined skill—a skill so clean it can be orchestrated internally within a company—it is also clean enough to be discovered and hired externally. An open agent labor market doesn’t need to be created; it emerges as a byproduct of millions of companies optimizing themselves.

At that point, the smallest pricable unit was pushed finer than internal corporate divisions: a single verification, a single API call, a segment of translation—each could be priced, traded, and settled individually. In other words, microtransactions measure not shopping, but the hiring of machine labor—in Allaire’s words, models become cost items, and agents become businesses.

1.2 Why Is It Exploding Now?

The HTTP 402 "Payment Required" status code was incorporated into internet standards as early as 1997, but remained unimplemented for years: traditional payment systems had high fixed costs, making micropayments commercially unviable, so the internet long relied on advertising and subscriptions for monetization.

Du Ke Fintech believes that high settlement costs are only half the problem; the deeper half lies in the psychological transaction cost of human decision-making—for a payment to be worthwhile, the cost of deciding must be lower than the payment amount itself. Making a human decision over a few cents whether it's "worth it" is inherently worth several cents. Micropayments have not failed in human economics because they were poorly executed—they are mathematically infeasible; agents have inverted this inequality.

The real turning point comes from AI Agents. The report anchors the crossing of the capability threshold in mid-2025—with the emergence of models like Claude 4.5 and GPT Codex 5.2—when Agents begin autonomously discovering unfamiliar APIs, evaluating prices, and deciding whether to pay, without requiring manual approval for each transaction. For the first time, programmatic, on-demand purchasing of data and compute becomes widespread. Meanwhile, specialized blockchains like Tempo offer transaction fees in the thousandth-of-a-cent range and finality within 500 milliseconds, while Gas costs on Base and Solana drop to fractions of a cent.

However, the report emphasizes that dedicated chains have existed for years without spontaneously generating a pay-per-use API market—it was Agents that created the demand, and infrastructure merely enabled that demand to be commercialized.

The report also issues a straightforward warning to traditional institutions: the risk of inaction is disintermediation. If a small number of platforms control both the demand entry points and transaction routing for agents, they can divert payment flows outside existing networks—particularly in the micropayment sector, where the card-based economic model is weakest. Institutions that participate only through large-volume agent transactions will miss out on the highest-frequency segments of the agentic economy.

The difference between traditional e-commerce and micro-machine transactions isn't just smaller transaction amounts—it's a fundamental shift in the entire transaction logic.

Subscription and pay-as-you-go models can alleviate some issues, but they still require prior registration, negotiation, and partnership establishment, failing to support the ad-hoc transaction model of agents: “discover interface—evaluate price—pay immediately—execute call.” Therefore, the value of the new protocol is not to replace credit cards, but to add a payment architecture on top of existing payment systems that is optimized for machine-to-machine interactions.

DuKe Fintech believes that the significance behind this table goes far beyond "smaller amounts, higher frequency": Agents are not merely faster consumers—they are an entirely different species of consumption.

• Atomization of consumption—humans bundle their purchases through monthly subscriptions, annual fees, or one-time purchases, essentially using upfront payments to simplify decision-making; agents have no such need, as they can precisely calculate the marginal value of each invocation;
• Decision streaming—humans have tolerance for delays in payments, such as waiting while filling out forms, entering passwords, or confirming transactions; agents, however, can perform hundreds of tool calls within milliseconds, and each step may incur a cost.
• Dehumanization of the主体—The most fundamental change is that the payer is no longer a human; every design in traditional payment systems—from KYC to CVV verification to chargeback mechanisms—is built on the assumption that "the user is a human." The first two are efficiency issues that engineering can resolve; the third is an assumption issue requiring a complete overhaul of the system.

Before entering the two protocols, let’s first revisit the map previously drawn by DuKe Fintech.

Today, Agent spends money through three paths; the essential difference among these three paths lies in where the responsibility resides:

• Wallets and credits: Users deposit funds first and spend until depleted, prepaying their responsibility—at the cost of being locked into an ecosystem loop. This is the fastest to implement but often overestimated; most so-called "AI payments" are merely shopping assistants within the ecosystem.
• Track-based, governed by the network rules of Visa and Mastercard; the fifty-year framework of disputes remains in place; TAP, AP2, ACP, and UCP from Chapter Five are all under construction on this track;
• On-chain direct payment, wallet-to-wallet, with finality upon receipt; responsibility is either unassigned or replaced by code and reputation.

