Visa's Intelligent Commerce Connect: one integration, four agentic protocols, and the case for an abstraction layer

On April 8, Visa announced Intelligent Commerce Connect — a single on-ramp that lets businesses accept AI agent-initiated payments across four major agentic commerce protocols through one integration on the Visa Acceptance Platform.

The press release framed it as protocol-agnostic, network-agnostic, and token-vault-agnostic. Strip the marketing language and what Visa shipped is something more consequential: an explicit bet that brands shouldn’t have to pick the winning protocol, and an infrastructure layer that lets them stop trying.

That bet — abstraction over selection — is arguably the most consequential architectural move in agentic commerce so far this year. Here’s why, and what brands should do about it.

---

What Intelligent Commerce Connect actually is

Three things in one product:

1. A multi-protocol acceptance layer. Intelligent Commerce Connect supports payment initiation through the four agentic protocols currently in the field: Trusted Agent Protocol, Machine Payments Protocol, Agentic Commerce Protocol (ACP), and Universal Commerce Protocol (UCP). An agent built against any of those four can initiate a payment that lands cleanly on the merchant side without the merchant having integrated against each protocol separately.
2. A multi-network acceptance layer. The product integrates both Visa’s own Intelligent Commerce APIs (for processing agent purchases using Visa cards) and other networks’ APIs. Agents can pay with Visa or non-Visa cards through the same merchant integration.
3. A discovery and acceptance bridge. The product also includes functionality to help merchants expose their product catalogs — descriptions, specifications, and pricing — to AI platforms, so consumers can discover, select, and complete purchases inside an AI-driven interface.

The launch quote from Andrew Torre, President of Value-Added Services at Visa, captures the framing:

“From small businesses to the world’s biggest retailers, Visa powers how people pay every day, millions of times over. Intelligent Commerce Connect brings that same, trusted payment acceptance infrastructure into the emerging world of AI-driven commerce, so businesses can let AI agents buy on behalf of consumers, securely and at scale.”

Currently in pilot with Aldar, AWS, Diddo, Highnote, Mesh, Payabli, and Sumvin. Broader availability planned across 2026.

The four protocols, in one diagram

The point: agent → protocol → ICC → network → merchant. The merchant only has to integrate at the ICC layer. The protocol the agent chose and the network the card belongs to are abstractions ICC handles.

Why this matters: the abstraction-vs-selection bet

The four-protocol problem is not academic. Each of ACP, UCP, Trusted Agent Protocol, and Machine Payments Protocol has its own technical spec, its own backers, and its own go-to-market motion. A brand that integrates against ACP today gets access to ChatGPT’s Instant Checkout flow. A brand that integrates against UCP gets access to Google’s AI Mode and the growing Google Universal Cart ecosystem. A brand that picks one and not the other is making a multi-quarter bet on which AI surface its customers will use most.

Most brand operators do not want to be making this bet. They want their products to be accepted across whichever surface their customer happens to be using when intent fires. Intelligent Commerce Connect is the first piece of major-network infrastructure that lets them defer the protocol decision indefinitely — by abstracting it to the acceptance layer.

That’s an architectural pattern with precedent. The card networks built durable businesses partly because they built acceptance layers that abstracted the network choice away from the merchant, and the merchant could accept whatever card the consumer brought. Intelligent Commerce Connect is making the same play one level up: abstract the protocol choice away from the merchant, so the merchant can accept whatever agent the consumer brings.

If the analogy holds — and it’s a strong analogy — the brands that benefit most from agentic commerce won’t be the ones who picked the right protocol early. They’ll be the ones who picked the right acceptance partner and let that partner handle the protocol layer.

The pilot partner list, decoded

The seven launch partners aren’t a random mix. Each one represents a class of integration ICC is meant to serve:

• Aldar — a large real-estate group. Shows ICC is intended for verticals beyond pureplay ecommerce.
• AWS — the leading cloud platform by market share. Signals that ICC’s APIs are being designed to be embedded in agent infrastructure at the hyperscaler level.
• Diddo — an agent platform. Tests the agent-side integration.
• Highnote, Payabli, Sumvin — issuer-side and acquirer-side processors. Tests the network-grade plumbing.
• Mesh — payment infrastructure for digital assets. Signals that ICC is engineered to coexist with non-card payment rails as those evolve.

The shape of the pilot suggests Visa is validating not just “can we accept an agent payment?” but “can we accept an agent payment from any of these agent stacks, on any of these card platforms, through any of these processors, into any of these merchant types?” That’s the stress test the abstraction layer needs before it goes broadly available.

What this doesn’t solve

Important to be honest about scope. Intelligent Commerce Connect solves the acceptance problem. It doesn’t solve the discoverability problem.

A brand that integrates ICC and ships great catalog data still needs the AI agent to choose its product in the first place. ICC handles the payment after the agent has decided. The agent’s decision is upstream — and the agent decides based on the product data quality the brand exposes to the AI surfaces. That part is unchanged by anything Visa ships.

This is the gap that Juniper Research’s $1.5T projection sits inside. The infrastructure to capture agentic commerce dollars is being built fast. The catalog readiness to be picked by agents in the first place is lagging — and it’s the brand’s job, not Visa’s.

What brands should do this quarter

If you’re a brand operator reading this, three concrete moves:

1. Stop debating which protocol to bet on. The acceptance layer is being commoditized faster than any individual protocol can win. The right architectural call is to plug into a network-grade abstraction (ICC is arguably the most credible one shipped to date) and let it handle the protocol question.
2. Move the agent-readiness work forward. The lever that determines whether an agent picks your product is data quality on your side: identifier coverage, descriptive depth, real-time pricing, structured data completeness. Every brand will end up with payment acceptance through an abstraction like ICC. Not every brand will be the one the agent recommends. The AI Readiness Score is the operator-grade signal for where this work needs to land.
3. Talk to your existing acquirer about agentic acceptance. ICC is in pilot with seven partners as of April 8, but every major payment platform is racing to ship a comparable layer. If you’re already working with a tier-1 acquirer, the protocol-abstraction conversation is one they’ll be ready to have within the next two quarters. Get on the early adopter list. The brands that integrate at the abstraction layer first will be the ones picking up the incremental agentic AOV that the $1.5T Juniper projection sizes — without having had to pick a winning protocol to do it.

---

The big takeaway from April 8: brands don’t have to choose between agentic protocols anymore. They have to choose whether to invest in being chosen by the agent in the first place. That’s a much better problem to be solving.

If you want to know where your catalog stands on the data quality that determines whether AI agents recommend you, get in touch.