Stablecoin rails colliding with agentic commerce

Every payment system is, at its core, a set of answers to a specific set of questions about payer identity, fund availability, destination routing, payee certainty of receipt, and dispute resolution. The card network architecture that has dominated retail payments for sixty years provides answers to all five questions within a single credential and a single settlement protocol. The cost of that integration is the interchange fee, the chargeback mechanism, the processing latency, and the requirement that every participant maintain a bank account in good standing.

Stablecoin rails emerged, in part, to provide different answers to the last three questions (settlement timing, finality certainty, and dispute resolution) while leaving the first two largely unaddressed. A USDC transfer from one wallet to another settles in seconds with cryptographic finality, but the question of who authorized the transfer and whether the sending party had legitimate ownership of those funds sits entirely outside the protocol. That gap is precisely the space that KYC layers, custodial wallets, and compliance wrappers have rushed to fill.

Gap-filling is now happening simultaneously across four distinct architectural approaches, each built by an organization with a different strategic position and a different theory of where value concentrates in the stack.

PYUSD, issued by PayPal, operates as an instrument: a stablecoin that travels on Ethereum and Solana, denominated in US dollars, with PayPal providing the identity and compliance wrapper through its existing consumer account infrastructure. The friction PYUSD addresses is primarily custody and cross-chain settlement for users who already have PayPal accounts. It does not solve account-opening friction for new participants.

Circle's CPN is a network-layer announcement rather than a consumer product: a connectivity layer that routes USDC flows between regulated financial institutions, removing the correspondent banking inefficiencies that slow cross-border B2B payments without requiring a new consumer interface. The friction it addresses is interoperability between existing compliant entities.

Bridge, acquired by Stripe in late 2024, operates as an orchestration layer above multiple stablecoin instruments and rails. It provides a single API through which a business can move value in stablecoins, converting between instruments and corridors at the routing layer. The friction it addresses is the developer and operational cost of connecting to multiple separate rails.

Tempo operates as a cross-border stablecoin corridor, specifically designed for remittance and B2B payment flows in corridors where correspondent banking costs are high. Its friction target is the cost and delay of foreign exchange conversion in illiquid corridors.

Card-led Agent Commerce Protocols, as developed and piloted by the major card networks, address a different friction entirely: the authorization event in an agentic commerce context, where a software agent (rather than a human) initiates a purchase. The friction here is credential delegation and spending policy enforcement across automated workflows.

The problem this paper addresses concerns which frictions each system can actually reduce, whether those frictions sit in the same position in the payment flow, and under what transaction archetypes genuine substitution occurs. That framing requires a map of the friction landscape. Presupposing a single competitive dimension and ranking competitors accordingly would obscure the structural distinctions that matter most. A friction-based analysis, rather than a competitive ranking, is the appropriate instrument for this purpose.

The paper proceeds as follows. Section 2 establishes why this convergence is time-sensitive. Section 3 positions the analysis against existing literature on payment stack layering, platform economics, and rail selection. Section 4 describes the friction-based framework used to structure the analysis. Section 5 presents the friction profiles of each system. Section 6 analyzes where substitution occurs and where complementarity dominates. Section 7 synthesizes the competitive architecture. Sections 8 and 9 address limitations and future work. Section 10 traces a concrete supply chain payment through each architecture.

Three structural changes have arrived in proximity. Their simultaneity compresses the decision window for merchants, financial institutions, and platform operators choosing which rail architectures to integrate and on what timeline.

Stablecoin regulatory definition is being forced. The Markets in Crypto-Assets Regulation (MiCAR), now operative for asset-referenced and electronic-money tokens in the European Union, has produced a concrete restructuring of the stablecoin issuer market [20]. Circle's USDC has pursued authorization as an e-money token; Tether's USDT does not meet MiCAR requirements and has been removed from MiCAR-compliant EU venues, though enforcement across EU-accessible platforms remains uneven and USDT remains accessible in some contexts outside MiCAR-compliant exchange infrastructure. This bifurcation is a current market condition, not a speculative future state, even where its edges remain contested. The GENIUS Act in the United States, though still moving through its legislative process, will similarly define which stablecoin issuers qualify as payment stablecoin issuers and which fall into more restrictive investment-company or money-market-fund classifications. The resolution of that classification question will determine which rail operators can legally scale in the world's two largest payment markets [8]. Regulatory compliance is therefore a primary competitive filter, not a background condition.