The data section of this Visa report focuses on the two most important examples along the third path. x402 launched in May 2025, incubated by Coinbase and Cloudflare, and was transferred to the Linux Foundation for open governance in April 2026; MPP launched in March 2026, jointly developed by Stripe and Tempo, with Visa participating in protocol development, and is currently undergoing the IETF standardization process.

3.1 Common Foundation: HTTP 402

Both protocols are based on HTTP 402. The typical flow is as follows:

1. The client requests resources from the server;
2. The server returned a 402 status code, along with the price, currency, and payment terms;
3. The client evaluates the conditions and submits payment proof;
4. After verifying the payment, the server will grant access to the resources.

Payments are directly embedded in HTTP requests and responses, eliminating the need to redirect to a checkout page or requiring buyers and sellers to establish prior account relationships. A server only needs to return a machine-readable payment request to function as a payment interface.

3.2 Key Differences

x402 introduces a facilitator in transactions: after the facilitator advances payment on behalf of the buyer, it verifies the transaction conditions and completes settlement. The facilitator layer handles payment verification, merchant onboarding, risk control, and fund settlement, and charges fees accordingly; thus, x402’s industrial value is more likely to be captured by facilitators and settlement service providers.

MPP emphasizes direct settlement between buyers and sellers, with its underlying Tempo protocol enabling finality in approximately 500 milliseconds. It supports two modes: Immediate Charge, which delivers payment instantly and allows a single transaction to be atomically split among up to ten recipients, with market settlement handled directly at the protocol level; and Session Payment, which involves depositing funds into escrow, signing off-chain vouchers based on consumption, and settling in bulk at the end while returning any remaining balance—ideal for ongoing services where the total cost cannot be determined in advance, transforming “monthly subscriptions” into task-level, real-time, and cancellable cost structures. By minimizing intermediaries, MPP’s long-term value is more likely to accumulate in upper-layer applications, aggregation platforms, and developer tools.

Both protocols are still in early stages, with potentially converging functionalities, but differing trust models and locations of value accumulation.

3.3 x402 Data: Live for 11 months

First, clarify the number mentioned in the report. The on-chain raw data for x402 shows a cumulative transaction volume of $135.7 million and 178.3 million transactions—but after filtering out wash trading and test transactions, only:

• $15 million adjusted volume
• 109.6 million valid transactions
• 422,000 valid buyers
• Approximately 5,300 active merchants

(Agentic Payment from the Ground Up)

The real market is an order of magnitude smaller than on-chain surface data. This is why discussing Agentic Commerce must begin with data cleansing—Noah Levine, partner at a16z crypto, highlighted a more fundamental issue: for the same x402, 30-day trading volume is reported as $24 million by x402.org, $3 million by Allium, and under $2 million by Artemis—a tenfold difference among the three sources. When there is no consensus on how to measure this market at all, every growth narrative must be questioned—and this report uses the strictest of the three measuring sticks.

November 2025 saw a volume peak of approximately 38 million transactions and $5.15 million in adjusted gross volume. By March 2026, transaction volume had declined to around 2.1 million, yet gross volume remained at $1.64 million, with the average transaction value reaching a new all-time high since launch—x402 is no longer just an “a few cents API payment” protocol, but is expanding into broader value ranges.

(Agentic Payment from the Ground Up)

DuKe Fintech offers another interpretation of this curve: the boundary of x402 is defined by the potential for dispute over goods. Surviving transactions are concentrated on buying tokens, calling APIs, and renting computing power—goods that are consumed upon delivery and cannot be undone. x402 has no authorization, no recourse, and no dispute resolution channel; for instantly consumed goods, "naked" is precisely an advantage—any product with potential for error cannot enter a naked protocol.

Concentration: The top 1% of buyers account for approximately 90% of transaction volume, while only 0.02% of buyers contribute about 48%. A small number of professional institutions dominate the majority of business, a typical characteristic of an early-stage market.

Distribution: Base accounts for approximately 90% of valid transactions and 93% of processed volume. By transaction count, agent-to-agent service procurement is the largest category; by volume, a large share consists of unclassified long-tail interfaces—supply-side fragmentation remains high, and aggregation and discovery tools are still lacking.

