Payment rails for machine-initiated transactions in Europe

The term agentic commerce refers to commercial transactions in which software agents, operating under delegated authority from human or institutional principals, initiate, negotiate, and settle payments without per-transaction human review. The agent may be an AI model executing a procurement workflow, an IoT sensor triggering a micro-payment upon confirmed delivery of a service, or an orchestration layer that coordinates multiple sub-agents each settling fractional obligations in sequence. What distinguishes agentic commerce from earlier forms of automated payment, such as direct debit mandates or scheduled wire transfers, is the combination of high transaction frequency, conditional execution logic, and the absence of a synchronous human approval step at the moment of settlement.

This combination creates infrastructure requirements that differ materially from those of conventional retail or corporate payment. The settlement rail must respond within latency bounds compatible with machine interaction rather than human interaction. It must expose confirmation signals in formats that a downstream process can parse without human intermediation. It must support, or at minimum not impede, conditional execution, where payment release is tied to an externally verifiable event such as a delivery confirmation, a sensor reading, or a service-level attestation. It must provide finality that is sufficiently rapid and legally certain that the agent can proceed to the next step in a workflow without waiting for a manual reconciliation cycle. And it must operate at per-transaction costs that do not render micro-transaction patterns economically unviable.

Europe presents a particularly instructive case for evaluating candidate rails because the region hosts three architecturally distinct settlement infrastructure candidates at different stages of maturity. SEPA Instant Credit Transfer, operated under the governance of the European Payments Council and reaching the majority of eurozone account-holding institutions, represents the incumbent pan-European instant rail. Wero, the multi-bank overlay scheme launched by EPI Company SAS and progressively rolling out across France, Germany, and Belgium among its initial markets, represents a second-generation consumer-facing instant payment product built atop national instant rails; this list of launch markets is illustrative of its current operational geography rather than exhaustive of its eventual scope. The digital euro, under development by the European Central Bank and subject to a separate legislative instrument, represents a prospective central bank digital currency that proponents position as capable of programmable and offline settlement.

This paper conducts a structured comparative evaluation of these three rails against the five technical criteria most directly relevant to agentic commerce: settlement speed, programmability, finality, machine-readable confirmation, and cost at scale. The evaluation is not a ranking intended to produce a single winning recommendation. The three rails serve partially overlapping but not identical markets and design objectives. The purpose of the comparison is to identify, for each rail, the specific gaps that separate its current or near-term capability from what autonomous agent workflows require, and to identify the structural changes, whether technical, regulatory, or standards-based, that would be necessary to close those gaps.

The paper proceeds as follows. Section 2 establishes why settlement infrastructure is a binding constraint for agentic commerce rather than an incidental engineering detail. Section 3 positions the contribution against prior literature on fast payments, programmable money, and autonomous commerce architectures. Section 4 describes the evaluation framework, data sources, and decision rules. Section 5 presents the comparative findings rail by rail. Section 6 interprets the results and their implications for infrastructure design. Section 7 concludes. Supplementary sections address limitations, future work, a traced case study, and data provenance.

Two terminological clarifications are necessary at the outset. First, the paper uses the term finality to denote the condition under which a payment cannot be reversed by the sending institution unilaterally, consistent with the usage in settlement law and European Central Bank documentation. Second, the paper distinguishes programmability, the capacity to attach conditional execution logic to the payment instrument itself, from API accessibility, the capacity to initiate or receive notifications about payments through a machine interface. Both matter for agentic commerce, but they are distinct capabilities that individual rails may possess in different combinations.

The urgency of evaluating payment rails for agentic commerce arises from the convergence of three distinct pressures: regulatory timelines that will materially alter the European payment landscape within a two-to-three-year horizon, documented operational friction in current cross-border euro-denominated commerce, and the specific technical constraints that autonomous transaction initiation imposes on settlement infrastructure.

On the regulatory front, the Instant Payments Regulation, which entered into force in 2024, mandates that eurozone payment service providers offer instant credit transfer as a standard product at pricing no higher than standard credit transfer by a defined implementation deadline. This obligation transforms SEPA Instant from an optional, premium service into a universal baseline. In parallel, the Payment Services Directive 3 legislative package advances the open banking framework established under its predecessor, strengthening API access obligations and introducing enhanced strong customer authentication provisions. The digital euro legislative proposal, introduced by the European Commission, establishes the legal tender status of a potential central bank digital currency and sets out distribution obligations for payment service providers. Each of these instruments reshapes the competitive and technical landscape in ways that directly affect which rails are viable candidates for agentic settlement by the mid-decade period.

