Visa is moving to position itself at the center of machine-led commerce with the launch of Intelligent Commerce Connect, a payments platform built to let autonomous AI agents discover products and initiate purchases on behalf of consumers and businesses. The product is designed as a bridge between conventional payment rails and agent-driven transaction flows, giving merchants and developers a way to plug AI commerce into an existing global network without exposing raw payment credentials.
The company introduced the platform in early April and said it has already entered a pilot phase with selected partners ahead of a broader rollout later in 2026. Visa is presenting the system as infrastructure for what it calls a new B2AI model, combining tokenization, spend controls and network-level fraud protections in a single integration. The strategic ambition is larger than a product launch: Visa wants to define the payment layer for autonomous commerce before that market fully takes shape.
A payment orchestration layer for AI agents
At its core, Intelligent Commerce Connect is structured as an orchestration layer that abstracts payment initiation, identity checks and risk controls for agentic transactions. Instead of forcing merchants or developers to stitch together separate systems for credentials, compliance and fraud prevention, the platform centralizes those functions inside Visa’s existing acceptance stack. That architecture is meant to reduce engineering complexity while giving AI agents tightly bounded authority to act.
Tokenization sits at the center of the model. Visa said digital tokens replace raw card credentials and can carry the identity and authorization logic needed to confirm that an AI agent is acting on behalf of a consumer. That allows an agent to initiate a payment without directly handling sensitive card data, while policy rules can limit what it is allowed to buy, from which merchants, and for how much. The system is built to make delegated payments possible without turning delegation into a security liability.
The platform also extends beyond simple payment initiation. Visa has incorporated real-time controls that can restrict merchant categories, transaction types and spending amounts, while applying machine-learning fraud models before a transaction is completed. At the same time, it is making merchant catalogs more queryable through standardized feeds and APIs so that product discovery can happen natively inside agent workflows. The goal is not only to let AI agents pay, but to let them shop within a controlled and searchable commercial environment.
Interoperability will determine whether the model scales
A major part of the platform’s pitch is protocol flexibility. Visa said Intelligent Commerce Connect supports several major agent protocols, including Trusted Agent Protocol, Machine Payments Protocol, Agentic Commerce Protocol and Universal Commerce Protocol, and can process both Visa and non-Visa credentials through integrated APIs. That protocol-agnostic approach is essential if Visa wants to become infrastructure for the broader agent economy rather than for a single closed ecosystem.
The company has also assembled a broad early network around the initiative. It said the pilot includes partners such as Aldar, AWS, Diddo, Highnote, Mesh, Payabli and Sumvin, while also pointing to relationships with AI platform providers including Anthropic, IBM, Microsoft, Mistral AI, OpenAI, Perplexity and Samsung. Visa additionally highlighted a technical collaboration with Stripe around AI-ready payment credentials. Those relationships suggest the company is trying to seed adoption simultaneously across payments, commerce and AI development layers.
The business case is straightforward. A single integration reduces the engineering burden for merchants that want to accept agent-initiated transactions, while centralized PCI orchestration and policy enforcement lower the operational risk of opening payment flows to software agents. For third-party enablers building agentic commerce tools on behalf of merchants, Visa is also offering orchestration and compliance support. What Visa is really selling is managed trust in a category where trust will be the hardest thing to scale.
The real test, however, will come in live deployment. The platform’s long-term relevance will depend on whether it can maintain low integration friction, enforce spend policies reliably across different agent frameworks and apply fraud controls without adding too much latency or operational overhead. If the pilot proves that autonomous payments can be both interoperable and tightly governed, Visa could secure a foundational role in the next phase of digital commerce.