(Agentic Payment from the Ground Up)

3.4 MPP Data: First 33 Days Since Launch

After the MPP update, cumulative transaction volume reached $25,000, with approximately 115,000 valid transactions, over 2,800 valid buyers, and around 90 valid merchants, averaging about 4,000 valid transactions per day. Instant settlement accounted for approximately 97% of transactions, with an average amount of $0.17; session-mode transactions had an average of $2.15 and were still primarily used for developer testing.

The current ratio of buyers to sellers is approximately 32:1, with demand growing significantly faster than supply. The report notes that the number of merchants on x402 during its first few months was at a similar level before surging sharply—DuKe Fintech believes that, based on this, a 32:1 ratio may not be a negative signal, but rather likely reflects a timing lag on the supply side.

4.1 The biggest issue is not payment, but trust

The payments industry has upgraded its authorization technologies six times over fifty years—from signature stamps and magnetic stripes with real-time authorization networks, to EMV chip cryptography, NFC, tokenization, and now Agentic Token—yet all these changes have only altered the verification methods, while always verifying the same thing: whether the person at the point of transaction is the genuine cardholder. Agentic Payment breaks not the methods, but the underlying assumption: the payer is no longer a human making real-time decisions, but a software program holding authorization and acting autonomously; authorization and transaction are now temporally separated, with intent expressed in natural language—which cannot be validated by risk control engines. This introduces four types of risk:

1. Incorrect procurement: The agent misunderstands, selects the wrong service provider, or overpays. Humans can correct these errors before settlement, but an uncontrolled agent can generate a large volume of erroneous transactions in a short time;
2. Malicious attack: Prompt injection can alter the agent's execution logic, inducing unauthorized purchases or fund transfers. The higher the autonomy, the greater the potential loss if compromised;
3. Unclear responsibility: The authorizer, agency platform, model service provider, wallet service provider, and merchant may all claim they are not at fault;
4. Cascading failure: When an agent makes bulk purchases of services from other agents, the failure of any link in the chain may render payments already made upstream worthless.

x402 and MPP can address some issues with payments and delivery, but they cannot individually answer the question of who is responsible when an agent makes a mistake.

4.2 Current Solution

Protocol Layer: x402 settles payments only after intermediary verification of delivery; MPP manages risk through session custody; flexible limits and spending caps restrict single-instance losses.

Identity Layer: Many agents still rely on API keys and wallet addresses for authentication, which can only prove access rights, not historical履约 capability. Even Visa itself does not deny this—Forestell acknowledged in a public conversation in March 2026 that agentic transactions carry greater risk than e-commerce and mobile payments of that era: an additional agent is introduced, requiring identity, protection, verification, and more data to ensure security. The industry is addressing this gap from two directions—cross-chain identity and reputation standards within the crypto ecosystem, and the existing authorization, KYC, and fraud prevention capabilities of the card payment system.

Strategy Layer: Tools such as Coinbase Agent Wallet, Turnkey, Privy, and Safe provide whitelist, spending limits, and merchant category restrictions; the card system binds额度 and merchant restrictions to proxy identities via tokenized credentials.

4.3 Issues Still Unresolved

The industry still lacks three key infrastructures: a universal proxy identity standard bridging cryptocurrency and bank card ecosystems; dispute resolution, refund, and evidence mechanisms suitable for high-frequency microtransactions; and clear delineation of responsibilities among authorized parties, proxy platforms, model service providers, and merchants.

These gaps represent both the primary risk of Agentic Commerce and the area where traditional payment institutions are best positioned to fill.

DuKe Fintech believes that the three gaps背后 are manifestations of the same structural shortfall. In legal terms, delegated payment cannot avoid four fundamental questions: whether authorization is valid, where its boundaries lie, who is responsible for overstepping those boundaries, and who ultimately bears the loss—the framework of agency law for the past two millennia rests on these four questions. The first two are matters of proof: whether authorization exists and whether its boundaries are clear—cryptography can answer these. The latter two are matters of adjudication: who is at fault for overstepping and who bears the loss—only rules can answer these.