Current cross-border euro-denominated commerce, even within the single euro payments area, involves friction that compounds when the transacting party is a machine rather than a human. Correspondent banking arrangements introduce settlement delays measured in hours or days on transactions not covered by real-time interbank connections. Reconciliation relies on narrative reference fields that are not standardised to machine-readable schemas across institutions. Confirmation of finality is communicated through mechanisms, such as statement delivery and account credit notification, designed for human review rather than programmatic ingestion. These frictions are tolerable when a treasury team performs end-of-day reconciliation but become structural blockers when an agent must confirm settlement before releasing a resource or initiating the next transaction in a conditional workflow. This last observation is an inference drawn from the structural requirements of autonomous workflow design; the agentic payment protocol literature [9, 10] demonstrates the same constraint operating within cryptocurrency-native settlement contexts, and the underlying mechanism, an agent cannot proceed past a decision node without a machine-parseable confirmation signal, applies regardless of the specific monetary instrument.

The technical constraints of machine-initiated transactions are qualitatively different from those of human-initiated ones in three respects. First, latency tolerance is narrower: an agent executing a procurement workflow may need settlement confirmation within the duration of an API timeout window, typically seconds, rather than the minutes or hours acceptable in conventional treasury operations. Second, the absence of a human in the confirmation loop means that any ambiguity in the confirmation signal, such as a status code that requires contextual interpretation, will produce either a process halt or an erroneous downstream action. Third, the volume profile of agentic transactions differs from conventional retail: a single orchestrated workflow may produce hundreds or thousands of micro-transactions in rapid succession, creating throughput and cost-per-transaction pressures that do not arise in human-paced commerce [10].

The combination of these regulatory and technical pressures means that the question of which settlement rail is most suitable for agentic commerce is not a speculative future-state question. It is a design decision that infrastructure operators, payment service providers, and AI system architects must engage with now, because the architectural choices embedded in current rail specifications will govern what is technically feasible for agentic applications for at least the medium term.

The literature relevant to this paper spans four distinct bodies of work: research on fast payment system design and governance, research on central bank digital currencies, research on distributed ledger technologies in post-trade and settlement contexts, and a nascent literature on payment protocols for autonomous and AI-driven commerce, represented by preprints published in 2025 and 2026.

Fast payment systems and European instant payment governance. Iglesias-Rodriguez [7] examines the future of instant payments in the European Union through the lens of bank ownership structures, market contracting costs, and regulatory objectives. The paper analyses how the governance architecture of instant payment schemes interacts with the competitive dynamics between pan-European solutions and nationally anchored alternatives. That work establishes the institutional and regulatory context for SEPA Instant and the arena model of European instant payment competition but does not address the technical requirements that autonomous agent workflows impose on settlement infrastructure. The present paper extends this context by applying the institutional analysis to the specific performance dimensions that agentic commerce demands.

Central bank digital currency design. Opare and Kim [5] provide a compendium of CBDC design practices across multinational financial infrastructures, surveying architectural choices including two-tier distribution, programmable payment layers, and interoperability mechanisms. Allen, Gu, and Jagtiani [2] examine the Chinese digital currency experience as a case of CBDC deployment within a complex fintech ecosystem, surfacing the interaction between central monitoring capabilities and the displacement of private payment instruments. Cannataci and colleagues [8, 11] engage directly with the digital euro design, revisiting frequently asked questions with particular attention to the tension between the ECB's transaction monitoring architecture and privacy preservation under European data protection law. These works collectively establish the design space for the digital euro as a candidate rail but do not evaluate it against the operational criteria of agentic commerce. The present paper applies the architectural findings of this literature to the five-criteria evaluation framework.

Distributed ledger technologies in settlement contexts. Ruttenberg and Pinna [6] analyse the application of distributed ledger technologies to securities post-trade processes, distinguishing evolutionary from revolutionary change scenarios and identifying the conditions under which DLT offers a genuine settlement efficiency gain over incumbent central securities depository architectures. Rauchs and Hileman [3] provide foundational benchmarking of cryptocurrency networks, including transaction throughput, latency, and cost characteristics. These works inform the present paper's treatment of programmability and finality as distinct settlement properties and provide a baseline for comparing DLT-native payment approaches with European public rail candidates.

Autonomous commerce payment protocols. The most directly relevant prior work for this paper's contribution comes from two recent preprints. Li and colleagues [9] introduce the A402 protocol, a mechanism for binding cryptocurrency payments to service execution in agentic commerce contexts. The A402 design addresses the specific problem of atomic settlement, ensuring that payment release and service delivery are coupled in a single indivisible operation, using on-chain conditional logic. That work demonstrates the technical feasibility of programmable, machine-readable, atomic settlement, but does so entirely within a cryptocurrency and blockchain context, without evaluating European public rail candidates. See and Tan [10] examine compliance-aware agentic payments on stablecoin rails, proposing a framework in which compliance obligations, including AML screening and transaction limits, are embedded as on-chain guardrails executed at settlement time rather than as separate post-transaction workflows. This design preserves settlement speed while satisfying regulatory obligations, a combination directly relevant to the European agentic commerce context.

