Agentic Commerce
What Is Agentic Commerce?
Agentic commerce describes commercial transactions initiated and executed by autonomous AI agents acting on behalf of a human principal—without requiring real-time human confirmation at the point of purchase. An AI travel assistant that books flights and hotels; a procurement AI that replenishes inventory against pre-approved supplier contracts; a personal finance agent that switches between subscription plans to minimise cost—these are all instances of agentic commerce.
The defining property is the separation between the authority to transact and the act of transacting. The human grants authority in advance—defining spending limits, permitted merchant categories, approval thresholds, and validity windows—and the agent executes within that authority without interrupting the user for per-transaction confirmation. This model radically improves the efficiency of commercial processes that are currently human-bottlenecked but follow predictable, rule-governed patterns.
Why Agentic Commerce Requires New Infrastructure
The Human-Not-Present (HNP) Problem
Existing payment infrastructure is designed for human-present, human-initiated transactions. Strong Customer Authentication (SCA) assumes a human who can respond to an OTP or biometric challenge. Fraud detection models are calibrated against human behavioural patterns. Chargeback dispute frameworks assume a human cardholder who can assert non-authorisation. Agentic commerce breaks every one of these assumptions: the agent cannot respond to a 3DS2 challenge, its transaction velocity is systematically different from human behaviour, and the delegation chain (human → agent → transaction) must be cryptographically verifiable.
Delegation Credentials and Network-Level Trust
For an agentic payment to be trusted by the issuing bank, card network, and fraud system, the transaction must carry verifiable evidence of delegation: who authorised the agent, with what parameters, and within what time window. This requires a tokenized identity layer that binds payment credentials to agent-specific authorisation tokens. Visa Intelligent Commerce (VIC) defines the credential and signal requirements for verifying agent-initiated transactions at network level.
Machine-Speed Velocity and Infrastructure-Level Controls
AI agents can initiate transactions at machine speed across multiple merchants simultaneously. Risk systems must handle this volume while distinguishing authorised high-frequency agent activity from fraud. Spending limits and merchant category restrictions must be enforced at the payment infrastructure level—not by the agent itself—to prevent a compromised or malfunctioning agent from exceeding its authorised scope.
How Hellgate Enables Secure Agentic Commerce
Hellgate's Composable Payment Architecture (cpa) provides native support for agentic commerce through its Human-Not-Present (HNP) architecture. Hub's flow engine handles HNP flows as a distinct transaction class: 3DS step-up challenges are replaced by agent identity token validation; spending limits and merchant category restrictions are enforced at the routing layer as hard constraints; every agentic transaction is logged with the full delegation chain for audit and dispute purposes.
Guardian manages the machine-readable payment credentials and identity delegation tokens that agents carry. Specter evaluates agentic transactions against risk rules calibrated for machine-initiated patterns, with Visa Intelligent Commerce (VIC) signal integration providing network-level confidence. Pulse provides the complete, immutable audit trail of every agent-initiated transaction—the authorisation chain, the parameters in effect at execution, and the outcome—supporting both operational monitoring and DORA-compliant regulatory reporting.
