TL:DR
The last structural shift in payments moved commerce from the store to the browser. The one underway now moves it from the browser to the agent. Instead of a person clicking buy on a trusted website, an autonomous system discovers options, compares them, negotiates, and initiates payment under rules its owner set in advance. For consumer shopping this is already visible in AI assistants that assemble and place orders. For cross-border businesses, the more consequential version is quieter: procurement systems that source and settle with suppliers across borders, and software that pays other software by the API call.
This guide is written for the businesses that will have to support that shift rather than the consumers who trigger it: marketplaces, platforms, fintechs, and enterprises that already move money across borders and now have to decide how agent-initiated payments fit their infrastructure. It covers where agentic payments show up, the instruments agents settle with, the protocols and network programs competing to standardize them, and the unsolved trust problem that determines whether any of it is safe.
Why agentic payments matter now
The market forecasts are large and, importantly, they disagree on scope rather than direction. McKinsey estimates that AI agents could mediate 3 to 5 trillion dollars of global consumer commerce by 2030, with as much as 1 trillion dollars in orchestrated United States retail revenue alone [1][2]. Those figures capture a broad definition that includes B2B, logistics and payment infrastructure, which is why they sit far above narrower estimates that count only direct sales on AI platforms. The useful takeaway is not the exact number. It is that every serious forecaster now models agent-mediated commerce as a material share of digital spend within five years.
Adoption signals sit underneath the projections. Research from ICSC and McKinsey found that a majority of consumers had already used at least one AI tool as part of a shopping experience, and a large share had used AI to compare brands, prices or reviews before buying [2]. The behavior is running ahead of the infrastructure, and the gap between the two is where the next few years of payments investment will concentrate. Today's payment systems assume a human is present and clicking. When an autonomous agent initiates the payment, that assumption breaks, and with it the fraud models, authentication flows and dispute frameworks built around human behavior.
What agentic payments actually are
An agentic payment is a transaction initiated, and sometimes negotiated and completed, by an autonomous AI agent acting on behalf of a person or business under predefined rules and constraints. The defining shift is from card-not-present to what is better described as person-not-present. A human delegated authority in advance; the agent exercises it at the moment of transaction without that human in the loop.
That single change ripples through every assumption in the payment stack. Authentication was designed around a person proving they are present, through a device, a biometric, or a one-time code. Fraud models were trained on human behavioral patterns. Dispute resolution assumed four parties: the buyer, the issuing bank, the acquiring bank and the merchant. Agentic commerce inserts a fifth actor, the AI platform, and raises questions courts have not yet answered about who is liable when an agent buys the wrong thing or pays the wrong party. None of these are reasons to avoid agentic payments. They are the specification for the infrastructure that has to be built underneath them.
Where agentic payments show up
The consumer use case is the most visible. A shopper tells an assistant to find and book the cheapest direct flight under a set price, and the agent searches, selects, and pays with stored credentials inside the conversation. It is real, and it is where the largest protocols are focused. It is also the most crowded and the least differentiated for a cross-border payments provider.
The use cases that matter more for cross-border businesses are three:
B2B procurement and marketplace settlement. A mid-market brand's procurement system receives a restock signal, evaluates suppliers across borders, negotiates terms, selects the payment rail that fits on cost and speed, and settles, all under corporate policy rules. This collapses the traditional separation between sourcing, contracting and settlement. Payment terms and settlement speed become inputs to the purchasing decision itself rather than an afterthought handled by finance days later.
Cross-border routing and FX optimization. Agents that specialize in cross-border payments can monitor rates and corridor conditions and route each payment through the rail that is cheapest and fastest for that specific corridor and moment. For a business paying suppliers, sellers or contractors across many countries, an intelligent routing layer is where a large share of the value sits.
Machine-to-machine micropayments. The most structurally new case is software paying software with no human involved: an agent paying per call to consume another agent's API, a fleet of devices settling usage of compute or data. There is an acceptance side to all of this as well: as agents begin to check out on behalf of buyers, businesses selling into these flows will need to accept agent-initiated payments through their payment gateway as reliably as they accept human ones. This category needs rails that are low cost, programmable and able to settle around the clock, which is exactly where stablecoins and real-time payment networks fit better than card rails. Mastercard extended its agent program to machine payments in 2026, a signal that the networks themselves see M2M as the next layer rather than a fringe [4].