Agent-initiated payments are transitioning from pilot to production. The large enterprise software platforms (both horizontal workflow orchestrators and vertical commerce platforms) have begun deploying AI agents that initiate procurement, subscription management, and vendor payment transactions without human approval on each individual transaction. The design question for a software agent choosing a payment rail is structurally different from the design question for a human consumer choosing a payment method. An agent can evaluate settlement finality, conversion cost, and corridor availability programmatically and select the optimal rail for each transaction. A human consumer reaches for the card in their wallet. This programmatic selection capacity means that in agentic B2B flows, substitution between rails becomes technically feasible in a way it never was for human-initiated retail transactions.

The orchestration layer is consolidating. The Stripe acquisition of Bridge is the most visible signal, but it is consistent with a broader pattern: platform operators are acquiring stablecoin infrastructure precisely to control routing decisions. A merchant that processes payments through Stripe does not choose payment rails; Stripe chooses them. If Stripe routes a transaction over Bridge's stablecoin infrastructure rather than over Visa's network, the merchant's interchange cost changes, the settlement timeline changes, and the dispute resolution mechanism changes, but the merchant's technical integration does not. That routing-level control, exercised at the orchestration layer, is where value capture concentrates as the underlying instruments commoditize [1].

These three pressures (regulatory definition, programmatic rail selection, and orchestration consolidation) converge on a single practical question for merchants, financial institutions, and platform operators: which system governs which transaction, and why. Getting this wrong in either direction carries cost. Over-investing in rail substitution that does not materialize produces stranded integration spend. Treating genuinely competitive rail architectures as permanently complementary creates exposure when routing decisions shift.

The academic literature on payment infrastructure provides three bodies of work directly relevant to this analysis: two-sided market theory, digital money and CBDC design, and platform infrastructure layering. Each offers tools that are necessary but individually insufficient for the question this paper addresses.

Two-sided market theory establishes that payment networks create value by connecting two distinct user groups (payers and payees) and that pricing and adoption dynamics in these markets follow rules distinct from those of single-sided markets [1]. A key structural implication is that the side charged lower prices is typically the side whose participation is more elastic. Card networks have historically applied this logic by subsidizing consumer participation (through rewards) while charging merchants interchange fees. The two-sided structure also generates network effects: a card is more valuable to a consumer when more merchants accept it, and more attractive to merchants when more consumers carry it. However, Liebowitz and Margolis [2] draw a precise distinction between network effects (which are real) and network externalities (which require that the individual-level adoption decision fails to internalize social costs). Their empirical finding, developed in the context of software markets, is that network externalities are difficult to document in practice. Applied by analogy to stablecoin rail dynamics, noting that the analogy is imperfect because stablecoin rail switching costs are primarily integration costs rather than the coordination costs that generate software market lock-in, the implication is that adoption scale creates coordination value but does not automatically produce the unassailable lock-in that competitive narratives around stablecoin rails sometimes assume. The structural difference is that a merchant switching between stablecoin rails abandons a technical integration, not a social coordination equilibrium, which makes the lock-in materially more fragile.

The mobile payment value network literature, structurally similar to the present inquiry, shows that banks and fintechs in mobile payment ecosystems occupy different nodes in the value chain and capture value through different mechanisms: card issuers through interchange, wallets through data, and network operators through transaction fees [21]. The ING and WeChat Pay cases demonstrate that the same underlying transaction can support different value capture points depending on which layer the customer-facing interface occupies.

Digital money and CBDC design literature provides the closest structural analysis of how new monetary instruments insert themselves into existing payment stacks. Ahnert et al. [14] analyze the economic trade-offs in CBDC design, particularly the tension between disintermediating commercial banks (by giving consumers direct central bank accounts) and preserving financial stability by routing CBDC through existing intermediaries. This tension directly parallels the stablecoin design choice between custodial wallets (which reintroduce intermediary friction) and non-custodial wallets (which remove KYC anchors). Da Cunha et al. [12] trace the transition from Bitcoin through stablecoins to CBDC proposals, identifying that each generation addresses the same settlement certainty problem but with different trust anchors. The IMF's multi-currency exchange and contracting platform research [16] proposes a unified multilateral settlement layer specifically for cross-border flows, structurally similar to Circle's CPN network-layer ambition, but with central bank participation providing the compliance anchor that CPN currently seeks through regulated financial institution membership.

Wang and Gao [15] analyze the CBDC network as a regulatory object, noting that its international coordination requires either new treaty structures or regulatory equivalence agreements, precisely the governance gap that private stablecoin cross-border corridors like Tempo currently fill by operating below the regulatory threshold that would trigger those requirements. The implication is that CBDC network maturation, if it occurs at regulatory speed, will not displace private stablecoin rails quickly in corridors where institutional adoption is already established.

