The acquirer counter-move to network-led agentic rails

In the standard account of payment infrastructure evolution, networks define the rails, issuers manage liability, and acquirers compete on execution. From magnetic stripe through EMV, tokenization, and 3DS, each time the network defined a new standard, acquirers adapted their technical stacks to comply and then competed on cost and service quality within the standard's boundaries.

Agentic commerce breaks that pattern in one specific way. The transactions it generates are initiated by a software agent acting on standing instructions, often without synchronous human approval, across multi-step commerce tasks (procurement workflows, subscription management, dynamic service bundling). The agent, not the cardholder, is the proximate decision-maker. That shifts the question of who governs the decision from an issuer-cardholder relationship to a framework question about agent identity, delegated authority, and runtime policy. None of which existing scheme rules, as written for human-initiated transactions, were designed to address.

Visa's Intelligent Commerce initiative and Mastercard's Agent Pay programme each attempt to resolve this framework question by defining it at the scheme level. Both initiatives assign the anchoring of agent credentials, the scoping of spending authority, and the provisioning of trust tokens to the network layer. The implication is direct: whoever holds the credential-issuance and trust-anchor function holds the rule-setting authority for agentic commerce, and both major networks have moved to occupy that position before any acquirer or PSP has established an alternative.

The acquirer and PSP landscape entering this transition is not homogeneous. Adyen operates as a unified processor-gateway-risk stack, processing card-not-present volume for enterprise merchants globally, with software and platform revenue as a growing share of gross profit. Stripe operates as a developer-first payments platform whose margin profile is weighted toward software services layered above the payment credential. Worldline and Nexi are the product of successive consolidations in European processing, with a geographic concentration in interchange-regulated markets and a legacy infrastructure base that constrains the pace of technical change. PayPal, considered here specifically in its PSP role rather than as a wallet or network, operates a dual-sided platform where merchant checkout integration sits alongside consumer identity and credit products.

Each of these firms enters the agentic commerce transition with a different set of incentives. For Adyen, scheme-defined agentic rails threaten to commoditise the gateway and routing logic where some of its differentiation sits, but leave open competition at the risk and data layers. For Stripe, the software and orchestration services that sit above the payment credential are the primary margin source, and the primary surface that scheme-defined agent frameworks could bypass or absorb. For Worldline and Nexi, the question is whether their current processing scale translates into any leverage in a standards-driven environment, or whether scheme compliance becomes a cost with no corresponding competitive moat. For PayPal, the identity and consumer trust layer it owns is potentially a credentialing asset in an agentic world, but its PSP function faces margin pressure from the same structural forces as its peers.

This paper uses four signal streams (public statements from leadership and developer documentation, hiring patterns by role and function, partnership and integration announcements, and patent filings by subject area) to triangulate the strategic posture of each firm. The methodology section describes how these signals are selected, weighted, and reconciled when they conflict. The results section classifies each firm into one of three readiness postures: building, conditional, or waiting. The discussion section explains why business model rather than technical capacity is the primary determinant of that classification.

The payment industry has a well-established model for how scheme innovation propagates. The scheme defines a standard. The issuers and acquirers certify compliance. Vendors build to the certified specification. Merchants receive the benefit, typically lower fraud or higher authorisation rates, without needing to understand the underlying change. Each party in the chain adds margin by being the preferred certified implementation of the scheme's requirement.

Agentic commerce disrupts this model at the point of initiation. A human cardholder initiating a transaction is, legally and operationally, the party whose consent anchors the payment. Scheme rules, dispute frameworks, and fraud liability assignments all trace back to that moment of human initiation. When a software agent initiates the transaction instead, the consent structure changes in ways that the existing framework does not address cleanly.