The gap that the present paper addresses is the intersection of these two bodies of work. The agentic payment protocol literature has established what agentic commerce requires technically and demonstrated solutions on DLT-native rails. The European payment infrastructure literature has established the institutional and regulatory architecture of SEPA Instant, Wero, and the digital euro. No prior work has systematically evaluated the three European rails against the technical criteria derived from the agentic payment literature. The present paper performs that evaluation, identifying specific gaps in each rail relative to each criterion and characterising the structural changes that would be needed to achieve agentic readiness.

The broader fintech literature [1, 4] provides context on the dynamics of financial innovation and the conditions under which new payment instruments achieve adoption. Lerner and Tufano [4] specifically examine the counterfactual consequences of financial innovation, a methodological orientation relevant to assessing what the European payment ecosystem foregoes by not providing programmable settlement primitives on public rails.

The evaluation proceeds through a structured framework that applies five technical criteria to each of the three candidate rails. The criteria were derived from the requirements established in the agentic payment protocol literature [9, 10] and from the operational constraints of autonomous transaction workflows described in the motivation section. Each criterion is operationalised with a defined measurement dimension and a threshold that distinguishes adequate from inadequate performance for agentic commerce.

Criterion 1: Settlement speed. Measured as the elapsed time from payment initiation to the point at which the receiving institution has posted the credit to the beneficiary account. For agentic commerce, the relevant threshold is compatibility with API timeout conventions in networked service environments. The analysis treats ten seconds as the upper bound for a transaction that must complete before the initiating agent can proceed. Latency reported in technical specifications or regulatory instruments is taken at face value; where no specification exists, the analysis notes the absence and characterises the gap.

Criterion 2: Programmability. Defined as the capacity of the payment instrument to encode conditional execution logic, such that payment release is contingent on an externally verifiable event or a time-bound condition, without requiring human intervention at the point of execution. This is distinct from API accessibility, which is necessary but not sufficient. Programmability requires either a smart-contract execution layer, a conditional payment primitive in the scheme rules, or an escrow mechanism with machine-readable release conditions. The evaluation assesses whether each rail provides any such primitive in its published design documentation.

Criterion 3: Finality guarantees. Assessed as the legal and operational certainty that a credited payment cannot be reversed unilaterally by the sending institution after the confirmation timestamp. European settlement law, as implemented in the Settlement Finality Directive and its domestic transpositions, provides the legal baseline. The operational question is whether the finality signal is delivered to the receiving party in a form and on a timeline compatible with agentic workflow continuation.

Criterion 4: Machine-readable confirmation. Assessed as the availability of a structured, schema-validated confirmation event delivered to the payee's system automatically upon settlement, without requiring the payee to poll a statement or parse a narrative message. The reference standard is a webhook or push notification carrying a structured payload, such as an ISO 20022 camt.054 message or an equivalent JSON schema, delivered within the settlement latency window.

Criterion 5: Cost at scale. Assessed as the per-transaction fee structure at high volume, specifically at transaction volumes consistent with agentic micro-transaction patterns, which may reach thousands of transactions per hour per agent. The analysis notes where published fee schedules exist and where they do not, and characterises the structural cost model of each rail in the absence of published schedules.

Data sources. The evaluation draws on four categories of source material. First, published technical specifications and scheme rulebooks from the European Payments Council for SEPA Instant. Second, publicly available documentation and press materials from EPI Company SAS regarding the Wero scheme. Third, the European Central Bank's published reports and legislative proposal documentation on the digital euro, supplemented by the analysis in Cannataci and colleagues [8, 11]. Fourth, the academic and preprint literature on agentic payment protocols [9, 10] and European instant payment governance [7], which provides the analytical framing for interpreting infrastructure capabilities.

Decision rules. For each rail-criterion pair, the assessment assigns one of three characterisations: Adequate, meaning the rail meets the agentic threshold based on published evidence; Partial, meaning the rail meets some component of the criterion or approaches the threshold with documented gaps; or Absent, meaning no published evidence establishes that the rail meets the criterion. Where a single criterion contains distinct sub-dimensions that receive different characterisations, the binding characterisation is the weaker of the two, and the sub-dimension distinction is noted in prose. The characterisation Absent does not assert that the capability does not exist; it asserts that no published evidence reviewed in this analysis establishes it. This conservative decision rule is appropriate given the evidentiary gaps identified in the synthesis and the limitations discussed in the dedicated section below.

The analysis does not conduct primary empirical testing of any rail. It is a structured review and synthesis of published specifications and literature, supplemented by inference from architectural descriptions where direct evidence is absent. The limitations of this approach are enumerated in the Limitations section.