Today, nearly all players are focused solely on the first two questions: signature chains, audit trails, and cryptographic proofs of performance—fixing "what happened" at the cryptographic level, while saying nothing about "who should be held accountable for what happened." This is precisely what the dispute resolution company Chargebacks911 warns about:

The industry is building from the wrong end—while frontend authorization frameworks have all been launched, post-trade dispute infrastructure—how disputes are categorized, how responsibility is allocated, and how evidence is admitted—has barely been touched.

On the vacant lot of rulings, what was erected first was not law, but a price list: the conditions for liability transfer by card networks, the payout thresholds for insurance, Amex’s protection clauses for registered agents—all putting a price on “who pays when things go wrong”—they anchor the same variable. The precision of authorization boundaries is the price of responsibility.

Agentic Commerce requires coordinated protocols across communication, identity, authorization, payment execution, and settlement—no single standard currently covers the entire end-to-end process.

(Agentic Payment from the Ground Up)

Visa TAP enables merchants to verify that visitors are authorized agents of the user through cryptographically signed HTTP messages. It does not require merchants to rebuild their payment systems, but instead adds a machine identity layer on top of existing web pages and card networks—this is the most direct path for traditional payment networks to enter Agentic Commerce.

AP2 records purchase intent, specific cart details, and final payment authorization as cryptographically signed instructions, transforming natural language intent into verifiable, auditable structured authorizations while remaining neutral toward payment channels such as credit cards, bank transfers, and stablecoins. DuKe Fintech believes its ceiling lies here: budgets, slots, and time windows can all be captured in fields, but the reasons you didn’t voice cannot—authorization logs record what you said, not why you said it.

ACP vs. UCP: ACP, driven by Stripe and OpenAI, adopts a closed-market model where agent platforms and merchants pre-approve each other; UCP, led by Google and Shopify, enables agents to complete purchases directly within entry points like Search and Gemini, while merchants retain customer relationships. Both fundamentally compete for control of the agent commerce gateway—whichever side controls the process of agent discovery, transaction confirmation, and payment execution is likely to dominate the next generation of e-commerce traffic distribution.

Visa Intelligent Commerce aims to become a unified gateway: a single integration compatible with protocols such as TAP, x402, MPP, ACP, and UCP, offering tokenization, authentication, and programmable spending controls. This reflects the card networks’ strategic shift from merely securing card transactions to becoming the central hub connecting agents, merchants, and multiple settlement channels. At the Visa Payments Forum in June 2026, Visa even demonstrated a conceptual product enabling agents to make payments directly from the command line using tokenized credentials—Forestell’s goal being to “make cards the best payment method in the command line.” The card networks’ reach has now extended into the heart of machine-native transactions.

The long-term trend is integration, not replacement. MPP supports both stablecoins and bank cards; Visa has launched bank card specifications compatible with MPP; Stripe has integrated x402 stablecoin payments on Base. In the long run, large-scale agent payments will still rely on the mature merchant networks and dispute systems of bank cards, while machine-native micropayments are better suited for low-cost on-chain settlement. Both channels will coexist within the same workflow—this coexistence embodies the structural division of labor described in Chapter One: replacement and creation. The card network has not lost any existing categories, nor are new categories within its scope.

Both parties' public statements align with this. Forestell provided the shortest version of this positioning at the Visa Payments Forum: AI transforms the front end of commerce, stablecoins reshape the back end of finance, and Visa’s role is to make both ends operate securely, reliably, and globally; Tim Conard, the lead author of the report and Visa’s Head of On-chain Data, summarized the key point in the official accompanying article: Visa aims to build infrastructure that doesn’t force the ecosystem to choose sides or require starting from scratch.

The opposing side has offered a mirrored stance—Circle CEO Jeremy Allaire wrote in The Agentic Economy that on-chain doesn’t need to defeat Cardano in debate; it simply needs to be the place where the agent economy can natively operate. He even acknowledged that it’s a plausible equilibrium for stablecoins to handle funding while leaving payment processing and fees in the hands of companies like Visa. Both sides are leaving room for each other.

Du Ke Fintech believes that when these statements and reports are read together, the sequence becomes clear: Forestell’s position is the conclusion, and this report provides the supporting rationale—strategy comes first, data follows.

