By VisaLearning Objectives
- Understand what agentic commerce is and why AI agents need a new kind of payment credential
- See how tokenization lets an AI agent pay without exposing a consumer's card number
- Evaluate the open questions around fraud liability and consumer consent for agent-driven purchases
What Is Visa Intelligent Commerce?
Visa Intelligent Commerce is an application programming interface (API) suite that lets trusted AI agents make purchases on a consumer's behalf. It bundles four capabilities — tokenization, authentication, agent-payment instructions, and real-time transaction signals — so that a shopping assistant can find a product, confirm a user's intent, and complete checkout securely. The problem it solves is direct: as AI agents move from answering questions to actually buying things, handing an agent a raw 16-digit card number is unsafe. Intelligent Commerce instead issues the agent an "agent-ready" payment token that is scoped, monitored, and revocable.
Visa is one of the world's largest payment networks, tracing its roots to the 1958 BankAmericard and operating today as Visa Inc. (NYSE: V), headquartered in San Francisco after its 2008 public listing. It launched Intelligent Commerce in April 2025 alongside a broad set of AI partners, including Anthropic, OpenAI, Microsoft, and Stripe, deliberately building across competing AI platforms rather than betting on one — much as its card network already runs across competing banks.
💡Key Concept
Tokenized agent payments: Instead of giving an AI agent your actual card number, Visa replaces it with a payment token bound to that specific agent and use case. The token can carry spending limits, be watched in real time, and be switched off instantly. So even if the agent or its data were compromised, the underlying card stays protected.
✅Tip
Visit Visa Intelligent Commerce: usa.visa.com — built for AI platforms, merchants, and issuers integrating agent payments; enterprise partnership, pricing by arrangement.
Core Capabilities
Tokenized credentials for agents
Visa converts a cardholder's credential into a network token that an approved agent can use at checkout. The real card number is never shared with the agent or the merchant, and the token can be scoped to a single merchant, a spending cap, or a time window.
Agent authentication and intent
Before a purchase proceeds, the platform verifies that a legitimate agent is acting and that the consumer actually authorized the transaction. This "intent" layer is central to trusting a machine-initiated payment rather than a human tap.
Real-time transaction signals
Visa applies the same real-time authorization and fraud-monitoring infrastructure it uses across its global network to agent-initiated transactions, watching each one as it happens instead of relying only on after-the-fact review.
Cross-platform reach
Because Intelligent Commerce is delivered as APIs adopted by many AI providers at once, a single integration lets a merchant or issuer reach agents running on several different assistant platforms.
Strengths
- Built on proven rails: It layers agent payments onto Visa's existing tokenization and authorization network rather than inventing an untested settlement system from scratch.
- Card number never exposed: Scoped, revocable tokens mean a compromised agent cannot leak a usable card number.
- Platform-neutral: Adoption across Anthropic, OpenAI, Microsoft, and others reduces the risk of a merchant backing a single losing standard.
- Real-time controls: Spending limits and instant revocation give consumers and issuers levers that a shared card number never offered.
Limitations & Considerations
- Standards are still emerging. Agentic-commerce protocols are new and evolving across the industry; today's integrations may shift as the ecosystem converges on shared rules.
- Fraud liability is unsettled. When an autonomous agent buys the wrong item or is manipulated, who bears the loss — the consumer, the merchant, the platform, or the issuer — is not yet clearly established.
- Consent rules are immature. Proving that a consumer genuinely authorized a given agent purchase, and defining what "authorization" even means for a standing agent, remains an open governance question.
- Requires ecosystem participation. Value depends on merchants, issuers, and AI platforms all adopting compatible standards; a partial rollout limits where agent checkout actually works.
Best Use Cases
| Task | Why Visa Intelligent Commerce |
|---|---|
| Letting a shopping assistant complete checkout | Issues a scoped token so the agent pays without seeing the card number |
| Enforcing spending limits on an agent | Tokens can carry per-merchant and per-transaction caps |
| Reaching many AI platforms at once | One API integration spans multiple partner assistants |
| Fraud monitoring on machine payments | Applies Visa's real-time authorization and monitoring to each agent transaction |
Getting Started
- Review the Intelligent Commerce program materials on Visa's developer and partner pages to understand the token, authentication, and signals APIs.
- Decide your role — AI platform, merchant, or card issuer — since each integrates a different slice of the suite.
- Engage Visa through its partner channels to provision access and agree on commercial terms.
- Pilot with a narrow use case and strict token limits before expanding to broader agent-driven checkout.
Key Takeaways
- Visa Intelligent Commerce lets trusted AI agents pay on a consumer's behalf using scoped, revocable payment tokens instead of raw card numbers.
- It combines tokenization, agent authentication, payment instructions, and real-time signals, launched in April 2025 with partners including Anthropic, OpenAI, Microsoft, and Stripe.
- Its main strength is layering agent commerce onto Visa's existing, battle-tested payment network.
- The honest caveats are unfinished standards and unsettled fraud-liability and consumer-consent rules for autonomous purchases.