SEPA Instant Credit Transfer

Speed. SEPA Instant specifies a maximum end-to-end processing time of ten seconds from the moment the payment order is received by the sending payment service provider to the point of credit at the receiving institution. This specification places SEPA Instant within the threshold identified as necessary for agentic workflow compatibility. The mandatory adoption timeline established under Regulation (EU) 2024/886 applies in three phases: eurozone payment service providers must offer instant credit transfer for sending by October 2025 and for receiving by January 2026; non-eurozone EU member payment service providers must meet the equivalent obligations by July 2027 for sending and January 2028 for receiving. This phased structure materially expands SEPA Instant's reachability across the EU by the end of the decade. The ten-second maximum is a scheme-level commitment; actual network performance may vary by institution and peak load, and no published stress-test data specific to high-frequency automated transaction loads has been identified in the course of this review.

Programmability. The SEPA Instant Credit Transfer scheme, as documented in the EPC rulebook, does not include a conditional payment primitive, a smart-contract execution layer, or an escrow mechanism with machine-readable release conditions. The scheme transmits a credit instruction unconditionally upon initiation. Any conditional logic must be implemented by the initiating application prior to submitting the payment order, which places the conditional execution burden on the agent's orchestration layer rather than on the settlement rail. This architecture means that the payment and the service delivery event are decoupled at the rail level: the rail does not guarantee atomicity between payment and service execution [9]. Characterisation: Absent.

Finality. SEPA Instant provides irrevocability at the scheme level upon confirmation of credit, consistent with the Settlement Finality Directive framework applicable to designated systems. Under the EPC SCT Inst rulebook, a positive confirmation message (pacs.002) is delivered to the originating payment service provider upon successful credit to the beneficiary account, within the ten-second processing window. This confirmation message is the mechanism through which the originating PSP learns that the credit has been applied; it is not a separately designated finality signal but rather the scheme's standard positive acknowledgement, which carries irrevocability as a consequence of the scheme rules. Whether this acknowledgement is propagated onward to the payee's system in a machine-readable form within the same window depends on the implementation choices of individual payment service providers, and no universal standard for payee-side credit notification has been identified in the scheme documentation reviewed. The binding characterisation for this criterion, applying the framework's rule that the weaker sub-dimension governs, is Partial, reflecting the absence of a mandated payee-side notification standard. The scheme-level irrevocability within the ten-second window is noted as the adequate sub-dimension.

Machine-readable confirmation. The ISO 20022 message set, which SEPA Instant uses for interbank messaging, includes the camt.054 credit notification message and related report structures capable of carrying structured settlement confirmation data. However, the delivery of these messages to end-client systems, as opposed to between financial institutions, is governed by individual bank API implementations rather than a mandatory scheme-wide standard. The open banking API frameworks mandated under PSD2 and its successor provide a baseline for account information access but do not mandate a push-notification or webhook standard for real-time settlement confirmation at the payee level. Characterisation: Partial.

Cost at scale. Published EPC scheme documentation does not specify per-transaction fees; pricing is determined bilaterally between payment service providers and their clients. Under Regulation (EU) 2024/886, the fee for instant credit transfer must not exceed the fee for standard credit transfer, establishing a ceiling on pricing without specifying a floor or a volume-tiered structure. For micro-transaction patterns at agentic scale, the relevant cost driver is whether pricing is flat per transaction or volume-tiered, a distinction with large consequences at thousands of transactions per hour. No published evidence establishes the fee structure at micro-transaction volumes for SEPA Instant. Characterisation: Partial (regulatory ceiling established; micro-transaction volume economics undocumented).

Wero

Speed. Wero is built atop national instant payment rails, including SEPA Instant in the markets where it operates, and inherits the latency characteristics of those underlying rails. In consumer-facing deployments in France and Germany, the scheme delivers payment confirmation within seconds. The end-to-end latency for machine-initiated transactions is not separately documented in publicly available Wero scheme materials. Characterisation: Adequate in consumer-facing operation; undocumented for non-human initiators.

Programmability. No publicly available Wero scheme documentation describes a conditional payment primitive, a programmable release mechanism, or a smart-contract layer. Wero's publicly described functionality is oriented toward peer-to-peer and point-of-sale consumer payments. The EPI Company's roadmap includes merchant payment capabilities, but no documentation of programmable payment primitives accessible to automated systems has been identified. Characterisation: Absent.

Finality. Wero inherits the finality characteristics of the underlying SEPA Instant rail where that rail is used. The overlay's own finality treatment in markets where it does not use SEPA Instant as the settlement layer is not separately documented in reviewed materials. Characterisation: Adequate where SEPA Instant underlies the settlement; Partial or undocumented elsewhere. Applying the framework's binding-characterisation rule: Partial.