The question of where agent authority is anchored (at the issuer, at the merchant, at the PSP, or at the agent's operating environment) is a governance question rather than a technical one with a unique correct answer. Whoever answers it first defines the commercial relationships that follow. Visa Intelligent Commerce and Mastercard Agent Pay have each staked out answers: both centre the credential and trust-anchor at the network, with delegated scopes provisioned through the issuer. That is a deliberate structural choice, not a technical necessity. A PSP-anchored alternative, where the acquirer or gateway issues the agent credential and the network accepts it downstream, was available, and both schemes bypassed it.

For acquirers and PSPs, this choice has direct margin implications. The software and platform revenue that Adyen, Stripe, and others have built above the payment credential depends on merchants purchasing orchestration, fraud management, and decisioning services from the PSP layer rather than receiving those services as part of the scheme's standard offering. If scheme-defined agentic rails absorb the decisioning layer (determining which agent can spend, up to what amount, under what conditions), the PSP's value-added services compress toward commodity execution.

The exposure is not uniform. Firms whose revenue is concentrated in processing volume are less immediately threatened, because agent-initiated transactions still require processing. Firms whose revenue depends on the software services layered above the credential (routing logic, fraud scoring, checkout experience) face a more direct displacement risk. The interchange economics underlying European acquirers (Worldline, Nexi) are already compressed by the EU Interchange Fee Regulation (IFR), which caps interchange rates on consumer card transactions, making software revenue proportionally more critical; scheme absorption of the agentic software layer would therefore hit those firms harder than it hits a processor operating in uncapped markets.

The other stake is liability. Machine-initiated transactions without synchronous human consent create new dispute surfaces. The responsible party when an agent overruns its spending scope is not yet defined in scheme rules or acquirer contracts. The allocation of liability when an agent is manipulated through a prompt injection attack that redirects a payment [7] remains similarly unresolved in scheme rules, consumer protection law, and acquirer contracts. Firms that build proprietary agent infrastructure before these questions are settled take on legal exposure that scheme-defined standards, once ratified, will redistribute. That is a reason some acquirers might rationally wait: the liability framework has not yet been written, and early builders absorb the risk of operating in its absence.

The history of acquirer and PSP response to scheme-driven standards offers a set of structural precedents that illuminate the current transition, and clarifies where the agentic case departs from prior pattern.

3D Secure. When Visa deployed 3DS 1.0 in the early 2000s, the scheme defined the authentication framework, issuers held the cardholder verification step, and acquirers were required to transmit the resulting liability shift indicators. The acquirer's role was compliance, not innovation. Acquirers who integrated quickly gained access to the liability shift; those who were slow faced higher chargeback rates on non-authenticated transactions. The competitive differentiation available to acquirers was in the smoothness of the authentication experience and the accuracy of their risk scoring that determined whether to challenge or pass-through. 3DS 2.x, deployed from the mid-2010s, expanded the data exchange and gave issuers richer context for frictionless authentication, but the structural logic was the same: the scheme defined the protocol, the acquirer executed within it.

The agentic case is structurally different from 3DS in one important respect. 3DS was entirely about authenticating a known human cardholder (the scheme's interest was in establishing that the person presenting the credential was its legitimate owner). Agentic commerce requires authenticating a non-human agent acting under delegated authority from a human cardholder. The authentication object changes, and with it the entire liability and consent structure. No analogue to the 3DS liability shift exists yet for agent-delegated transactions.

Tokenization. The EMV tokenization standard, codified through the network token programmes run by Visa and Mastercard, replaced primary account numbers with scheme-generated tokens for card-on-file and device-based transactions. Acquirers and PSPs who integrated network tokenization early gained authorisation rate benefits and fraud rate reductions that they could pass to merchants as a value proposition. Stripe's developer-first compliance posture, documented in its technical publications and developer changelog, is consistent with early adoption of scheme standards as a merchant acquisition lever, a pattern observable across its integration with network tokenization, mandate frameworks, and 3DS 2.x. That posture is an inferred structural inference rather than a dated proprietary commitment, but it is consistent across the public record of Stripe's compliance-driven product launches.