Platform infrastructure layering literature, particularly Constantinides, Henfridsson, and Parker [4], establishes that digital platforms create value by providing generative boundary resources (APIs, data access, and protocol specifications) that allow third parties to build complementary services. The payment stack is a layered architecture in this sense: instrument, network, orchestration, and interface layers can be operated by different entities with different governance models. That layering logic provides the structural vocabulary for the friction decomposition in Section 4; it does not, however, address the specific case of autonomous software agents selecting among available rails, which is an agentic commerce dynamic absent from the platform infrastructure literature. The fintech entrant pattern identified by Fatorachian et al. [17] is directly relevant: regulatory approval functions as a barrier to entry, but once cleared, the incumbency advantages of the first-mover are eroded by the superior unit economics of the entrant's narrower focus. This pattern is consistent with the hypothesis that compliant stablecoin orchestrators (Stripe/Bridge) are positioned to erode card network margin in specific corridors, while non-compliant or unclassified stablecoin issuers face a different competitive ceiling set by regulatory risk rather than market preference.

Santner and Taudes [20] (working paper, 2025) provide the most current structural analysis of MiCAR's effect on the European stablecoin market, documenting the bifurcation between compliant e-money token issuers and non-compliant issuers. As a working paper, this source should be read as the current best available analysis of a regulatory environment that is itself still evolving, rather than as settled empirical finding.

What the existing literature does not address is the specific interaction between programmatic rail selection by software agents and the layering logic of payment infrastructure. The agentic commerce context, in which the selection decision is made by a system with no consumer preference inertia and no cognitive switching cost, is absent from the two-sided market, CBDC, and platform layering literatures. This paper addresses that gap directly.

The analytical method in this paper is a structured friction decomposition applied to each of the five rail architectures under examination. The method proceeds in three stages: friction taxonomy construction, friction profile assignment, and overlap and substitutability assessment.

Stage 1: Friction taxonomy construction. This framework begins by decomposing the end-to-end payment flow into the discrete points at which a transaction can fail, slow, or cost more than a minimum baseline, rather than analyzing rails against a generic list of capabilities. Five friction categories are identified:

1. Account opening friction: the cost and time required for a new participant (whether a merchant, individual, or autonomous agent) to establish a relationship with the rail operator sufficient to initiate or receive a payment.

2. Identity and compliance friction: the cost of KYC, AML screening, sanctions checking, and ongoing compliance monitoring required to legally move funds between parties. This friction is distinct from account opening because it recurs at each transaction for high-value or cross-border flows, not only at onboarding.

3. Settlement speed and certainty friction: the gap between payment initiation and the payee's unconditional access to cleared funds. This encompasses both timing and finality. A payment that settles in two seconds but can be reversed by the sender's bank is structurally different from one that settles in two seconds with cryptographic finality.

4. Custody and self-sovereignty friction: the constraints placed on how and where funds are held between transactions. Custodial arrangements reduce operational burden but introduce counterparty risk and regulatory dependencies. Non-custodial arrangements shift burden to the participant.

5. Interoperability friction: the cost of moving value between different rails, instruments, or jurisdictions. This includes explicit conversion costs (FX spread, bridge fees), integration costs (API development, maintenance), and protocol mismatch costs (different finality models, different dispute resolution systems).

Stage 2: Friction profile assignment. Each rail architecture is assessed against all five friction categories. The assessment is ordinal, with each rail classified as reducing, neutral to, or adding friction relative to the prevailing alternative in its target use case. The prevailing alternative for cross-border B2B payments is correspondent banking. For B2C retail payments, it is the card network. For agentic commerce, the prevailing alternative is the human-in-the-loop card authorization flow, because that is what agents are being designed to replace or extend.

The assessment draws on published architectural documentation, regulatory filings, and academic literature on the mechanics of each system [8, 12, 14, 16, 20]. Where direct documentation of a system's friction profile is unavailable, as is the case for components of Circle's CPN network architecture that have not been publicly specified in detail, the assessment is based on the structural logic of the layer the system occupies and explicitly flagged as inferred rather than documented.

Stage 3: Overlap and substitutability assessment. Two rails are classified as directly substitutable for a given transaction archetype if and only if: (a) they reduce friction in the same category for the same transaction type, (b) the friction reduction is of comparable magnitude, and (c) the regulatory and technical prerequisites for both rails are met by the same set of participants. The third condition is critical: a rail that requires a regulated financial institution as a gateway node is not substitutable, for a small merchant, with a rail that requires only a software wallet, even if both produce the same settlement finality.