Beyond statements, there is a more honest record: investment history. Visa Ventures invested in Nekuda, which generates authorization documents, and Payman, which handles bank-side approval bottlenecks—the former’s authorization occurs at the moment of user-agent interaction, outside VisaNet; the latter’s funds never traverse the card network. One investment targets the upstream; the other targets outside the track. These two investments reveal that Visa understands its protocol only covers the segment “when money passes through me.” To understand what an institution wants you to believe, examine its protocol; to understand what it doesn’t believe itself, examine its investments.

Putting together the CLI concepts, investment records, and the stance of "not forcing a side," Visa's actual actions become clear.

DuKe Fintech believes that Visa is doing one of the most misunderstood things:

It is not competing with stablecoins; instead, it is quietly repositioning itself from a settlement network to an authorized infrastructure—VIC manages tokenization, TAP manages identity, verifiable records manage auditing, and the MPP card standard manages multi-track access, betting on all four fronts to cover every link in the authorized chain.

In the agent era, what Visa truly wants to sell is not card swiping, but the very fact that "this payment is legitimately authorized." The card network may not win the settlement in the machine economy, but it could win the authorization.

The original report’s scenario chapter consists primarily of a table listing eight industries and three subsections on patterns. Section 6.1, the industry breakdown, is an extended analysis by DuKe Fintech based on the report framework and does not represent the views of Visa or Artemis; Sections 6.2–6.4 are derived directly from the original report.

6.1 Eight Industry Scenarios

The most valuable aspect of Agentic Commerce is not the protocol itself, but the transactions that were previously unfeasible and are now made commercially viable by the protocol. These can be categorized into eight industries:

1. Corporate Procurement and Back-Office Operations: IT, administrative, finance, travel, and procurement agents complete recurring procurement and payments within established policies and budgets, relying more on mature card and expense management systems;
2. Sales and customer data: Sales agents purchase enterprise information, email verification, and CRM enrichment on a per-lead basis, comparing multiple providers to select only the most cost-effective option—transforming fixed subscriptions into task-based variable costs;
3. Research and financial data: Procure financial reports, on-chain data, and regulatory information around the same issue, and use multi-source cross-verification instead of relying on a single data provider—particularly suitable for high-value, low-frequency, long-tail data.
4. Content idea: The marketing agent pays five cents to each of ten image service providers, using only the best output—payment is embedded into the production and screening process;
5. Engineering and Computing Power: The code Agent temporarily procures GPUs, inference, testing, and sandbox environments; it exits automatically upon task completion, without requiring prior account setup or signing.
6. Security and Compliance: Pay-per-use KYC, sanctions screening, on-chain risk analysis, and deepfake detection. Most valuable and sensitive—legitimacy of data sources and accountability for results must be addressed simultaneously;
7. Sentiment and Monitoring: Continuously purchase social signals and competitor insights based on data volume, duration, or trigger count—more suitable for temporary projects than fixed monthly fees;
8. Maps and logistics: Continuously purchase route, inventory, and price data—pay for decision-making capability first, then decide whether to purchase real-world goods.

DuKe Fintech believes these eight scenarios will not arrive simultaneously; the order of implementation is determined not by technological maturity, but by who bears the risk: scenarios with protocol-backed guarantees (API, data, computing power) move fastest—delivering instantly, with clear billing and no refund disputes; nearly all real transactions on x402 and MPP today are concentrated here. Scenarios with enterprise-backed guarantees (B2B procurement, internal standardized spending) follow second—budget pools and approval workflows already existed in manual form before the advent of agents. Consumer retail scenarios with individual-backed spending are the slowest, as they require a complete consumer protection framework. The most visually compelling scenarios are the last to be implemented.

6.2 Pay-as-you-go Streaming

Not all services can be priced in advance: continuous computing power over hours, transcription billed by the minute, and real-time market data feeds cannot have a total price quoted before the transaction. MPP’s session-based payments are designed precisely for this—merchants earn based on actual delivery, and buyers are not required to commit to fixed subscriptions for fluctuating demands.

6.3 Agent Purchases from Agent

Agent-to-Agent services represent the most iconic use case of Agentic Commerce and have become the top category on x402 by transaction volume. An Agent does not need to possess all capabilities; instead, it outsources subtasks to specialized Agents—the primary Agent purchases search services from a retrieval Agent, which in turn buys reports from a data Agent, with analysis, compliance, and reporting each being procured in sequence. A single user request thus forms a multi-layered machine supply chain.