Tokenization is the closer structural precedent for agentic credential provisioning, because both involve the scheme issuing a derived credential that scopes and governs usage. The difference is that network tokens are issued to static storage contexts (a device, a merchant's vault), while agentic credentials must be issued to dynamic, runtime agent contexts whose scope may change mid-session. That dynamism is what makes the agent credential problem technically harder and why the governance layer matters more.

Recurring and instalment mandates. The migration from merchant-held recurring credentials to scheme-registered mandates (Visa Stored Credential framework, Mastercard standing instructions) followed the same pattern: the scheme defined what a valid recurring authority looked like, acquirers built the transmission plumbing, and merchants who used the scheme-defined mandate received better authorisation rates. PSPs differentiated on how well they abstracted the complexity from the merchant.

In each of these cases, the acquirer's ability to differentiate depended on the gap between scheme minimum compliance and best-practice implementation. The question the agentic transition poses is whether that gap will be large enough, and will persist long enough, for PSP-side differentiation to generate durable margin, or whether the scheme's definition will be comprehensive enough to close the gap faster than prior standards cycles.

Recent research on production agentic deployments suggests the gap is real and structural. The POLARIS framework for governed agentic execution [4] demonstrates that generic multi-agent setups fail systematically on auditability and policy alignment, problems that a scheme-defined rail will not automatically solve unless governance primitives are baked into the standard itself. Adobe's production deployment of ReAct agents for e-commerce observability [3] shows that the operational reliability requirements for agentic systems in commerce contexts are substantially higher than for conventional automated processes, and that the build effort to meet those requirements is non-trivial. These findings suggest that the compliance-to-differentiation gap in agentic rails will be wider than in prior scheme innovation cycles. That advantage will favour the firms that invest in the governance and reliability layer, rather than those who simply certify to the credential standard.

Inferring strategic posture from publicly available signals requires a structured approach to source selection, signal weighting, and conflict resolution. The firms under analysis (Adyen, Stripe, Worldline, Nexi, PayPal-as-PSP) do not publish strategy documents that map directly to agentic readiness. What they do publish, in aggregate and over time, is a pattern of public commitments, workforce configuration, external alliances, and intellectual property claims that, read together, allow a probability-weighted inference about where each firm is directing resources.

Signal stream one: public statements. This stream includes investor communications, developer documentation, conference presentations, and press releases. The limitation is that public statements are curated for audience effect: investor calls emphasise growth narratives, developer documentation reflects current capability rather than roadmap, and press releases announce partnerships that have already closed. Public statements are therefore treated as confirming signals rather than leading indicators: they establish a floor of capability or commitment, not a ceiling.

Signal stream two: hiring patterns. Job postings, LinkedIn role distributions, and reported headcount changes in specific engineering functions are the highest-velocity signal available. A firm hiring at volume for ML infrastructure engineers with stated experience in stateful agent pipelines, or for product managers with backgrounds in identity and credentialing, is committing capital before any public announcement. The limitation is that job postings reflect authorised headcount, not filled roles, and a posting can be withdrawn or frozen. Hiring signals are weighted above public statements for leading-indicator purposes, but discounted when the role descriptions are generic enough to be interpreted across multiple strategic directions.

Signal stream three: partnership and integration announcements. Announced integrations with LLM providers (OpenAI, Anthropic, Mistral), agent orchestration platforms, or identity and credentialing infrastructure providers are a strong signal of build commitment. They represent contractual relationships, not aspirations. The limitation is that the terms of announced partnerships are rarely disclosed, making it difficult to distinguish a deep technical integration from a co-marketing arrangement. Partnership signals are treated as confirming when accompanied by developer documentation or API availability, and as tentative when the announcement precedes any technical artefact.

Signal stream four: patent filings. Patent applications in the relevant technology areas (session-coherence for agent pipelines, context propagation across multi-step transactions, agent credential scoping, anomaly detection in machine-initiated payment flows) represent the most durable form of build signal, because they require institutional commitment and engineering effort to produce. The limitation is publication lag: patent applications are typically published 18 months after filing, meaning the current visible patent landscape reflects activity from 18 to 36 months prior. Patent signals are therefore treated as confirming prior investment rather than current direction.

