State of Agentic Payments Q1 2026 | AXD

The Quarter in One Sentence

'In the first quarter of 2026, every major payments company announced infrastructure for machines to spend money - and not one of them announced infrastructure for humans to trust the machines spending it.'

That sentence contains the entire story of Q1 2026. Everything that follows is detail. Stripe published a maturity model and a protocol. Mastercard completed the first authenticated agentic transaction. Visa launched a platform for AI-ready credentials. Alipay processed 120 million agent-initiated transactions in a single week. FIS declared "agentic commerce" the word of the year. JP Morgan counselled patience. Razorpay built agentic payments into ChatGPT. Coinbase opened AI wallet infrastructure. Venture capital poured $4.2 billion into agentic AI companies in ninety days.

The payment rails are being built at extraordinary speed. The trust architecture those rails require is not being built at all. This is the defining tension of the quarter, and it is the defining tension of agentic commerce as a field. The industry is solving the wrong problem first - not because the payment problem is harder, but because it is familiar. Trust is unfamiliar. Trust requires a design discipline that does not yet exist inside any of these organisations. That discipline is Agentic Experience Design. The AXD Institute's analysis of AI agent payments design and agentic banking examines these trust gaps in detail.

This essay is the first quarterly briefing from the AXD Institute on the state of agentic payments. It synthesises the developments, identifies the patterns, names the gaps, and offers the design perspective that the payments industry has not yet developed internally. It is written for product leaders, payment strategists, and design practitioners who need to understand not just what happened this quarter, but what it means for the systems they are building.

The Scoreboard

Before analysis, the facts. Q1 2026 produced more concrete agentic payment implementations than the entire preceding year. The following table captures the key developments in chronological order, stripped of marketing language and reduced to what actually happened.

Eight major developments in ninety days. Every one of them addresses the same question - how do machines pay for things? - and not one of them addresses the prior question: how do humans trust machines to pay for things on their behalf? The scoreboard reveals an industry moving with remarkable speed in a single direction. The direction is infrastructure. The missing direction is trust.

Three Philosophies of Agent Identity

The most consequential pattern of Q1 2026 is not any single announcement. It is the emergence of three fundamentally incompatible philosophies about what an agent is in a payment context. These are not technical disagreements. They are ontological ones. And they will determine the shape of agentic commerce for the next decade.

Philosophy One: The Agent as Entity (Mastercard)

Mastercard's Mastercard Agent Pay treats the agent as a new kind of participant in the payment network. The agent receives its own Agentic Token - a credential that identifies it as a distinct entity, separate from the human who delegated authority to it. When the Westpac transaction completed in New Zealand on February 17th, the payment network could see that an agent was transacting. The agent had visibility. It had identity. It was not pretending to be the human.

This is the most radical position any card network has taken. It implies that the payment system needs to accommodate a new category of participant - not a human, not a merchant, not an acquirer, but an agent. The downstream consequences are enormous. If agents are entities, they need governance frameworks. They need liability models. They need reputation systems. Mastercard has built the identity layer. Everything else remains unbuilt.

Philosophy Two: The Agent as Extension (Visa)

Visa's Intelligent Commerce takes the opposite position. The agent is not a new entity. It is an extension of the human - a delegate operating within the human's existing credential space. The DBS Bank pilot in Singapore used AI-ready credentials: tokenised card details tied to a specific AI agent, with intent-driven transaction controls managed by the issuer. The agent transacts using the human's identity, constrained by rules the human (or the issuer) has defined.

This is the conservative position, and it has a powerful advantage: it fits within existing regulatory frameworks. The human remains the accountable party. The agent is a tool, not a participant. But it has a powerful disadvantage too: it cannot scale to autonomous agent-to-agent commerce. If every agent transaction must trace back to a human credential, the system cannot support agents that transact with other agents without human involvement. Visa's philosophy works for Level 2-3 of Stripe's framework. It breaks at Level 4-5.

Philosophy Three: The Agent as Protocol (Stripe)

Stripe's x402 protocol sidesteps the identity question entirely. In x402, the agent is neither an entity nor an extension. It is a protocol participant. A server returns HTTP 402 ("Payment Required"), the agent responds with a payment token, the server verifies and fulfils. The agent's identity is irrelevant to the transaction. What matters is the token - the pre-authorised credential that carries spending authority.

