Payment Methods & Rails

Agentic Commerce

What Is Agentic Commerce? Definition and How It Works

Definition

Agentic commerce is a model of online commerce in which autonomous AI agents, acting under delegated authority from a human user, discover, evaluate, and complete purchases on the user's behalf without the user directly navigating a checkout interface. The agent interprets the user's intent, queries merchant systems, selects a product or service, and executes the transaction using scoped payment credentials.

How it works

In agentic commerce, a human user provides high-level intent to an AI agent: for example, "book the cheapest nonstop flight landing before noon on Friday" or "reorder my usual office supplies when stock falls below five units." The agent then operates autonomously across discovery, decision, and checkout.

Discovery relies on machine-readable data structures. Merchants that expose structured product catalogs, pricing, inventory, and policy data via APIs or Model Context Protocol (MCP) servers become accessible to agents. Merchants without structured data are effectively invisible to agent-driven traffic, in the same way an unindexed website is invisible to search engines.

Payment execution requires new credential protocols because agents cannot type card numbers or click through a checkout page. Agent payment protocols, including OpenAI's Agentic Commerce Protocol (ACP), Mastercard's Agent Pay, and Visa's Intelligent Commerce for cards, and Google's payment-agnostic Agent Payments Protocol (AP2), issue scoped tokens or cryptographically signed mandates that bind the user's consent to a specific merchant, amount, and session. Stablecoin rails use on-chain wallet signatures to achieve the same scoping without card networks.

Reconciliation in agentic commerce attaches agent identity and mandate evidence to each transaction record, enabling merchants and issuers to attribute transactions back to the originating agent, user, and consent event, which is essential for dispute resolution when a human was not present at checkout.

Why it matters

Agentic commerce shifts the fundamental unit of merchant competition from page-level UX to data quality and checkout infrastructure. Merchants optimised for human browsers, with marketing copy, imagery, and frictionful checkout flows, are not automatically optimised for agents. Agents prioritise structured, machine-readable data: clean product schemas, reliable inventory signals, unambiguous pricing, and fast API responses.

Payment infrastructure must evolve in parallel. A checkout flow that requires human authentication steps, CAPTCHAs, or browser-rendered pages will fail for agent-initiated transactions. Merchants need payment stacks that support programmatic authorisation, agent-compatible tokenisation, and audit trails that satisfy card scheme chargeback rules in the absence of a human at checkout.

Fraud and risk models also require recalibration. Traditional fraud signals, browser fingerprint, session behaviour, device type, are absent or meaningless for agent-generated requests. New signals, such as agent registry verification, mandate integrity checks, and spend-scope validation, are being built into card network infrastructure and must be incorporated into risk engines.

With PXP

PXP supports merchants and partners across the payments value chain. To talk through agentic commerce as part of your payment strategy, get in touch with our team.

Talk to a payments specialist

Frequently asked questions

How is agentic commerce different from regular e-commerce?

In standard e-commerce, a human navigates the merchant's interface, selects products, and completes checkout manually. In agentic commerce, an AI agent performs all of those steps autonomously, using structured data and programmatic payment credentials rather than a browser and a typed card number. The merchant experience is similar; the buyer-side interface is entirely different.

What payment protocols power agentic commerce?

The main agent payment protocols are OpenAI's Agentic Commerce Protocol (ACP, developed with Stripe), Mastercard Agent Pay, and Visa Intelligent Commerce for cards, alongside Google's payment-agnostic Agent Payments Protocol (AP2). These issue scoped tokens or signed mandates that let agents authorize payments without exposing full card credentials. Stablecoin rails, notably Coinbase's x402, offer an alternative for machine-to-machine transactions.

How do chargebacks work in agentic commerce?

Chargeback handling in agentic commerce relies on mandate evidence rather than traditional authentication records. Under card-rail protocols, the merchant retains a cryptographically signed mandate showing the user's explicit consent to the specific transaction. This mandate serves as dispute evidence in the same way a 3DS authentication record does today. Issuers and networks are building agent-specific data fields into transaction records to support this.

Do merchants need to rebuild their checkout for agentic commerce?

Not entirely, but incremental changes are required. Merchants need to expose structured, machine-readable product and pricing data; ensure their checkout flows support programmatic payment authorisation without human-intervention steps; and adopt tokenisation and mandate-verification capabilities from their payment provider. Merchants already using API-first payment infrastructure are best positioned to enable agentic checkout quickly.