Conflict resolution. When signals contradict, for example a public statement asserting agentic readiness combined with no observable hiring or patent activity in the relevant functions, the hiring and patent signals are given priority over the public statement. The reverse case, where hiring and patent signals are strong but public statements are absent or vague, is interpreted as deliberate pre-announcement posture rather than as evidence against the inferred direction.

Evidentiary basis and scope of inference. The research corpus available for this analysis is weighted toward academic and technical sources rather than competitive intelligence sources [3][4][5][6][7][8]. Direct patent filings and granular hiring data for the specific firms analysed are not available in the corpus at the level of detail the methodology would ideally require. The results and case study sections therefore operate as illustrative structural inferences: they apply the business-model logic and the signal-weighting framework to each firm's publicly known architecture and revenue structure, and they identify the posture that is most consistent with those structural facts. Individual signal attributions (hiring patterns, partnership signals, patent activity) are used as directional indicators consistent with the inferred posture, not as directly observed and verified data points. The limitations section addresses the uncertainty this introduces, and readers should treat the firm-level classifications as probability-weighted structural inferences rather than empirically confirmed positions.

The signal triangulation produces a classification of each firm into one of three readiness postures with respect to scheme-led agentic rails. These postures are defined by the convergence of structural inferences and available signals rather than by any single data point. Where the available signals are thin or contradictory, the classification carries wider uncertainty bounds, noted explicitly below. The posture labels (building, conditional, waiting) describe the direction of current capital allocation as inferred from the structural analysis; they are deterministic forecasts in no sense.

Visa and Mastercard: standard-setters, not acquirers. Visa Intelligent Commerce and Mastercard Agent Pay are the structural anchors of this analysis. Both programmes have been announced publicly, both involve the scheme taking responsibility for the agent credential and trust-anchor function, and both are positioned to reach general availability within a timeframe that creates a hard deadline for acquirer response. The key structural feature of both programmes is that they define agentic commerce as a network service, with the credential provisioned through the issuer and the scope governed by scheme rules. This is the stated design, and it is not a prediction about where authority will sit. The consequence for acquirers is that their position in the agentic transaction flow is determined by the scheme's specification rather than by their own technical choices, unless they can establish a credentialing function that the scheme accepts as equivalent or complementary.

Adyen: Building with structural advantage. Posture: Building. Adyen's integrated stack (processor, gateway, risk engine, and data platform operating under a single technical architecture) provides a structural basis for building agentic infrastructure that pure-play acquirers lack. The unified architecture means that session-coherence and context propagation for agentic transactions can be implemented once, at the platform level, rather than stitched across vendor boundaries. The case study section of this paper traces Adyen's specific signal profile in detail; the summary finding is that Adyen's business model creates the strongest structural incentive to build a proprietary agent layer, because both its platform revenue and its processing volume are at stake if scheme-native rails absorb merchant decisioning. Adyen's risk and data products represent the surface most directly threatened by scheme-defined agentic governance, and the surface where proprietary build is most defensible.

Stripe: Building at the orchestration layer. Posture: Building. Stripe's developer-first positioning and its documented pattern of early adoption of scheme standards are consistent signals of a building posture. Stripe's margin is concentrated in the software and platform services layered above the credential: Radar (fraud), Billing (subscription and mandate management), and the broader Connect infrastructure. Each of these is a direct surface of scheme-defined agentic rails.