This is the most technically elegant position and the most philosophically evasive. By making identity irrelevant, Stripe avoids the hardest design questions. But those questions do not disappear because a protocol ignores them. When an x402 transaction goes wrong - when the agent buys the wrong thing, or buys at the wrong price, or buys from a fraudulent endpoint - the identity of the agent matters enormously. The protocol has no mechanism for accountability, because it was designed without one.

The Protocol Proliferation Problem

Q1 2026 did not produce a standard for agentic payments. It produced at least six competing ones. This is not healthy competition. It is fragmentation at the foundation layer - the kind that creates decades of interoperability problems.

Six protocols in a single quarter. None of them interoperable. None of them addressing the same trust model. This is the early internet before TCP/IP - multiple incompatible network protocols competing for dominance, each backed by a different corporate interest, each solving a slightly different version of the problem.

The difference is that the early internet's protocol wars were about data transmission. These protocol wars are about money. The stakes of fragmentation are not inconvenience - they are financial loss, fraud, and the erosion of consumer trust before it has been established. An agent built for x402 cannot transact on Mastercard's network. An agent built for Visa's TAP cannot use Stripe's Shared Payment Tokens. The merchant who wants to accept agentic payments must integrate with multiple incompatible systems, each with different identity models, different authentication requirements, and different liability frameworks.

From an AXD perspective, protocol proliferation is not a technical problem. It is a trust architecture problem. Every new protocol creates a new trust boundary. Every trust boundary requires a new set of design decisions about delegation, consent, visibility, and recovery. The more protocols, the more trust boundaries. The more trust boundaries, the more opportunities for trust to fail. The industry is building complexity at the exact layer where it needs simplicity.

The Alipay Anomaly

Between February 5th and 11th, 2026, Alipay's AI Pay processed over 120 million transactions. In one week. While Mastercard was celebrating a single authenticated transaction in New Zealand and Visa was piloting with one bank in Singapore, Alipay was processing transactions at a rate that dwarfs the entire Western agentic payments ecosystem combined.

This number demands attention, but it also demands context. Alipay operates within a fundamentally different trust environment. Chinese consumers have spent a decade building comfort with mobile-first, platform-mediated payments. The regulatory framework is different. The cultural relationship with AI is different. The competitive dynamics are different. Alipay's scale does not prove that Western markets are ready for the same volume. It proves that scale is technically possible - and that the bottleneck in Western markets is not technology but trust.

The Alipay anomaly reveals something the Western payments industry has not yet internalised: the constraint on agentic commerce is not infrastructure. The infrastructure works. Alipay proved it. The constraint is the trust relationship between humans and agents - the relationship that must be designed, not assumed. Western consumers have not spent a decade building comfort with agent-mediated transactions. They are being asked to trust systems they have never used, built by companies whose AI capabilities they have no way to evaluate, operating under regulatory frameworks that do not yet exist.

Alipay's 120 million transactions are not a roadmap. They are a warning. Scale without trust design is possible - in an environment where trust has been built through other means over many years. In environments where that trust does not exist, scale without trust design is a recipe for the kind of consumer backlash that sets entire industries back by years.

The Consumer Trust Data

The industry's optimism about agentic commerce rests heavily on survey data. BCG reports that 81% of consumers are "inclined to use" agentic commerce, with $1.3 trillion in spending potentially affected. McKinsey projects up to $1 trillion in US retail and $5 trillion globally by 2030. These numbers are cited in every investor deck and every corporate strategy document. They are also misleading.

The more granular data tells a different story. Checkout.com's research found that UK consumers are comfortable with AI agents spending an average of just £204.53 on their behalf. US consumers: $233. These are not the numbers of an industry about to process trillions. These are the numbers of consumers who are willing to let an agent buy groceries but not a car. The same research found that 42% of consumers fear losing control over purchases, and 28% cite lack of transparency as their biggest concern.

Read together, the data says something precise: consumers are interested in the concept of agentic commerce but do not yet trust the implementations. The gap between "inclined to use" and "willing to spend significant amounts" is the trust gap. It is not a marketing problem. It is not an education problem. It is a design problem. The systems being built do not yet contain the trust mechanisms that would make consumers comfortable delegating meaningful financial authority to agents.

The Venture Capital Signal

Venture capital invested $4.2 billion in agentic AI companies in Q1 2026 alone. Circuit & Chisel raised $19.2 million specifically for the ATXP agent transaction protocol. These numbers signal conviction. They also signal a particular kind of conviction - conviction in infrastructure, not in trust.