Du Ke Fintech believes that payments serve not only as a settlement tool but also as a coordination mechanism for resource allocation and specialized division of labor among agents. This explains why the number of transactions grows much faster than consumer spending: a single task may cost only a few dollars, yet be broken down in the background into dozens or even hundreds of machine payments.

6.4 Long-tail Services and "API as Merchant"

Machine payments have significantly lowered the barrier for digital services to become merchants: developers no longer need to build full SaaS products; they simply need to encapsulate their capabilities as machine-readable paid APIs. Many services that were previously impossible to commercialize independently have now become viable—such as single-document conversion, corporate registration queries in a specific country, single-image authenticity verification, and risk scoring for an address. This model is already being productized: Merit Systems’ AgentCash bundles wallet management, merchant discovery, and over 420 paid APIs into a single integration, compatible with Agent frameworks like Claude Code.

However, the low barrier to deploying interfaces also means that similar services are easily replicated—individual long-tail merchants struggle to build barriers; where the real barriers lie will be addressed in the conclusion.

7.1 How to Qualify Protocol Roles

In x402, intermediaries may temporarily hold funds, verify delivery, and transfer payment; in different jurisdictions, they may be classified as providing money transmission, payment processing, or custodial services. Without a standardized intermediary in MPP, another issue arises: in the absence of a clearly defined intermediary, who is responsible for KYC, anti-money laundering, transaction monitoring, and dispute resolution?

Cross-border transactions further amplify uncertainty: intermediaries, service providers, wallets, and infrastructure may be distributed across multiple countries, while existing rules are based on clearly defined transaction parties, clear jurisdiction, and lower transaction frequencies.

Circle CEO Jeremy Allaire provides a more thorough articulation of this in The Agentic Economy: the three layers of this economy—money, contracts, and working agents—are all software, all running on the internet, with none having an inherent geographic identity; the work performed by cloud-based models has "no home"; globality is not a feature added to the system, but a structural property of its very composition.

Du Ke Fintech believes that, following this line of reasoning, the fiat currency pathway fails here on two premises: the agent lacks legal personhood—which pertains to the human dimension—and the action lacks a physical location—which pertains to the spatial dimension. This is not a regulatory vacuum, but rather regulatory overload: a single action may simultaneously fall under multiple jurisdictions—such as the consumer’s location, the data subject’s location, and the market’s location—yet there is no definitive location to resolve jurisdictional conflicts. For payment institutions, compliance architecture may evolve from a back-office function into a source of competitive differentiation.

7.2 Stablecoin Regulation

x402 is heavily reliant on USDC, and MPP also depends significantly on on-chain settlement; therefore, stablecoin regulation directly impacts the speed of Agentic Commerce’s expansion. In the United States, the GENIUS Act has established a federal licensing and reserve framework for payment stablecoins; with clear regulatory status, stablecoins serve as a trusted channel for banks and acquirers to build upon. If other jurisdictions remain fragmented over the long term, the cost of cross-border deployment will rise significantly.

Allaire also made a distinction that helps explain why the U.S. is willing to embrace this framework: the channel through which money flows and the money flowing through that channel are two different things—the protocol layer is neutral and stateless, but the funds moving across it are still denominated in sovereign currencies and represent fiat liabilities issued under the jurisdiction of that country.

DuKe Fintech believes that, viewed through this lens, stablecoins are not the opposite of fiat currency—they are fiat currency. What is being replaced is not money itself, but the national banking channels that have welded money to national borders. Stablecoins win not by defeating fiat currency, but by liberating fiat from these national channels—for Washington, this is almost a dollar globalization strategy that doesn’t require a CBDC. The GENIUS Act is less an acceptance of something new and more an acknowledgment of this very strategy.

7.3 The Unique Challenges of AI Agents

Existing KYC and anti-money laundering rules are centered on human customers. Agentic Commerce must readdress: Who is being verified—the authorized agent, the agency platform, the wallet, or the model service provider? Who is responsible for monitoring the agent’s ongoing transactions? Who bears the obligation to report suspicious transactions and fulfill tax duties? How can thousands of cross-border microtransactions be audited and tracked?