Machine-readable confirmation. Wero's documented interface is a consumer-facing mobile application. No published API specification for programmatic access by non-human initiators, and no webhook or push-notification schema for machine-readable settlement confirmation, has been identified in reviewed materials. This represents the most significant gap for agentic commerce: the absence of a documented machine-interface layer means an agent cannot initiate a Wero payment or receive structured confirmation without a bespoke integration that is not supported by published standards. Characterisation: Absent.

Cost at scale. Wero's consumer-facing fee model is zero or low-cost for peer-to-peer transactions in its launch markets. Merchant pricing has been announced in general terms but not published as a fee schedule accessible for micro-transaction volume analysis. The cost model at agentic transaction volumes is therefore uncharacterised. Characterisation: Absent (merchant and high-volume micro-transaction pricing undocumented).

Digital Euro

Speed. The digital euro's design documentation, as reviewed in the ECB's published reports and analysed in Cannataci and colleagues [8, 11], targets retail payment speeds consistent with existing instant payment rails for online transactions. The offline payment mode, intended for transactions without network connectivity, introduces a distinct latency model in which value transfer occurs locally between devices with later reconciliation at the intermediary PSP level, rather than in real time against any central ledger; final settlement is deferred to the point at which the intermediary PSP synchronises with the central infrastructure. For agentic commerce, which presupposes network connectivity in the majority of use cases, the online settlement latency is the relevant parameter. No specific maximum latency figure has been published for the online mode at the time of this review. Characterisation: Partial (online settlement targets instant-compatible latency; no published maximum confirmed).

Programmability. The digital euro legislative framework and ECB design reports acknowledge the concept of conditional or programmable payments as a potential feature of the instrument, particularly for machine-to-machine and Internet of Things payment scenarios. The architectural approach to programmability, whether through a smart-contract execution layer, a conditional payment API exposed to third-party developers, or a restricted set of ECB-defined payment conditions, has not been finalised in published specifications. The privacy tension identified by Cannataci and colleagues [8, 11] is directly relevant here: programmable payments require transaction-linked attestation and execution audit trails that are architecturally in tension with the privacy preservation requirements under GDPR and the political commitments made during the legislative process. Characterisation: Partial (programmability acknowledged in design; implementation unspecified).

Finality. A digital euro transaction settled on the ECB's infrastructure carries central bank finality, the strongest form of settlement finality available in the European monetary system, structurally equivalent to the finality of a TARGET2 settlement. The finality of the instrument is not in question as a structural matter. The operational question of how quickly and in what format the finality signal is delivered to a transacting agent's system remains unspecified in published materials. Characterisation: Adequate in principle; undocumented operationally. Applying the framework's binding-characterisation rule: Partial.

Machine-readable confirmation. No published digital euro API specification or developer interface documentation has been identified in the course of this review. The distribution model envisages intermediary payment service providers as the client-facing interface, which means the machine-readable confirmation format will be determined by intermediary implementation choices unless a mandatory technical standard is specified in the legislative or regulatory instrument. Characterisation: Absent (no published specification).

Cost at scale. The digital euro legislative proposal establishes that the ECB will not charge end-users directly, and that distribution costs will be borne by intermediary payment service providers subject to compensation arrangements to be determined. The per-transaction cost to a high-volume agentic commerce operator would therefore depend on the pricing model adopted by the intermediary PSP, which is not determinable from current published materials. No marginal cost structure for micro-transaction volumes has been published. Characterisation: Absent (cost model undetermined at distribution level).

The comparative assessment produces a clear structural finding: across the fifteen rail-criterion pairs produced by the three-rail, five-criterion framework, no rail achieves an Adequate characterisation on all five criteria. To make the aggregate pattern explicit: SEPA Instant receives Adequate on speed and Partial on finality, machine-readable confirmation, and cost, with Absent on programmability; Wero receives Adequate on speed in consumer-facing operation and Partial on finality where SEPA Instant underlies the settlement, with Absent on programmability, machine-readable confirmation, and cost; the digital euro receives Partial on speed, programmability, and finality, with Absent on machine-readable confirmation and cost. The criteria on which all three rails score weakest, programmability and machine-readable confirmation, are precisely the criteria most directly implicated in what distinguishes agentic commerce from conventional automated payment.

SEPA Instant meets two of the five criteria at the Adequate level, compared with zero for the other two rails in their fully evaluated state. This differential is attributable to its settlement architecture: the ten-second maximum latency is a scheme-level commitment derived from a published rulebook, and the scheme-level irrevocability upon the pacs.002 confirmation is grounded in the Settlement Finality Directive framework. The gap between SEPA Instant's current state and agentic readiness is concentrated in two specific areas: the absence of a mandatory payee-side machine-readable confirmation standard, and the absence of any conditional payment primitive. Both gaps are, in principle, addressable through incremental standardisation without requiring architectural redesign of the scheme. The EPC could mandate a push-notification standard for payee-side credit confirmation, defining a required JSON or ISO 20022 schema and delivery timeline. The EPC or a regulatory instrument could introduce a conditional payment annex to the scheme rulebook, defining permissible condition types and the execution environment for evaluating them. These are non-trivial standardisation exercises but are structurally feasible within the existing SEPA Instant architecture.