The building posture here is reinforced by the fact that this software layer is also the surface most threatened by scheme-defined rails. A PSP whose software services could be absorbed by scheme-native agent governance has two available strategies: build an orchestration layer that is demonstrably more capable than the scheme's native offering before general availability, or accept the margin compression that follows from scheme absorption. Stripe's posture, inferred from its product investment pattern and compliance-first track record, is the former. The absence of a proprietary credential-issuing function (Stripe is not an issuer) means it cannot compete at the trust-anchor layer. It can, however, build a decisioning and orchestration layer that operates above the scheme credential and provides merchants with capabilities the scheme's native rail does not: richer context propagation, governance primitives for multi-step procurement, and dispute management infrastructure for machine-initiated flows. The window for that build closes at scheme general availability, which is why the posture is building rather than conditional.

Worldline: Conditional posture. Waiting on specification. Worldline's signal profile is less consistent. Its scale in European processing gives it volume leverage, and its geographic concentration in IFR-regulated markets has already compressed interchange margins, making software revenue proportionally more critical. The signals available do not indicate a committed build programme for agentic infrastructure beyond standard scheme-compliance preparation. The more probable inference is that Worldline is waiting for Visa and Mastercard to publish technical specifications before committing capital, a rational choice given the cost structure of large-scale legacy infrastructure and the uncertainty about where scheme-defined agent governance will sit. This conditional posture is distinct from Nexi's waiting posture: Worldline retains the organisational scale and merchant relationships to execute a build programme once specifications are published, and the conditional classification reflects deferred commitment rather than structural incapacity. The risk is that waiting until specification publication may leave insufficient time to build a differentiated layer before scheme general availability, converting the conditional position into functional equivalence with the waiting cohort.

Nexi: Waiting, with structural exposure. Posture: Waiting. Nexi's position is structurally more exposed than Worldline's, and the distinction matters. Its revenue base is concentrated in processing volume across Southern European markets, where IFR interchange caps constrain processing margin. As a structural inference (and not an assertion supported by direct merchant-mix data), Nexi's core merchant base skews toward segments where agentic purchasing agents will appear later in the adoption curve than in the high-volume enterprise e-commerce environments served by Adyen and Stripe. That inference follows from the geographic and market-segment concentration of Nexi's disclosed business, not from a direct measurement of e-commerce penetration rates. The practical consequence is that the near-term pressure to build is lower, but the revenue available to fund a build programme is also lower. The structural risk is that when agentic commerce penetration reaches Nexi's core merchant base, its competitors will carry 24 to 36 months of build advantage. Unlike Worldline's conditional posture, Nexi's waiting posture does not appear to be accompanied by a deferred build plan triggered by specification publication.

PayPal-as-PSP: Building at the inference layer. Posture: Building, narrowly. PayPal's fine-tuning of NVIDIA's NeMo framework for its commerce agent, documented in the production corpus [8], represents the most concrete evidence of PSP-side agentic build activity available in this analysis. The NeMo framework fine-tuning effort specifically targeted retrieval latency as the dominant bottleneck in agent response time, accounting for more than half of total agent response latency, and resolved it through a domain-specific small language model. This is production infrastructure investment, not prototype work. The open empirical question the signal raises is whether this effort is a narrow retrieval optimisation for PayPal's existing commerce agent, or the visible surface of a broader strategy to own the merchant-side agent orchestration layer. The identity assets PayPal holds (consumer accounts, transaction history, credit products) are potential inputs to agent credentialing, which could position PayPal as a trust anchor for agent identity independently of the scheme programmes. Whether that positioning is being actively built, or whether the NeMo framework work is a standalone performance improvement, is not determinable from available public signals alone.

The three-way classification (building, conditional, waiting) maps with reasonable consistency onto the business model topology of the firms analysed. That correspondence is not coincidental. It reflects the mechanism by which business model shapes the return on investment from building agentic infrastructure, and the mechanism by which existing margin structures determine what is at risk from abstaining.

Why integrated stacks build first. Adyen and Stripe both exhibit building postures, and the reason is that both have software margin concentrated in the layer that scheme-defined agentic rails are most likely to absorb. For Adyen, the risk layer and data products depend on merchant routing decisions and fraud-scoring relationships that could migrate to scheme-native governance. For Stripe, Radar, Billing, and the Connect infrastructure provide the orchestration services that an agent framework would otherwise require from the merchant. The value proposition in both cases depends on the merchant choosing the PSP's software layer over the scheme's native offering, and that choice is only available if the PSP's layer exists and is demonstrably superior before the scheme's offering reaches general availability.