Substitutability is therefore assessed at the level of specific transaction archetypes (B2B cross-border vendor payment, B2C retail purchase, B2B supply chain automated procurement, and agentic-initiated subscription or microtransaction) rather than at the level of the rails in the abstract. This specificity is required because the same rail can be complementary in one archetype and substitutable in another.

Assumptions and scope. This analysis treats the regulatory status of each system as it stands at the time of writing, with MiCAR operative in the EU and the GENIUS Act pending in the US, and does not project future regulatory outcomes. It treats the fraud and consumer protection constraints of irrevocable stablecoin settlement as fixed costs of those rails rather than deficiencies to be corrected, because no current framework resolves them. It does not model adoption rates or switching costs, as the empirical data required for that modeling is not publicly available at transaction-level resolution.

The friction profiles below are organized by friction category. Table 1 (described narratively) summarizes the ordinal assessments; the following subsections develop each profile in detail.

Table 1: Ordinal Friction Profiles by Rail Architecture

| Friction Category | PYUSD | Circle CPN | Bridge/Stripe | Tempo | Card ACP | |---|---|---|---|---|---| | Account opening | Reduces (for PayPal users) | Neutral (FI-to-FI only) | Reduces (API-first) | Reduces (corridor-specific) | Neutral (card-dependent) | | Identity / KYC | Neutral (delegates to PayPal) | Reduces (shared KYC layer) | Reduces (unified compliance API) | Reduces (licensed corridors) | Neutral (delegates to issuer) | | Settlement speed | Reduces (blockchain finality) | Reduces (near-real-time) | Reduces (instrument-agnostic) | Reduces (corridor-specific) | Neutral (T+0 auth, T+1/2 settle) | | Custody / finality | Mixed (custodial wrapper) | Reduces (institutional) | Mixed (orchestration abstraction) | Reduces (sender corridor) | Adds (chargeback-reversible) | | Interoperability | Adds (chain-specific) | Reduces (network layer) | Reduces (multi-instrument) | Adds (corridor-specific) | Neutral (universal acceptance) |

PYUSD: Instrument Layer

PYUSD operates on Ethereum and Solana, with PayPal providing the custodial wrapper and compliance anchor. For PayPal's existing account holders (a substantial installed base), account opening friction is eliminated: the PYUSD balance is a feature of an existing account, with no new relationship required. For new participants with no PayPal relationship, the friction profile is structurally identical to opening any PayPal account, including KYC requirements.

Settlement on-chain is fast and produces cryptographic finality, a material improvement over card settlement, which is authorized at point of sale but settled one to two days later. Card scheme chargeback windows vary by scheme and reason code: card-present transactions are typically disputable within 60 days under most scheme rules, while card-not-present transactions carry longer windows, up to 120 days under several reason codes. In both cases, the authorization-to-final-settlement gap is measured in the duration of the dispute window, during which the payment is conditionally reversible. Stablecoin settlement, by contrast, is irreversible at block confirmation. However, PYUSD inherits chain-specific interoperability constraints: a PYUSD balance on Solana cannot directly interact with a system expecting ERC-20 USDC without a bridge operation, which introduces its own latency and counterparty exposure.

The custody model is custodial by default for retail users, which reduces operational burden but introduces PayPal as a counterparty risk and regulatory dependency. Card payments are also custodial, so this is not a deficiency relative to that alternative; it does mean PYUSD does not reduce custody friction relative to the card alternative. It relocates custody, with different counterparty properties.

In B2B agentic flows, PYUSD's primary friction advantage is settlement finality. Its primary friction disadvantage is the requirement that counterparties also hold PayPal accounts or PYUSD-compatible wallets. This interoperability constraint limits PYUSD's substitutability for card payments in any flow where the payee does not already have compatible infrastructure.

Circle CPN: Network Layer

Circle's CPN is designed as a connectivity protocol for regulated financial institutions transacting in USDC. Its friction target is explicitly the interoperability gap between institutions: a bank in Brazil and a payment processor in Europe can, in principle, route a USDC settlement through CPN without establishing a bilateral correspondent banking relationship. The KYC delegation model in CPN is significant, as each institutional participant is responsible for KYC on its own customers, while CPN provides shared data infrastructure that reduces duplicate compliance work at the network level.

For institutional participants that are already regulated, CPN meaningfully reduces interoperability and identity friction at the network level. For smaller merchants or individuals, CPN is a product accessed through an institutional intermediary rather than directly. This means CPN does not reduce account opening friction for the tail of participants; it shifts that friction to the institutional gateway.

CPN's settlement and custody model is the strongest among the five systems for institutional B2B flows: near-real-time USDC settlement between regulated institutions, with the finality properties of the underlying blockchain, and compliance anchored by each institution's regulatory status. The interoperability advantage is real and meaningful, but it is conditional on the counterparty network density. CPN's value increases as the number of participating institutions grows, following a standard network effect dynamic [2].