Examine what the money is funding. Payment protocols. Agent orchestration platforms. Transaction processing infrastructure. API layers. Token management systems. The capital is flowing toward the plumbing - the pipes through which agent transactions will flow. It is not flowing toward the design of the trust relationships that will determine whether consumers use those pipes.

This is not surprising. Venture capital funds what it can measure. Transaction volume is measurable. Processing fees are measurable. API calls are measurable. Trust is not measurable in the same way - not in the quarterly metrics that venture capital requires. The result is a funding landscape that systematically underinvests in the thing that will determine whether agentic commerce succeeds or fails.

The Collison brothers, in Stripe's annual letter, offered the most honest assessment of the quarter: "Like much in AI, agentic commerce suffers from having been overhyped too early." They placed the industry at "hovering on the edge of levels 1 and 2" of their own five-level framework. This is the founders of the company that built x402 saying that the industry is still at the beginning. The $4.2 billion in venture capital is a bet on the future. The present is considerably more modest than the investment suggests.

What Nobody Built This Quarter

The most important story of Q1 2026 is not what was built. It is what was not built. The following capabilities are absent from every agentic payment system announced this quarter. Not partially implemented. Not in beta. Absent.

Agent reputation systems. No payment network has a mechanism for tracking agent behaviour over time. An agent that has completed ten thousand successful transactions is treated identically to an agent completing its first. There is no reputation, no track record, no accumulated trust. Every transaction starts from zero.

Cross-network agent identity. An agent authenticated on Mastercard's network has no identity on Visa's. An agent using x402 has no identity on either. There is no portable agent identity - no way for an agent's reputation or credentials to follow it across payment networks. This means every network boundary is a trust boundary, and every trust boundary resets the relationship to zero.

Graduated delegation controls. Every system announced this quarter offers binary delegation: the agent can transact or it cannot. There are no graduated controls - no mechanism for saying "this agent can spend up to £50 on groceries but must ask permission for anything over £200 or anything from a new merchant." The Delegation Design Framework describes seven dimensions of delegation. Current implementations address one.

Recovery architecture. When an agent transaction goes wrong - wrong item, wrong price, wrong merchant, fraudulent endpoint - there is no designed recovery pathway. Traditional chargebacks assume a human made the purchase. Agent transactions break this assumption. Who is liable when an agent buys something the human did not want? The agent's developer? The payment network? The merchant? The human who delegated? No system announced this quarter answers this question.

Transparency infrastructure. No system provides real-time visibility into what an agent is doing during a transaction. The human delegates, the agent acts, and the human sees the result. The process between delegation and result is opaque. This is the hesitation layer problem: the human friction that traditionally absorbed errors has been removed, and nothing has been designed to replace it.

These are not edge cases. They are the core requirements of any system where humans delegate financial authority to autonomous agents. The fact that none of them were addressed in Q1 2026 - by any company, in any market - is the clearest signal that the industry is building payment infrastructure without trust infrastructure. The rails are being laid. The safety systems are not.

The KYA Question

FIME, the global consulting firm specialising in payment security, introduced a concept this quarter that deserves to become canonical: Know Your Agent (KYA). The parallel to Know Your Customer (KYC) is deliberate and revealing.

KYC exists because the financial system learned, through decades of fraud and money laundering, that you cannot process transactions safely without knowing who is transacting. The entire anti-money-laundering infrastructure - billions of dollars of compliance technology, thousands of pages of regulation, entire departments within every financial institution - exists because identity matters in financial transactions.

KYA asks the equivalent question for agents: if you are going to let an autonomous system transact in your payment network, do you not need to know what that system is? Not just its token. Not just its credentials. Its capabilities. Its limitations. Its failure modes. Its developer. Its training data. Its decision-making logic. Its history of transactions. Its error rate. The things you would want to know about any entity you are trusting with money.

FIME offered a useful metaphor: the agent as a "corporate intern with a prepaid card." You would not give an intern unlimited spending authority without knowing who they are, what they are capable of, and what constraints they operate under. You would not let them transact without oversight. You would not hold them personally liable for mistakes - you would hold the organisation that employed them. The metaphor is imperfect, but it captures something essential: agents need governance, and governance requires knowledge.