The instruments agents pay with
No single instrument wins across every agentic use case. The right answer is a multi-rail one, chosen per transaction. Cards bring universal acceptance, mature fraud protection and established dispute rights, and both major networks have moved to make card rails agent-ready. Their weaknesses for agentic commerce are cross-border cost and the fact that the rails were never designed for machine-scale micropayments. Stablecoins settle in seconds, operate continuously, and are programmable, which makes them a natural fit for cross-border B2B and M2M, though they carry on-ramp and off-ramp friction and thinner direct merchant acceptance. Bank transfers and real-time payment networks remain the backbone for high-value B2B, with real-time rails solving the speed problem domestically while cross-border interoperability is still maturing.
For a business moving money across borders, the practical conclusion is that the winning infrastructure is not a single rail but the layer that routes intelligently across all of them. That is the same problem cross-border providers already solve for human-initiated payments, applied to a new initiator. Stablecoin settlement in particular tends to offer the clearest return on the highest-cost corridors, where it collapses the correspondent-banking chain, a mechanism covered in depth in our settlement guide.
The protocols and network programs
2025 produced a wave of competing standards, and they fall into two groups: open protocols and card-network programs.
Three open protocols anchor the landscape. The Agentic Commerce Protocol, developed by Stripe and OpenAI, powers checkout inside ChatGPT and introduces shared payment tokens scoped by business, time and amount. Google's Agent Payments Protocol, AP2, centers on cryptographically signed mandates that create a verifiable record of what a user authorized, and it is payment-agnostic across cards, bank transfers and stablecoins. The x402 protocol from Coinbase revives the dormant HTTP 402 status code to enable stablecoin payments directly over the web, which makes it the most natural fit for machine-to-machine use where both parties are API-connected [5].
The card networks have moved in parallel. Mastercard launched its Agent Pay program in April 2025, built on agentic tokens that bind a tokenized card credential to a specific agent and policy, and completed its first live agentic transaction in September 2025 [3]. Visa's Intelligent Commerce gives developers tokenized credentials, spend controls and authentication designed for agents, and its later Intelligent Commerce Connect packages support for several agent protocols into a single merchant integration, with AI platform partners including Anthropic, OpenAI, Microsoft and Perplexity [6]. The two approaches are converging rather than colliding: open protocols increasingly handle agent identity and intent, while network tokens handle the credential and settlement, so a single agent payment may travel through both.
The trust problem: Know Your Agent and verifying the payee
The foundational challenge is not the rails. It is trust. Two questions have to be answered before an agent payment is safe: is this agent genuinely authorized to act for this person or business, and is the party being paid the party the payer intended.
The first question is converging on a concept the industry calls Know Your Agent, extending KYC and AML frameworks to cover the identification, registration and monitoring of agents. Network programs already require agents to be registered and verified before they can transact, and mandate-based protocols create a verifiable chain linking a user's intent to an agent's action. For a cross-border provider the burden is heavier, because agent identity has to hold up across jurisdictions with different rules.
The second question, confirming the payee, is where regulation is already moving and where the agentic case is sharpest. When a human sends a payment, they can eyeball the recipient. When an agent initiates a payout, no one is checking the beneficiary details at the moment money moves. Europe has made this check mandatory: under the Instant Payments Regulation, Verification of Payee has applied to euro-area payment providers since 9 October 2025, with providers outside the euro area required to comply by 9 July 2027 [7][8]. The mechanism is a name-and-account consistency check that returns a result such as match, close match or no match before a transfer is authorized, and it exists specifically to reduce fraud and misdirected payments [9]. These beneficiary-integrity controls sit alongside existing cross-border obligations such as the Travel Rule, which already requires originator and beneficiary information to travel with a payment, and together they point at a future where verifying who is paid is a default step rather than an optional one. The same control that regulators are mandating for human payments is precisely what agent-initiated payouts need, because the agent cannot supply the human judgment the check replaces.
Fraud, liability and disputes
Agentic commerce widens the fraud surface. Prompt injection can manipulate an agent into unauthorized purchases or redirected payments. Generative tools make synthetic identities cheaper to produce. A distinct problem is first-party dispute: a consumer authorizes an agent transaction and later disputes it, exploiting the ambiguity of whether the agent correctly interpreted intent. Because the agent followed its technical authorization, the usual evidence for resolving a dispute is weaker.