Bridge/Stripe: Orchestration Layer

Bridge, as integrated into Stripe's infrastructure, operates above the instrument and network layers. A developer building on Stripe's API does not specify whether a given transaction routes over USDC on Solana, USDC on Ethereum, or a fiat stablecoin equivalent; Stripe's routing logic makes that determination based on corridor availability, cost, and settlement speed. This abstraction is the core value proposition: account opening friction for the merchant is zero incremental work beyond existing Stripe integration, KYC is handled by Stripe's existing compliance stack, and settlement speed is determined by the optimal rail selected at routing time.

The orchestration layer captures value precisely because it abstracts instrument selection. A merchant using Bridge/Stripe does not benefit directly from USDC price discovery or PYUSD liquidity depth, as those are routing inputs that Stripe optimizes on the merchant's behalf. The merchant's observable outcome is settlement timing and cost. This positions Stripe as the entity with the most durable revenue model among the five systems, because its value does not depend on any single instrument remaining dominant [1].

The custody model at the orchestration layer is mixed: Bridge holds stablecoin positions in transit, creating transient custody exposure, but the merchant-facing settlement is typically in fiat, so the merchant does not bear stablecoin custody risk in most configurations. The finality properties depend on the underlying rail selected.

Tempo: Cross-Border Corridor

Tempo addresses the specific friction profile of cross-border payment corridors where correspondent banking costs are high and FX liquidity is thin. Its friction reductions are material in the corridors it serves: account opening for senders is straightforward, settlement speed is competitive with wire transfers, and the stablecoin denomination eliminates the FX conversion at the midpoint of the corridor (though conversion still occurs at the receiving end). Identity and compliance friction is reduced because Tempo operates as a licensed money services business in its corridors, centralizing the compliance function.

Tempo's interoperability constraint is the inverse of CPN's: it is deep within specific corridors and thin outside them. A merchant with diversified international vendor relationships cannot use Tempo as a general-purpose B2B rail the way it could use Bridge/Stripe. Tempo is substitutable for correspondent banking in its corridors and not substitutable for anything outside them.

Card-led ACP: Authorization Layer

Card-led Agent Commerce Protocols address the authorization event specifically in agentic commerce: the moment at which a software agent requests the right to spend funds on behalf of a human or a business entity. The friction card ACP reduces is the credential management friction, specifically ensuring that an agent's spending is bounded, auditable, and revocable without requiring human approval on each individual transaction.

Card ACP does not reduce settlement friction; card settlement timelines and chargeback exposure are unchanged. It does not reduce account opening friction; the underlying card account must exist. Its friction reduction is narrow but important: it is the only architecture among the five that provides a native chargeback-backed dispute resolution mechanism for agentic transactions. In B2C agentic flows (where a consumer's agent is purchasing on their behalf and the consumer retains chargeback rights), this is a regulatory and commercial precondition for consumer trust, rather than a marginal feature [17].

The friction profiles in Section 5 establish that the five rail architectures do not compete uniformly. They share transaction volume only where they reduce friction in the same category for the same transaction archetype. The analytical task in this section is to draw those boundaries precisely and identify the mechanisms that govern them.

B2B cross-border vendor payment: the substitution case is real

In a B2B cross-border vendor payment (a manufacturer in one jurisdiction paying a supplier in another), the prevailing alternative is a SWIFT wire through correspondent banking chains. The friction profile of that alternative includes: multi-day settlement, multiple correspondent bank fees, FX conversion at opaque spreads, and KYC re-verification at each intermediary. All five stablecoin-adjacent systems address at least one of these frictions. CPN and Bridge/Stripe address most of them simultaneously.

In this archetype, CPN and Bridge/Stripe are substitutable for each other, but only for the subset of participants that qualify for both: specifically, businesses that are either regulated financial institutions (and thus have direct CPN access) or are already Stripe-integrated merchants (and thus have Bridge access through the orchestration layer). A corporate treasury function at a bank can access CPN directly. A mid-market manufacturer cannot; it reaches CPN, if at all, through an intermediary institution, whereas it can reach Bridge through a Stripe API integration. For this mid-market participant population, CPN and Bridge/Stripe are not substitutable; they occupy different points in the access hierarchy. For the institutional participant that qualifies for both, the competitive tension between them is a layer-level negotiation over where value is captured: Circle captures it if CPN pricing is visible to the merchant, Stripe captures it if Bridge abstracts CPN as one of many routing options at Stripe's margin.