No payment system announced in Q1 2026 implements KYA. Mastercard's Agentic Tokens identify that an agent is transacting, but they do not capture what the agent is, what it can do, or how it makes decisions. Visa's Intelligent Commerce constrains what the agent can spend, but it does not assess the agent's competence or reliability. Stripe's x402 does not even ask the question. The KYA gap is the trust gap expressed in regulatory language. It is the gap that will need to be closed before agentic payments can scale beyond pilot programmes and controlled environments.

Five Predictions for Q2 2026

Quarterly briefings earn their value through prediction. The following are not hopes or recommendations. They are assessments of what the patterns of Q1 2026 suggest will happen next, viewed through the lens of Agentic Experience Design.

One: Protocol consolidation will begin. Six competing protocols is unsustainable. By the end of Q2, at least one will be abandoned or absorbed. The most likely consolidation is around x402 (which has Stripe's distribution) and the card network protocols (which have Mastercard and Visa's distribution). The startup protocols - ATXP, ACP - will struggle to compete with incumbents who control the rails.

Two: The first major agentic payment failure will occur. With more pilots going live and volumes increasing, the probability of a significant failure - an agent making a large unauthorised purchase, a security breach in an agent credential system, a consumer losing money through an agent error - approaches certainty. This failure will be the catalyst for the trust conversation the industry has been avoiding.

Three: Regulators will issue their first formal guidance. The EU, the UK's FCA, or the US CFPB will publish the first regulatory framework specifically addressing agent-initiated financial transactions. It will be cautious, principles-based, and insufficient - but it will establish that agents operating in financial systems are within regulatory scope.

Four: Mastercard will launch Agent Suite. The Q2 launch of Mastercard's broader Agent Suite will be the first attempt to move beyond single-transaction pilots to a comprehensive agent commerce platform. Its success or failure will signal whether agent-native identity is viable at scale or whether Visa's delegated model is more practical.

Five: The trust design conversation will begin in earnest. The combination of increasing volume, protocol competition, and the inevitable first failure will force the industry to confront the question it avoided in Q1: how do you design trust into agentic payment systems? This is the conversation AXD was built for. The discipline exists because this moment was coming. It has arrived.

State of Agentic: Implications for Practitioners

Q1 2026 confirms what the AXD Institute has argued since its founding: the design of human agent interaction in financial contexts is the most consequential design challenge of the next decade. The payment infrastructure is being built. The trust infrastructure is not. The gap between them is the space where Agentic Experience Design operates.

For practitioners, the quarter produces five actionable implications. First, design for protocol plurality. Do not assume a single protocol will win. Design trust architectures that work across x402, Agentic Tokens, TAP, and whatever emerges next. The trust layer must be protocol-agnostic because the protocol layer is not going to consolidate quickly.

Second, design for the £204.53 ceiling. The consumer trust data is your design brief. The average amount consumers will trust an agent to spend is not a marketing problem - it is a trust design problem. Every mechanism you build that increases transparency, provides graduated control, or enables easy recovery raises that ceiling. Measure your success by whether the number goes up.

Third, design the recovery pathway before the happy path. Every payment system announced this quarter designed the successful transaction first. AXD practitioners must design the failed transaction first. What happens when the agent buys the wrong thing? What happens when the merchant is fraudulent? What happens when the agent exceeds its authority? The recovery pathway is the trust pathway. It is where trust is built or destroyed.

Fourth, design for KYA. The Know Your Agent framework will become regulatory requirement within eighteen months. Practitioners who build KYA-ready systems now - systems that can identify, assess, and monitor the agents operating within them - will be ahead of the regulatory curve rather than scrambling to comply.

Fifth, design the hesitation layer replacement. The hesitation layer - the human friction that traditionally absorbed errors in financial transactions - is being removed by every system announced this quarter. Something must replace it. Not friction for friction's sake, but designed checkpoints that catch errors before they become losses. The replacement for human hesitation is not speed. It is designed trust.

Q1 2026 was the quarter where agentic payments moved from concept to competing implementations. Q2 2026 will be the quarter where the consequences of building payments without trust become visible. The AXD Institute will continue to observe, analyse, and provide the design perspective that the industry needs but has not yet built internally.

The payment rails are being laid at extraordinary speed. The question is not whether agents will transact. They already do. The question is whether humans will trust them to. That question is not answered by infrastructure. It is answered by design.