AWS Enters Agentic Payments, Building a 'Trust Layer' with Visa

AWS Unveils New Strategy Focused on Payment Execution Layer

AWS leaps into agentic payments

Amazon Web Services, the tech giant's cloud computing network, is gearing up for armies of AI-powered agents and transactions.

Amazon Web Services (AWS) is making a full-scale move into building infrastructure for autonomous payment processing by AI agents. According to Payments Dive, John Kain, AWS's Head of Global Financial Services Market Development, stated that "the real purpose of agentic is to have agents that act and execute on behalf of users."

What stands out is that AWS is focusing specifically on the "payment execution layer" rather than "commerce as a whole." Kain emphasized that "commerce is certainly interesting, but fundamentally, the most important layer is being able to execute secure payments at scale in a way that enables the entire agentic commerce ecosystem."

Industry Context

In realizing agentic commerce -- a system where AI agents autonomously search, compare, and purchase products on behalf of users -- "payments" has been the biggest bottleneck. Even when AI agents could find products, there was no standardized way to securely complete payments.

The industry has been moving rapidly to address this challenge. Stripe announced its "Agentic Commerce Suite", introducing Shared Payment Tokens (SPT) as a new payment primitive. Mastercard has also deployed "Agent Pay", and Visa has built an authentication framework through its "Trusted Agent Protocol."

With the largest cloud infrastructure provider entering this space, there is now potential for integrating the efforts of various payment networks. AWS offers customers the freedom to choose from various AI model providers including OpenAI, Anthropic, and AWS's own models, with a "network-agnostic" approach that avoids vendor lock-in.

Visa Intelligent Commerce and the Full Blueprint Picture

At the core of the AWS-Visa partnership is the integration of "Visa Intelligent Commerce" and "Amazon Bedrock AgentCore." Announced in a December 2025 press release, this initiative provides four foundational technologies.

Tokenization: Issues and manages payment tokens specifically for AI agents, enabling transactions without directly exposing credit card information.

Authentication: Identity verification features including passkeys and device binding ensure that agents are legitimate representatives of users.

Personalization: Enables transactions based on user preferences and past purchase history.

Intent Capture: Accurately captures user purchase intent and controls the scope of agent actions.

Technically, Amazon Bedrock AgentCore Runtime provides a serverless hosting environment, and through the Visa Intelligent Commerce MCP server, securely propagates user IDs, consent states, and authenticated payment credentials throughout the workflow. Amazon Bedrock AgentCore Gateway allows agents to access tools for flight search, hotel search, product search, and payment processing in an auditable manner.

Furthermore, the open blueprints published on AWS Marketplace are designed as ready-to-use implementation templates.

• Travel Agent: Three specialized agents -- supervisor agent, travel assistant, and cart manager -- work in coordination
• Retail Shopping Agent: Covers everything from product discovery and price comparison to shipping, discounts, cart management, loyalty, checkout payments, order tracking, and returns
• B2B Payment Agent: Supplier payments and reconciliation processing integrated with ERP, CRM, accounting, and banking systems

Visa's SVP of Global Growth Rubail Birwadker stated that "agentic commerce requires trust to turn intent into action. Visa Intelligent Commerce is the trust layer for the agent economy."

Impact and Action Items for E-commerce Businesses

Previously, adopting agentic payments required individual integrations with each payment network and building proprietary authentication infrastructure. With AWS's blueprints, e-commerce businesses can now build standardized workflows immediately through Amazon Bedrock AgentCore.

Using the retail shopping agent blueprint, a customer can simply instruct an AI agent to "find running shoes within a 10,000 yen budget that will arrive next week," and the entire process from product search to payment completion is handled end-to-end. The user never needs to interact with a checkout page.

On the other hand, as Digital Commerce 360 points out, delegating payment authority to AI agents creates risks for new fraud patterns. Countermeasures against threats different from conventional ones -- such as identity spoofing, merchant impersonation, and large-scale automation errors -- are needed. E-commerce businesses must simultaneously build governance frameworks including agent transaction limits and bot identification through the Visa Trusted Agent Protocol.

Currently the integration is Visa-centric, but "support for multi-network commerce flows is coming soon" has been announced. Stripe is similarly integrating tokens from Visa, Mastercard, Affirm, and Klarna, and the cross-network standardization competition is accelerating.

Summary

AWS's entry into agentic payments symbolizes this field's transition from the "experimental stage" to the "infrastructure development stage." With the largest cloud provider partnering with Visa to standardize the payment execution layer, the barriers to practical agentic commerce implementation have been significantly lowered.

There are three key points e-commerce businesses should focus on. First, early technical validation using Amazon Bedrock AgentCore blueprints. Second, monitoring the landscape of competing agentic payment standards including Stripe SPT, Mastercard Agent Pay, and Visa Intelligent Commerce. Third, establishing security policies and governance frameworks for delegating payment authority to AI agents.

Major companies such as Expedia Group, Intuit, and lastminute.com are already participating in blueprint design, and it is important to prepare ahead of the second half of 2026 when implementation examples are expected to increase.