Card ACP is not substitutable in this archetype. No consumer chargeback expectation exists in a B2B vendor payment, and the authorization-layer value of card credentials does not apply to a wire-equivalent transaction. Card ACP is absent from this competitive space entirely.

B2C agentic retail purchase: the substitution case does not hold

In a B2C agentic retail purchase (a consumer's software agent purchasing a product on the consumer's behalf), the consumer retains chargeback rights under card network rules, and those rights are a consumer protection requirement under several regulatory frameworks. The finality properties of stablecoin settlement, which are an advantage in B2B flows, become a liability in B2C flows because irrevocable settlement eliminates the chargeback mechanism that card networks provide.

This is a structural constraint, not a preference difference. A stablecoin rail that settles irreversibly cannot offer chargeback protection without the issuer backstopping reversals from its own balance sheet, which transforms the economic model. The practical result is that card-led ACP retains the authorization layer in B2C agentic flows because regulatory and commercial requirements make stablecoin settlement structurally inadequate for that use case without a separate dispute resolution overlay [14].

PYUSD is partially excepted from this analysis: PayPal's existing consumer protection policies, which predate PYUSD, provide some dispute resolution backstop for PayPal account holders. This is a function of PayPal's platform policies, not of the PYUSD protocol itself, and it depends on PayPal's continued willingness to absorb that liability.

The sequential dependency problem

A critical structural finding is that several of the friction reductions offered by these systems are sequentially dependent: one friction cannot be reduced until a preceding friction is already solved. CPN's interoperability advantage, for instance, is only available to participants that have already cleared the account opening and KYC frictions required for institutional membership. A long-tail merchant cannot access CPN's settlement efficiency without first establishing a relationship with a CPN member institution that carries the account opening and compliance burden on its behalf. Bridge/Stripe's routing abstraction is similarly dependent on the merchant having completed Stripe's onboarding process. These dependency chains mean that the friction map is layered in a specific sequence, and a rail that solves a downstream friction (settlement speed) cannot be adopted without the upstream frictions (account opening, KYC) being resolved first.

This sequential structure has a direct competitive implication: the system that controls the upstream friction controls access to all downstream friction reductions. Stripe's account opening and compliance onboarding is the gateway through which Bridge's stablecoin routing becomes accessible to the majority of merchants. PayPal's consumer account is the gateway through which PYUSD becomes accessible to retail users. The instrument- and network-layer systems (PYUSD, CPN, Tempo) are therefore partially dependent on orchestration- and interface-layer systems for distribution, even when their technical friction reduction is superior in isolation.

Agentic B2B procurement: where substitution is structurally enabled

The transaction archetype where substitution is most structurally enabled is automated B2B procurement in an agentic supply chain: a software agent purchasing inventory, services, or compute resources on behalf of a business, in a context where the business is the legal principal and no consumer protection regulation applies. In this archetype, the agent can evaluate stablecoin rails programmatically against card networks on settlement cost, settlement speed, finality certainty, and corridor availability, and select accordingly. This programmatic selection logic has no direct precedent in the platform infrastructure literature [4], which addresses human-developer API adoption rather than autonomous agent rail selection, but the structural implication follows from the absence of consumer preference inertia: an agent carries no habitual payment method, only an optimization function.

For this archetype, Bridge/Stripe's orchestration layer is the enabling infrastructure: it gives the agent a single API surface from which to access multiple rails without requiring the agent to carry per-rail authentication credentials. PYUSD, CPN, and Tempo all become routing options within that API rather than competing products the agent must evaluate directly. Card ACP in this context represents the fallback, the rail the agent uses when stablecoin routing is unavailable, the counterparty is not stablecoin-capable, or the transaction amount is below the threshold where settlement cost optimization matters.

The regulatory bifurcation and its competitive limits

MiCAR's restructuring of the EU stablecoin market has created a bifurcated competitive landscape in which compliant operators (Circle, Stripe/Bridge, and Tempo where it operates under licensed corridors) hold a near-term advantage that is regulatory in origin rather than technical [20]. Liebowitz and Margolis [2] demonstrate, in the context of software markets, that network externalities are difficult to document empirically, and that scale advantages which appear self-reinforcing often prove fragile when switching costs are integration-based rather than coordination-based. The analogy to stablecoin rail competition is structural but imperfect: software market network effects operate through user-side coordination, while stablecoin rail switching involves replacing a technical API integration and associated compliance workflows. Because the switching cost for a merchant moving between compliant stablecoin rails is primarily the engineering and compliance review cost of re-integration, rather than the cost of persuading a counterparty network to move simultaneously, the early-mover advantage for compliant rails does not compound in the way that card network acceptance does.