The atomicity gap is the central structural problem. The A402 protocol literature [9] demonstrates that the requirement most specific to agentic commerce is the binding of payment to service execution in a single atomic operation, a property that speed and finality alone do not provide. A human purchaser who pays and receives nothing can dispute the transaction, escalate to a human operator, or initiate a chargeback. An agent operating in a multi-step conditional workflow cannot perform any of these human-mediated recovery actions without halting the workflow and surfacing the failure to a human supervisor, at which point the latency advantage of autonomous settlement is extinguished. None of the three rails provide an atomic settlement primitive at the rail level. SEPA Instant settles unconditionally upon instruction; Wero provides a consumer overlay without a programmable layer; and the digital euro's programmability remains architecturally acknowledged but operationally unspecified. This atomicity gap reflects the fact that European public payment rails were designed for human-authorised transactions in which the decoupling of payment from delivery is managed through legal remedies rather than technical primitives, and no single rail's implementation has resolved this by design.

Wero's gap is more fundamental because it originates at the interface layer rather than the settlement layer. The absence of a documented machine-accessible API represents the absence of a specification altogether, rather than an incomplete technical specification. Wero's design, as publicly documented, is oriented toward consumer-to-consumer and consumer-to-merchant interactions mediated by a mobile application. Extending Wero to agentic commerce would require EPI Company to publish and maintain a developer API, define authentication and authorisation standards for non-human initiators, specify machine-readable confirmation schemas, and address the question of how agent identity is verified and delegated authority is established. See and Tan [10] identify the authentication and identity verification problem as one of the most structurally challenging aspects of agentic payment design, because existing strong customer authentication frameworks are built around human biometric or knowledge-based verification that agents cannot perform. Wero's current architecture provides no published resolution to this problem.

The digital euro's position is analytically distinctive. It carries the strongest structural potential for programmable and conditional settlement among the three rails, because its design process explicitly acknowledges machine-to-machine payment as a target use case, and because a central bank digital currency's architecture can embed conditional execution logic at the ledger level rather than as an overlay on a credit transfer scheme. However, the gap between architectural potential and operational specification is currently the widest of the three rails. The privacy-programmability tension identified in Cannataci and colleagues [8, 11] is not merely a policy preference; it reflects a genuine architectural constraint. Programmable conditional payments require the execution environment to evaluate conditions against transaction data at settlement time. If that evaluation occurs on a central ECB ledger, it creates a transaction-level monitoring capability that is structurally inconsistent with the privacy commitments in the legislative proposal. If it is delegated to intermediary PSPs, the programmability guarantee is only as strong as the intermediary's implementation, which reintroduces fragmentation. No published architectural resolution to this tension has been identified.

The cross-border dimension amplifies all three gaps. The case study in this paper traces a cross-border supply chain settlement scenario and finds that even where a single rail performs adequately within a national or eurozone context, cross-border reachability introduces additional latency, correspondent bank intermediation, and confirmation uncertainty. Wero's geographic concentration in its current launch markets means that an agentic supply chain involving counterparties in, for example, Poland, the Netherlands, or non-EU suppliers cannot rely on Wero for end-to-end settlement. SEPA Instant's reachability is broader but its cross-border performance depends on bilateral interbank connectivity that is not uniformly established across all eurozone institutions. The digital euro would, if deployed, provide a common instrument across eurozone jurisdictions, but its non-eurozone reach, critical for supply chains involving EU trading partners outside the currency union, remains an open architectural question.

The compliance embedding insight from the stablecoin literature points to a convergence opportunity. See and Tan [10] demonstrate that embedding AML and transaction limit compliance as on-chain guardrails executed at settlement time preserves settlement speed while satisfying regulatory obligations, a design pattern that resolves the apparent conflict between regulatory compliance and agentic latency requirements. This pattern is directly transferable to the digital euro's architecture if the programmability layer is designed to include compliance condition evaluation as a native function. It is partially transferable to SEPA Instant through the ISO 20022 structured data fields that can carry compliance attestations alongside the payment instruction. It is not transferable to Wero in its current form because no programmable or structured-data layer is exposed.

The cost dimension remains the most empirically underdetermined. The Instant Payments Regulation establishes a structural ceiling on SEPA Instant pricing, requiring that the fee for instant credit transfer not exceed the fee for standard credit transfer. This is a concrete cost-structural constraint derived from a regulatory instrument. However, it establishes a ceiling without specifying how that ceiling translates into per-transaction pricing at micro-transaction volumes, and standard credit transfer pricing is itself variable across institutions and markets. For an agentic system executing thousands of transactions per hour, the absence of published volume-tiered fee schedules across all three rails is an evidentiary gap with direct practical consequences for deployment decisions.