The inference from persistent agent-state research is relevant here. Aethon's reference-based instantiation approach [5] demonstrates that constant-time agent instantiation across sessions is a solved problem in principle but requires deliberate architectural investment to implement at production scale. The codified context approach [6] establishes that maintaining consistent agent behaviour across long-horizon tasks (multi-step procurement, subscription management) requires explicit context infrastructure that does not emerge automatically from LLM deployment. Both findings point to the same structural conclusion: the firms that build persistent-state infrastructure for agentic commerce pipelines will produce more reliable and commercially viable agent integrations than those relying on stateless scheme credential flows, and the performance differential will be observable in authorisation rates, fraud outcomes, and dispute rates.

Why consolidated processors wait. Worldline and Nexi exhibit conditional and waiting postures respectively, and the reason is structurally different from Adyen and Stripe's position. Both are primarily processors. Their margin comes from volume throughput, scheme certification, and geographic distribution, rather than from the software layer above the credential. A scheme-defined agentic standard does not threaten processing margin; it creates a new compliance requirement, which processors build to and certify. The incentive to build ahead of the standard is therefore weaker, because the return does not accrue from differentiated software but from being first to certify to the new standard, and certification is available to all compliant processors simultaneously.

The complicating factor for both firms is their European regulatory environment. PSD2 open banking requirements, NIS2 cybersecurity obligations, and the emerging European AI Act each create regulatory interactions with agentic commerce that are more complex in European markets than in the US or Asia-Pacific. The trust and security literature, particularly the cross-layer attack vector analysis and the limitations of static Software Bill of Materials for agentic pipelines [7], suggests that European regulators will demand runtime auditability of agentic transactions that static credential flows cannot provide. If that regulatory demand materialises before scheme general availability, it could create a compliance surface on which Worldline and Nexi need to build regardless of their strategic preference.

PayPal's dual-layer tension. PayPal's position is the most analytically complex in the sample. The production evidence of NeMo framework fine-tuning for agentic commerce [8] places it firmly in the building cohort at the inference-optimisation layer. The strategic question is whether that build effort connects to PayPal's identity and credentialing assets in a way that creates a durable competitive position, or whether it remains a performance optimisation within a platform that is structurally subject to the same margin compression as other PSPs.

The identity layer is the key variable. PayPal holds consumer identity, transaction history, and credit relationships that are direct inputs to the trust-anchor function that Visa Intelligent Commerce and Mastercard Agent Pay are building at the network level. If PayPal can establish that its identity infrastructure qualifies as a trusted delegation source under scheme rules, such that an agent credentialed through PayPal's consumer identity carries the same trust level as a scheme-provisioned token, it converts its consumer platform into a structural position in agentic commerce rather than a commodity PSP function. The public signals do not yet indicate whether this is the direction of the NeMo framework and broader commerce agent investment, or whether those investments are oriented toward checkout conversion optimisation within the existing PSP model.

The governance and auditability differentiator. The finding from the POLARIS governance framework [4], that generic multi-agent setups fail on auditability and policy alignment in ways that purpose-built governance architectures do not, has a direct implication for the competitive analysis. If the PSP layer is the appropriate site for agentic governance (as opposed to the scheme layer), then the firms building governance primitives into their agent infrastructure are building a position that is difficult to replicate quickly. Audit trail generation, anomaly routing for machine-initiated payment flows, and policy-aligned execution are features that cannot be added to a credential-pass-through architecture after the fact. They require architectural decisions at the platform level that determine what data is captured, how it is structured, and how it is surfaced to merchants and regulators.

