The Control Layer: Why Agentic Commerce Is Not Decided at Checkout
The battle is not fought over payment methods and reach. It is fought in the layer above: identity, authorization, policy, audit and settlement. Each major provider controls only its own system. The neutral control layer is the most valuable unclaimed position in the European market.
What the Control Layer Is
In the standard model of commerce, the decision happens at checkout. A human arrives at a product page, reviews the price, confirms their payment credentials, and clicks buy. The payment network validates the transaction. Governance is implicit in the human presence.
In agentic commerce, the decision happens earlier and in a different layer: the agent evaluates options, selects a preferred outcome, and executes the transaction without the account holder's real-time involvement. The checkout is an automated downstream consequence of a decision already made.
The control layer is the infrastructure that governs this earlier decision: agent identity, authorization scope, spending policy, audit logging, and dispute traceability. Whoever controls this layer controls the effective decision point in agentic commerce, not the checkout UI that processes the result. For the maturity taxonomy of the transaction itself, see The Agentic Commerce Framework.
The agentic commerce layer model. The control layer sits between agent and payment rail. Diagram: contraco original.
Visa Intelligent Commerce: Live Infrastructure
Visa Intelligent Commerce launched in July 2026 with more than 30 European issuers already connected, including lastminute.com, Frasers, Cleverbridge, and BrickDepot. The Trusted Agent Protocol defines how agents are enrolled and authenticated. The Agent Directory provides merchants with a query mechanism to confirm that an agent presenting for payment is a registered, issuer-authorized entity. [Q1, Q2]
What Visa controls in this model: enrollment, directory lookup, and transaction validation within the Visa network. What Visa does not control: the agent's behavior outside that transaction, multi-rail settlement, or authorization policy across competing networks.
Mastercard Agent Pay: Verifiable Intent and Multi-Rail
Mastercard Agent Pay introduces Verifiable Intent: a signed record linking the account holder's authorization to the specific agent and transaction. The dispute and chargeback mechanism is built on top of this signed chain. Programmatic spending limits are set by the issuer at the time of mandate issuance. [Q3]
Mastercard Agent Pay for Machines extends this to multi-rail settlement, meaning a single mandate can authorize transactions across payment networks and settlement systems. [Q4] This is technically significant: it is the first major attempt to build a mandate framework that is not network-bound.
The constraint remains the same as Visa's: Mastercard controls what happens within its own infrastructure. The authorization scope, the agent quality, and the behavior between the mandate and the execution are not addressed by the payment protocol.
Google AP2 and the Signed Mandate
Google's Agentic Payment Protocol version 0.2.0 introduced human-not-present payment execution via signed mandates, with more than 60 partners connected at launch. [Q5, Q6] The signed mandate is the key construct: the account holder issues a mandate to the agent, the agent signs transactions against that mandate, and the payment processor validates the signature.
AP2 is a developer-facing protocol. Its significance is in establishing the technical pattern for mandate-signed agentic payments, independent of any single network. Whether Google's ecosystem controls the dominant share of this pattern or whether it becomes an open standard is the competitive question for the next 18 months.
SAP AI Agent Hub: The ERP Integration Layer
Announced at SAP Sapphire 2026, the SAP AI Agent Hub provides granular permission management and a full audit log for AI agents operating inside enterprise workflows. [Q7, Q8] Its significance is not payment processing but ERP integration: when an agent purchases, the authorization, the transaction record, and the reconciliation entry all flow through the same governed infrastructure.
SAP's position is the back-office end of the control layer. It governs what happens after the payment clears, not before it is initiated. Connecting the pre-purchase authorization layer to the post-purchase reconciliation layer is the integration challenge that no single provider has fully resolved.
The Benchmark: Alipay at Scale
Ant International's Alipay is processing approximately 120 million AI-agent transactions per week. [Q9] This is not a pilot or a proof of concept. It is infrastructure running at consumer scale in a single-operator, single-market context where identity, payment, and the agent layer are controlled by one entity.
The European market is structurally different: multiple payment networks, multiple banking regulators, multiple identity frameworks, and a consumer base that surveys show is more skeptical of single-actor control. The Alipay benchmark demonstrates what is technically achievable. It does not provide the model for the European market.
The Structural Gap: Each Provider Controls Only Its Own System
Visa controls the Visa network. Mastercard controls the Mastercard network. Google controls AP2 within its ecosystem. SAP controls the ERP layer for SAP deployments. Each provider has built or is building a control layer for its own system. None of them controls, or is positioned to control, the layer across all systems.
The neutral control layer, which provides consistent agent identity verification, authorization policy, audit logging, and dispute resolution across payment networks, banking partners, and ERP systems, is the most structurally valuable position in the European agentic commerce market. It is also the most genuinely unclaimed.
For the trust dimension of this gap, see Trust Is the Bottleneck, and Banks Own It. For the attribution consequence, see Dark Agentic Traffic. For the agent accountability question, see Who Audits the Agent? For the market timing argument, see The Window. For the travel vertical where this plays out first, see Europe's First Agentic Purchase Was a Trip. For a dated reference chronicle of the 2026 announcements, see The Loud Summer of 2026. For the dedicated diagnostics property, see agentic.contraco.net.
Sources:
- Q1: Visa Intelligent Commerce press release, July 2026. Trusted Agent Protocol, Agent Directory, 30+ European issuers.
- Q2: Visa Intelligent Commerce technical documentation. Protocol detail.
- Q3: Mastercard Agent Pay press release. Verifiable Intent, programmatic limits, dispute and chargeback mechanism.
- Q4: Mastercard Agent Pay for Machines. Multi-rail settlement detail.
- Q5: Google AP2 v0.2.0 release notes. Human-not-present payments via signed mandates, 60+ partners.
- Q6: Google AP2 developer documentation.
- Q7: SAP AI Agent Hub announcement, Sapphire 2026. Granular permissions, full audit log.
- Q8: SAP AI Agent Hub technical detail.
- Q9: Ant International / Alipay. 120M AI-agent transactions per week.
Strategy Conversation on Agentic Commerce Readiness
Where does your organization sit relative to the control layer? We work with enterprises, financial institutions, and technology providers to map their position and design their path to the neutral infrastructure layer.
Start the strategy conversation