The infrastructure debate has been framed incorrectly as a binary choice between DLT-native and traditional rails. The agentic payment protocol literature [9, 10] operates primarily in a cryptocurrency and stablecoin context, which can create the impression that agentic commerce requires DLT infrastructure by definition. The comparative analysis in this paper does not support that conclusion. The capabilities that agentic commerce requires, programmability, atomic finality, machine-readable confirmation, are achievable through standardisation of European public rails, given sufficient regulatory mandate and technical investment. The structural advantages of DLT-native approaches, including programmability and atomicity, are real but arise from specific design choices that can, in principle, be replicated in a public rail context. The relevant analytical distinction concerns which specific capability gaps in European public rails can be closed through incremental standardisation versus which require architectural redesign, rather than any inherent superiority of one infrastructure type over another.

This paper has conducted a structured evaluation of three European payment rails, SEPA Instant Credit Transfer, Wero, and the proposed digital euro, against five technical criteria derived from the requirements of agentic commerce. The principal finding is that no single rail, in its present documented state, satisfies all five criteria simultaneously. The comparative assessment reveals that each rail carries a distinct and specific pattern of capability and gap, and that the nature of the gaps differs structurally across the three.

SEPA Instant meets the speed threshold and provides irrevocability upon the scheme-level pacs.002 confirmation, but the binding characterisation for finality is Partial because no mandatory payee-side notification standard propagates that irrevocability to the payee's system within the settlement window. The scheme also exposes no conditional payment primitive, leaving atomicity as an application-layer responsibility. The path to agentic readiness for SEPA Instant therefore runs through two specific standardisation actions. The first is the EPC mandating a push-notification schema for credit confirmation at the payee level, specifying both the message format, whether ISO 20022 camt.054 or a defined JSON equivalent, and the delivery latency obligation in seconds, together with a retry and failure-acknowledgement protocol that automated systems can consume without human intervention. The second is the introduction of a conditional payment annex to the scheme rulebook that defines the permissible condition types, such as delivery attestation events, time-bound escrow releases, and sensor-verified state transitions, and the technical execution environment in which those conditions are evaluated at the moment the payment instruction is submitted.

Wero's gap is located at the interface layer rather than the settlement layer. The scheme currently exposes no developer API, no authentication standard for non-human initiators, and no machine-readable confirmation schema. Geographic expansion to additional EU markets would increase Wero's reachability for cross-border agentic workflows, but reachability is secondary to interface accessibility: a rail that cannot be programmatically addressed by an autonomous agent is unavailable for agentic commerce regardless of the settlement performance of its underlying infrastructure. Closing Wero's agentic gap requires EPI Company to publish a versioned developer API aligned with the FAPI security profile, to define how delegated authority from a human principal is established and revoked for a non-human initiator, and to specify the structured confirmation event schemas that a payee's automated system would consume upon settlement.

The digital euro carries the strongest architectural potential of the three rails for programmable and conditional settlement. Its central bank finality is the most legally certain available in the European system. However, its programmability layer remains unspecified, its developer interface has not been published, and the tension between the transaction-monitoring requirements of a centralised ledger and the privacy commitments in the legislative proposal has not been resolved architecturally. The specific form that resolution must take is consequential: if conditional payment execution occurs on the central ECB ledger, the evaluation of conditions against transaction data creates a monitoring surface that the legislative privacy commitments were designed to foreclose; if it is delegated to intermediary PSPs, the programmability guarantee becomes dependent on intermediary implementation consistency, which reintroduces the fragmentation that a common instrument is intended to eliminate; and if a hybrid model is adopted, the boundary between the two execution environments must be specified with enough precision that third-party developers can build to it reliably. Publishing a technical architecture paper that resolves this question is the prerequisite for any subsequent developer interface work. The timeline for these design decisions, aligned with the pilot phases and eventual deployment schedule, means that the digital euro cannot be treated as an available rail for agentic commerce in the near term, and practitioners should not model it as a deployment option for workflows that must be operational within the next two to three years.

For practitioners designing agentic commerce systems in Europe in the current period, the consequence of this analysis is that SEPA Instant is the only rail through which a near-term agentic settlement workflow can proceed, and that workflow must manage three residual gaps explicitly: the absence of a guaranteed machine-readable confirmation at the payee side, which requires either polling through an account information API or accepting a probabilistic settlement assumption; the absence of a conditional payment primitive at the rail level, which requires the application layer to bear full responsibility for the ordering and recovery logic of conditional workflows; and the absence of published micro-transaction fee schedules, which requires cost modelling to proceed on assumptions about pricing ceilings rather than on measured per-transaction costs at volume.