Adobe's production deployment of ReAct agents for alert triage [3] demonstrates that the observability and auditability requirements for agentic systems in commerce contexts exceed what conventional monitoring stacks provide. That finding, applied to PSP infrastructure, suggests that the firms building agentic observability alongside agentic execution, rather than treating observability as a retrospective compliance requirement, will produce better risk outcomes and therefore better fraud and dispute rates. Better fraud and dispute rates are the commercial language that enterprise merchants understand, and they are the metric on which PSPs differentiate in practice.

Business model as the binding constraint. Worldline, for example, has the engineering scale and the merchant relationships to build a proprietary agentic layer. The signal pattern does not indicate it is doing so, and the explanation lies in incentive rather than capability. The processing business model rewards compliance speed over differentiated build. That is a rational response to the scheme-innovation history described in the related work section, and it is likely correct for the short-to-medium term. The risk is that it may be incorrect if scheme general availability arrives with specifications that are less comprehensive than currently signalled, leaving a larger gap for PSP-side differentiation than processors have budgeted for.

The central claim of this analysis is that Visa and Mastercard's decisions to define agentic commerce standards at the network level, rather than leaving that definition to the PSP and acquirer community, have created more than a new compliance requirement. They have determined the branching structure of the competitive responses available to each acquirer type, and they have done so before most acquirers have publicly committed to a direction.

The branching logic runs as follows. Once a scheme defines the credential-issuance and trust-anchor function, the available competitive spaces are: the governance and orchestration layer above the credential (where Stripe and Adyen are building); the identity layer that feeds into the credential (where PayPal has latent assets but unconfirmed direction); the processing and certification layer below the credential (where Nexi operates with low differentiation risk but equally low upside); and the data and risk layer that agent-generated transaction streams will produce in volume (where firms with integrated stacks have structural advantages over pure-play processors). Worldline occupies a distinct intermediate position: its conditional posture preserves the option to build a differentiated layer once scheme specifications are published, but that option degrades if the interval between specification publication and scheme general availability is shorter than the build effort requires.

The signal analysis presented here is a precise forecast. The evidentiary base is weighted toward academic and technical sources, with limited direct access to the patent filings, hiring data, and internal strategy documents that would allow finer-grained classification. The limitations section addresses this constraint explicitly. What the analysis does establish is the mechanism by which business model translates into readiness posture, and that mechanism is durable regardless of which specific firms ultimately occupy which positions.

The timing pressure is real and asymmetric. Firms in the building cohort face the risk of building to specifications that shift before scheme general availability, a stranded-investment risk that is most acute where build decisions depend on assumptions about the scope of scheme-native agent governance. Firms in the waiting cohort face the risk that scheme general availability arrives with a comprehensive specification that leaves no gap for PSP-side differentiation, converting their waiting period into a structural disadvantage that compounds over each subsequent scheme innovation cycle. The conditional posture adopted by some processors is rational as a capital-preservation strategy, but carries the embedded assumption that the specification gap will be large enough and the build window long enough to close the distance after publication, an assumption that prior scheme cycles do not uniformly support.

The three-tier differentiator hierarchy identified in this analysis, with persistent agent-state infrastructure as the foundational layer, governance and auditability primitives as the second tier, and agent-identity trust structure as the outer layer, clarifies where the competitive outcomes will register. Persistent agent state determines whether a PSP can maintain coherent context, enforce spending policy, and generate auditable records across multi-step commerce tasks; it is the prerequisite on which the upper tiers depend. Governance and auditability primitives determine whether that state infrastructure produces fraud and dispute rates that enterprise merchants can accept and regulators can audit; they are the commercial translation of persistent-state investment into merchant-facing outcomes. Agent-identity trust structure determines whether a PSP's credentialing function integrates with scheme-level trust anchors or operates as a parallel system; it is where the competitive boundary between PSP-layer and scheme-layer governance is ultimately drawn.