This lock-in fragility benefits entrants and constrains any single instrument-layer operator's margin durability, except at the orchestration layer, where the merchant's integration point is the API rather than the rail. A merchant integrated with Stripe does not re-integrate when Stripe changes its routing layer; it simply receives different settlement economics. This asymmetry between orchestration-layer stickiness and instrument-layer replaceability is the mechanism by which Bridge/Stripe is positioned to capture disproportionate margin as the underlying instruments commoditize.

The core finding of this analysis is that the five rail architectures under examination (PYUSD, Circle CPN, Bridge/Stripe, Tempo, and card-led ACP) do not compete for the same payment volume across all transaction archetypes. They compete only where they reduce friction in the same category for the same type of transaction. Mapping those intersections precisely reveals a competitive architecture that is friction-specific rather than rail-versus-rail.

For merchants and platform operators, this finding shapes integration and investment decisions along three specific vectors.

The first vector is transaction archetype classification. A merchant processing both B2C retail transactions and B2B supply chain procurement faces different competitive landscapes in each. The B2C retail flow preserves card network primacy at the authorization layer because consumer protection requirements create a structural dependency on chargeback-backed dispute resolution. No current stablecoin rail resolves this without an overlay that replicates card network liability, at which point the settlement cost advantage is substantially eroded. The B2B supply chain flow carries no such dependency. Stablecoin rails reduce settlement cost and timing friction in ways card networks cannot match without architectural changes: the two-day authorization-to-finality gap in card settlement becomes a direct cost in working capital terms for high-frequency B2B procurement, and the correspondent banking fee structure in cross-border corridors adds basis points that stablecoin corridors eliminate at the midpoint. For a merchant routing a material share of procurement spend through international suppliers, the arithmetic of settlement cost reduction in the B2B layer is separable from the consumer protection logic governing the B2C layer and should be evaluated independently.

The second vector is integration point selection. Merchants that integrate at the orchestration layer (specifically through Stripe/Bridge) receive instrument-agnostic routing and absorb none of the complexity of per-rail API management. The cost of this convenience is that the orchestration layer captures the routing margin and the merchant observes only the net settlement outcome. Merchants with sufficient transaction volume and technical capacity to manage per-rail integrations can, in principle, access instrument-layer cost directly: using CPN membership for institutional cross-border flows, PYUSD for PayPal-ecosystem counterparties, and Tempo for specific corridors. Whether this cost-benefit calculus favors direct integration depends on three specific inputs: the merchant's total volume in the relevant corridors, the concentration of that volume in specific counterparty ecosystems, and the internal compliance capacity required to operate as a direct participant rather than a routing customer. Below the threshold where those inputs justify the overhead, orchestration-layer integration is the cost-minimizing choice.

The third vector is regulatory positioning. The practical competitive filter for stablecoin rails in 2025 and into 2026 is regulatory status under MiCAR in the EU and investment-company classification in the US. A rail operator that does not achieve compliant status in both jurisdictions cannot serve the institutional and high-volume merchant segments where stablecoin settlement is most economically attractive. This constraint does not operate symmetrically across the five architectures. Circle's CPN, operating at the network layer between regulated institutions, is structurally aligned with MiCAR's e-money token framework because its participants are themselves regulated. Bridge/Stripe's orchestration-layer model places compliance responsibility at the API gateway, which Stripe's existing regulatory relationships support in the jurisdictions where it already operates. Tempo's licensed corridor model achieves compliance within corridors but must replicate that licensing in each new corridor, making geographic expansion a compliance project before it is a technical one. PYUSD's compliance posture is anchored to PayPal's existing regulatory status, which provides a stable foundation in jurisdictions where PayPal is licensed but limits expansion to the footprint of that license. Card ACP faces no novel regulatory classification question, given that the card schemes already operate under established regulatory frameworks, and it also does not benefit from the settlement cost economics that make stablecoin rails attractive in the B2B context.