The medium-term standardisation agenda that follows from this analysis has four principal components. The first is the EPC, in coordination with the EBA under the PSD3 framework, specifying and mandating the payee-side confirmation webhook for SEPA Instant. The second is EPI Company publishing a developer API for Wero that covers non-human initiation, authentication, and structured confirmation. The third is the ECB and the European Commission publishing a technical architecture paper that resolves the programmability-privacy tension for the digital euro and defines the API surface available to third-party developers. The fourth is an independent benchmarking study producing published micro-transaction cost data for all three rails at volumes representative of agentic commerce, covering the range from thousands to millions of transactions per day, so that operators can perform grounded deployment cost analysis rather than structural inference from pricing ceilings.

This analysis is subject to the following specific evidentiary limitations.

1. Absence of published agentic use case pilot data for European rails. No published pilot program, proof-of-concept report, or stress-test result has been identified that directly measures the performance of SEPA Instant, Wero, or the digital euro under high-frequency automated transaction loads representative of agentic commerce. The characterisations in the comparative assessment are therefore derived from scheme specifications and regulatory documents rather than empirical performance data. Actual latency, throughput, and confirmation behaviour under agentic load conditions may differ materially from scheme-level specifications.

2. Incomplete digital euro technical specifications. The digital euro remains in a design and pre-pilot phase. The specific API architecture, programmability mechanisms, developer interface specifications, and confirmation message schemas have not been published in a form that permits technical evaluation. The characterisations assigned to the digital euro in this analysis reflect the design documentation available at the time of writing and will require revision as specifications are published during the pilot and eventual deployment phases.

3. Absence of published Wero machine-interface documentation. The evaluation of Wero against all five criteria is constrained by the absence of published developer API documentation, authentication standards for non-human initiators, and machine-readable confirmation schemas. It is possible that EPI Company has developed or is developing such documentation that has not been made publicly available. The characterisation Absent in this context reflects the public evidence base, not a definitive conclusion about Wero's technical capabilities.

4. Lack of cross-border empirical performance data. The paper's case study and discussion of cross-border agentic settlement rely on the structural characteristics of each rail's geographic architecture rather than on measured cross-border transaction latency, finality propagation time, or confirmation reliability under cross-border conditions. Empirical testing of cross-border flows involving all three rails has not been identified in the literature.

5. Absence of micro-transaction volume pricing data. The Instant Payments Regulation establishes a concrete structural ceiling on SEPA Instant pricing, as discussed in the body of this paper. The evidentiary limitation is not an absence of cost information at the regulatory level but an absence of published micro-transaction volume fee schedules for any of the three rails. Standard credit transfer pricing, which sets the ceiling for SEPA Instant, is itself variable across institutions and markets. The cost characterisations in this paper therefore represent structural inferences from pricing frameworks rather than measurements at the transaction volumes characteristic of agentic commerce, and that inference does not substitute for published schedules when operators must make deployment cost decisions.

The gaps identified in this analysis point to five specific research and standardisation actions that would materially advance the evidence base.

1. Agentic commerce API standard for SEPA Instant. The European Payments Council, in coordination with the European Banking Authority under the PSD3 framework, should initiate a working group to specify a mandatory webhook standard for payee-side credit confirmation. The technical output required is a defined JSON or ISO 20022 schema for the credit notification event, a maximum delivery latency obligation in seconds, and a retry and failure-acknowledgement protocol compatible with automated system integration.

2. Wero developer interface publication. EPI Company should publish a versioned developer API specification covering payment initiation by non-human principals, machine-identity authentication and delegated authority standards, and structured confirmation event schemas. This specification should be developed in alignment with the FAPI (Financial-grade API) security profile to ensure compatibility with existing open banking frameworks.

3. Digital euro programmability architecture resolution. The European Central Bank and the European Commission should publish a technical architecture paper that resolves the programmability-privacy tension identified in the legislative process, specifying whether conditional payment execution occurs on the central ledger, at the intermediary PSP level, or through a hybrid model, and defining the API surface exposed to third-party developers in each case.

4. Cross-border interoperability pilot. A structured pilot involving at least three eurozone markets and at least two non-eurozone EU markets should be conducted using SEPA Instant, measuring end-to-end latency, finality propagation time, and confirmation reliability for automated transaction loads at rates representative of agentic commerce. The pilot should publish transaction-level performance data under an open data commitment.

5. Micro-transaction cost benchmarking study. An independent benchmarking study should be commissioned, covering per-transaction costs for SEPA Instant, Wero, and projected digital euro intermediary pricing at transaction volumes ranging from one thousand to one million transactions per day, to provide agentic commerce operators with the cost data necessary for deployment decisions.