Acquirers that build across all three tiers before scheme general availability own the infrastructure through which scheme-defined credentials operate. Acquirers that build at the processing and certification tier alone will process the transactions, at the margins available to certified processors in a scheme-governed market. The interval between now and scheme general availability is the period in which that distinction is made.

The analysis presented in this paper is an inference from publicly observable signals, not a direct observation of strategy. Several limitations constrain what can be concluded with confidence.

Signal noise in hiring and patent data. Job postings are approved headcount, not filled roles. A firm can post for machine learning infrastructure engineers and withdraw the posting without hiring. A patent filing reflects past engineering effort, not current strategic direction. Both signals are directionally informative in aggregate and over time, but neither is individually reliable. The analysis weights them as confirming signals rather than sole determinants of posture, but even as confirming signals they carry material noise.

Private strategy diverges from public signal. Large payments firms routinely build infrastructure and form partnerships without public announcement, particularly in competitive technology areas. The absence of a public signal (no press release, no developer documentation, no visible patent) does not confirm the absence of a build programme. Worldline or Nexi may have internal agentic commerce programmes that have not yet produced public-facing artefacts. The analysis interprets signal absence as weak evidence of non-build, not as evidence of confirmed non-build.

Evidentiary basis of firm-level inferences. The research corpus underpinning this analysis is drawn from academic and technical preprints [3][4][5][6][7][8] rather than from proprietary competitive intelligence. Firm-level signal attributions (hiring patterns, patent activity, partnership posture) are structural inferences derived from business model analysis and publicly known firm characteristics, not from directly observed and verified data points. The firm classifications should be read as probability-weighted structural inferences: the most probable posture given the available evidence, not a confirmed empirical finding.

Regulatory shifts can overturn inferred positions. The European AI Act, PSD3 developments, and potential national implementations of NIS2 as applied to AI-driven payment processing could each create compliance requirements that override the strategic choices inferred here. A regulatory mandate for runtime AI auditability in payment processing, for example, would force all processors to build the governance layer regardless of their current posture, collapsing the differentiation that the building cohort is currently constructing. The analysis does not model regulatory scenarios; it describes the current competitive structure.

Scheme specification uncertainty. The scope and completeness of Visa Intelligent Commerce and Mastercard Agent Pay specifications at general availability is not known in advance. The analysis assumes that scheme specifications will leave a meaningful gap at the governance and orchestration layer; if specifications are more comprehensive than currently signalled, the strategic value of PSP-side build programmes shrinks correspondingly. That assumption is the most consequential uncertainty in the entire analysis.

The inferred positions described in this paper will produce observable outcomes as scheme general availability approaches and passes. Several concrete inflection points will serve as validation or refutation signals.

API launch timelines. The publication of developer documentation and API specifications for Visa Intelligent Commerce and Mastercard Agent Pay will establish the scope of scheme-defined functionality and the gap available to PSP-side differentiation. The gap between scheme API capability and PSP API capability, measured across governance, state management, and dispute handling functions, is the primary empirical test of the hypothesis.

Merchant routing decisions. Enterprise merchants who operate agentic purchasing agents will make routing decisions (which PSP or scheme-native credential flow governs agent transactions) that are directly observable through public merchant announcements, case studies, and payment volume disclosures. Routing concentration toward scheme-native flows would validate the margin compression scenario for waiting PSPs; routing concentration toward integrated PSP stacks would validate the building-cohort thesis.

Dispute and fraud rate patterns. Machine-initiated transactions without synchronous human consent will generate a distinct dispute pattern. PSPs and schemes that have built governance primitives into the transaction flow should show lower agent-transaction dispute rates than those relying on post-transaction remediation. Scheme and acquirer dispute rate disclosures, where available, are the operational metric most directly tied to the governance infrastructure investments analysed here.

Hiring and patent publication convergence. As patent publication lag resolves and current hiring patterns translate into shipped products, the correspondence between inferred posture and observable technical output will become testable. Tracking the alignment between current signal-based classifications and subsequent product launches is the direct empirical test of the signal triangulation methodology itself.