Therefore, a traceable, auditable, and batch-reportable agency trading infrastructure will become a new business value point.

And vacuum is not abstract. The U.S. still lacks specific regulation for agent transactions; the binary "authorized or unauthorized" framework of Reg E/Reg Z is forcibly applied to gray areas—contract formation itself is not the issue, as the ESIGN Act of 2000 already recognized contracts formed by electronic agents; the vacuum lies after contract formation: there are no specific rules or precedents for allocating liability when agents act beyond their authority. The EU’s AI Liability Directive (AILD) was officially withdrawn in October 2025, just before large-scale commercial deployment of agents, due to the 27 member states’ inability to reach consensus; PSD3’s strong customer authentication was designed for humans holding smartphones—agents have no fingerprints.

All fill-in-the-blanks are private ordering: The rules most closely aligned globally and directly enforceable within payment chains remain card network rules—their enforcement comes not from courts, but from “I can decline this transaction”; Amex allocates responsibility through commercial commitments, and institutions like Munich Re have already begun offering AI liability insurance.

DuKe Fintech believes that the rules governing agent payments are being written by protocol specifications, card network rules, and policy terms—legislation is occurring outside of law.

For financial institutions, the most practical path into machine payments is hybrid settlement: retaining cards and existing merchant experiences on the front end, while using stablecoins on the back end to enable 24/7 settlement, faster payouts, and lower cross-border costs. At a higher level, x402, MPP, AP2, TAP, and ACP are currently independent; in the future, agents will need to automatically select authorization, identity, and settlement methods within a single workflow—platforms capable of cross-protocol recognition and routing will become the critical control layer.

Cross-border transactions will become the default feature of machine-to-machine trading rather than an edge case, but currency exchange, compliance, and cross-border settlement adapted to machine speed are not yet mature—this represents the clearest near-term opportunity in global payment networks.

The supply side will also change: currently, agents are primarily buyers, but as deployment costs decline, more agents will directly provide data, models, and professional services to other agents, leading to rapid expansion of supply. Overly刻 Fintech believes this forecast rests on a legal premise unmentioned in the report—the absence of legal personhood discussed in Chapter Seven becomes most concrete here: the foundation of the fiat account system is that “account holders must be natural persons or legal entities”; agents within the banking system are forever merely a set of API permissions on someone else’s account—they can spend on behalf of others but cannot hold their own funds, receive payments, or stake assets themselves. Spending on behalf of others is a payment issue and can be accommodated; holding one’s own funds is a matter of legal personhood, and the structure of fiat accounts fundamentally does not permit it. For agents to become merchants, they must either wait for the law to grant them legal personhood or adopt on-chain wallets—this is why the supply-side expansion of “agents becoming merchants” will likely occur first on-chain.

Agentic Commerce has moved from concept to real transactions: x402 and MPP are now processing actual volumes, with pay-per-query, streaming payments, and agent-to-agent services taking shape. While the new protocols address low-cost, machine-speed, and programmable payments, integration into the formal financial system still relies on the long-established identity, risk management, compliance, and dispute resolution capabilities of traditional payment institutions.

Returning to the map at the beginning of Chapter Three showing three paths, we can now define their respective boundaries. Degree Fintech believes that the three paths are not competing for the same market, but rather carving out their own domains based on the density of controversy: zero-controversy areas belong to chains and stablecoins, complex-controversy areas belong to cards, and avoided-controversy areas belong to wallets. The real battlefield lies at the intersections of these boundaries—agents first use credits to compare prices, then place orders via cards, and settle payments with stablecoins to suppliers, bringing all three paths together within a single transaction.

Because the supply side is easily replicable, long-term value will not remain with individual transactions or individual merchants, but will concentrate in three layers: traffic distribution (who controls access to agent discovery and transaction routing), identity and trust (who can verify agents, merchants, and authorization chains), and underlying settlement (who can complete global settlements cost-effectively, securely, and in compliance).

Traditional institutions have spent decades building up in the latter two layers; new protocols compete on speed, cost, and programmability—the future leaders will likely be platforms that combine both sets of capabilities. The ultimate competition is not about individual transaction fees, but about who can control identity delegation, transaction entry points, cross-protocol routing, and global compliance infrastructure.