For rail operators specifically, the strategic implication follows the layering logic directly. Instrument-layer issuers face the greatest commoditization pressure: USDC and PYUSD are differentiated by distribution relationships and compliance wrappers, rather than by the stablecoin mechanism itself, which is structurally similar across compliant operators. A second compliant issuer with equivalent reserve quality and distribution reach reduces the first mover's pricing power at the instrument layer immediately, because the switching cost (re-pointing API calls to a new contract address) is low relative to the settlement savings achievable. Network-layer plays like CPN face an adoption speed constraint: their value scales with participating institution count, and that count grows at institutional sales cycle speed, which is measured in quarters, rather than weeks. Orchestration-layer operators face the most durable position but also the most regulatory scrutiny, because controlling routing decisions across multiple stablecoin instruments raises the same concerns about payment infrastructure concentration that have historically attracted regulatory attention to card networks [1]. The parallel is structural: card networks achieved routing control over merchant-acquirer relationships and faced regulatory and antitrust examination as a result; an orchestration-layer operator that achieves equivalent routing control over stablecoin flows will face the same examination, on a timeline governed by the volume threshold at which that control becomes systemically significant. Authorization-layer card ACP retains structural primacy in B2C flows for as long as consumer protection regulation requires chargeback-backed dispute resolution, and no current revision to that requirement is underway in any major jurisdiction.

This analysis operates at the level of architectural mechanism and structural logic. It does not, and cannot, measure the following.

Actual adoption curves. The friction profile of a rail and the rate at which merchants and agents adopt that rail are different questions. A rail may reduce settlement friction substantially while facing adoption barriers rooted in counterparty network density, developer tooling quality, or institutional sales cycles. CPN's value, as noted, scales with participating institution count, but this analysis cannot measure whether that count grows at a rate sufficient to produce the claimed interoperability advantage within any specific timeframe.

User switching costs. The analysis identifies integration cost as the primary switching cost for merchants moving between rails. It does not quantify those costs, because the engineering hours, compliance review effort, and testing cycles required vary materially by merchant size, technical architecture, and existing integration depth. A merchant with a legacy ERP system faces fundamentally different switching costs than a natively cloud-architected platform. These differences affect whether the substitutability identified in B2B agentic flows translates into actual volume migration.

Regulatory trajectory. The analysis treats MiCAR as operative and the GENIUS Act as pending, and does not project outcomes. The investment-company classification question for stablecoin issuers under US law is unresolved, and its resolution could materially alter the competitive landscape, either by eliminating operators that cannot meet registration requirements or by clearing a path for new compliant entrants that currently sit outside the analysis. The analysis therefore provides a snapshot competitive architecture, not a durable forecast.

Opaque friction components. Several friction components are not publicly documented at sufficient detail to assess precisely. Circle CPN's shared KYC data architecture, for instance, has been described in commercial terms but not in the technical detail required to assess whether it genuinely reduces identity friction at the transaction level or merely at the institutional onboarding level. Bridge's routing logic (which determines when it selects stablecoin settlement versus fiat settlement) is commercially proprietary. These opaque components mean that some friction profile assessments are inferred from structural logic rather than documented mechanism.

Agentic commerce volume baseline. The analysis identifies B2B agentic procurement as the archetype where stablecoin substitution of card settlement is most structurally enabled. However, the current volume of genuinely autonomous agent-initiated B2B procurement transactions is not measurable from public data. The substitution hypothesis is structurally sound but empirically untestable without access to transaction-level data from platform operators.

The layering hypothesis advanced in this paper (that stablecoin rails and card-led ACP occupy different friction positions and compete directly only in B2B agentic flows without consumer protection requirements) is structurally derived but empirically unvalidated. Three concrete research directions would either confirm or falsify it.

Transaction-level corridor analysis. If Bridge/Stripe and CPN produce routing logs that distinguish between stablecoin-settled and card-settled B2B transactions by corridor, those logs would allow direct measurement of whether stablecoin settlement share is increasing in the B2B agentic archetypes identified in Section 6. A 36-month measurement window covering the period from material agent-commerce deployment to steady-state would be sufficient to test whether card settlement share declines in those specific archetypes as the hypothesis predicts.

Friction weighting under regulatory change. The GENIUS Act's investment-company classification outcome will alter the friction profile of PYUSD and potentially of Tempo, depending on how each is classified. A pre/post regulatory event study, comparing merchant adoption rates and switching behavior before and after classification, would provide direct evidence of whether regulatory compliance functions as the primary competitive filter this analysis identifies, or whether technical differentiation or distribution relationships dominate instead.

Agent rail selection behavior. As agentic commerce platforms accumulate transaction histories, analysis of agent rail selection decisions, controlling for corridor, transaction size, and counterparty capability, would reveal whether agents in practice select stablecoin rails at the rates predicted by their friction profiles, or whether platform-level defaults and integration inertia dominate. This behavioral measurement is the most direct test of the paper's central claim about agent-driven substitution. The platform infrastructure literature [4] addresses developer adoption of API boundary resources but does not model autonomous agent selection among competing rails; the empirical design would need to distinguish between decisions made by human developers configuring agent defaults and decisions made dynamically by the agent at transaction time. That methodological distinction requires access to execution-level logs rather than configuration-level data.