Liability is genuinely unsettled. Traditional card disputes involve four parties. Agentic commerce adds the AI platform as a fifth, and courts have not yet decided how agency law applies to an autonomous system that acted within its instructions but produced a bad outcome. For cross-border transactions the complexity compounds, because investigation and resolution already span jurisdictions before an agent is added. The practical response for infrastructure providers is to make agent actions accountable: keep the authorization record, the mandate chain and the verification result attached to each transaction so that when something goes wrong there is an evidence trail to reason from.
Compliance and the regulatory picture
Regulation is catching up unevenly. The European Union's AI Act brings additional oversight to higher-risk AI systems, a category regulators are actively mapping onto autonomous financial agents. Data privacy regimes apply with full force, because agents require deep access to financial and behavioral data, and cross-border data transfer adds another layer. There is no harmonized global framework for agent-specific payments yet, which means a provider operating across many markets has to build for the strictest applicable regime rather than the average. This is the same discipline cross-border payments already demand, which our guide to the global payment licensing landscape covers in depth. The businesses that will move fastest are the ones that already hold multi-jurisdiction licensing and can extend existing know-your-business and compliance processes to cover agents rather than building them from nothing.
Why B2B cross-border is the underserved segment
The largest protocols and the most visible products are aimed at consumer checkout, where the volume is high and the transactions are simple. B2B cross-border is the opposite: fewer transactions, higher value, more complexity, and a genuine need for specialized settlement, compliance and multi-party capabilities. That is exactly why it is underserved. Consumer-focused protocols do not solve corridor-specific routing, local settlement in a supplier's currency, or the compliance overhead of paying counterparties across many jurisdictions, the kind of corridor-level depth explored in our guide to stablecoins in emerging markets.
For a business paying suppliers, sellers and contractors across borders, the agentic layer that matters is the one that can take an agent's instruction and route it through the rail that performs best for that corridor, whether that is a local rail, a real-time network, or stablecoin settlement, while confirming the payee and keeping the whole action auditable. The initiator is new. The underlying cross-border problem is one that specialized providers have been solving for years.
Where Tazapay Fits
Tazapay is not selling a live agentic product, and the honest position in 2026 is that the category is still forming. What is live is the infrastructure that agent-initiated cross-border payments will need underneath them. Tazapay's cross-border payout network reaches beneficiaries in more than 170 countries, with the large majority of payouts completing in under 15 minutes, and settlement available through local rails, SWIFT in a wide range of currencies, and stablecoin settlement provided by Tazapay Canada Corp. for the highest-cost corridors. That multi-rail routing is the exact capability an intelligent agent needs when it selects a settlement path per corridor.
The trust layer is live too. Tazapay offers payee verification as an opt-in add-on that does exactly this: before a payout is initiated, a business can confirm the beneficiary's bank account and the account holder's name in real time, receiving a result that flags a strong match, a partial match or no match across a broad set of corridors. It runs through the dashboard and the API and is non-blocking, so it adds a fraud and error check to agent-initiated payouts without disrupting existing operations. In an agent-initiated flow, where no person is eyeballing the beneficiary, that check is the difference between a payout that reaches the right party and one that does not. Combined with named accounts and payments made in the actual on-behalf-of originator's name, these are the building blocks a cross-border business can rely on today while the agentic standards mature around them.
The shift to agent-initiated commerce is early, and most of the headlines will keep pointing at consumer checkout. The quieter, more valuable work is in cross-border B2B, where the rails, the compliance and the payee-verification controls that make agent payments safe are being built now. The businesses that get there first will not be the ones that waited for the standards to settle. They will be the ones whose cross-border infrastructure was already sound when the initiator changed.
Sources
[1] McKinsey & Company. "The automation curve in agentic commerce." January 2026.
[2] Retail Dive (ICSC / McKinsey). "US agentic commerce revenue forecast to reach $1 trillion by 2030." May 2026.
[3] Mastercard. "Mastercard unveils Agent Pay, pioneering agentic payments technology." April 2025.
[4] Mastercard. "Mastercard launches Agent Pay for Machines." June 2026.
[5] Digital Commerce 360. "Visa and Mastercard both launch new agentic AI payments tools." 2025.
[6] TechInformed. "Visa opens one integration for AI agent payments." April 2026.
[7] European Commission. "New EU rules make instant euro payments faster and safer." October 2025.
[8] EY. "Why EU payment firms must act now on Verification of Payee requirements." 2026.
[9] European Payments Council. "Verification of Payee Scheme Ruleb
