Visa, Mastercard and Google are building agentic payments. None are solving the real problem.

Visa, Mastercard and Google are all building agentic payments. None of them are solving the real problem.

Every major player in payments made the same move in the same quarter.

Google launched Universal Commerce Protocol. Visa announced Intelligent Commerce. Mastercard shipped Agentic Tokens. Checkout.com positioned as the single connection point. Razorpay built agentic payments on Anthropic’s Claude. Cashfree partnered with Mastercard and OpenAI for in-chat transactions.

The conclusion is unanimous: AI agents will transact on behalf of consumers and businesses, and the payment infrastructure underneath needs to support that.

What is less unanimous is who controls it.

$1 The checkout page is no longer the destination

For twenty years, every digital payment ended the same way. Redirect to a form. Enter card details. Submit. Whether the buyer arrived from a website, an app, or an email, the destination was fixed.

That model assumed a human at a screen. It assumed a single payment surface. It assumed the platform knew in advance which PSP would process the transaction.

None of those assumptions hold when an AI agent is completing a purchase mid-conversation, a voice agent is taking a payment mid-sentence, or an autonomous workflow is settling an invoice without a human in the loop.

The payment surface is no longer a page. It is a layer that exists wherever software is already running.

$1 The convergence nobody is connecting

Consider what happened in the contact centre market this quarter alone. Talkdesk shipped Automation Flows. Genesys launched agentic virtual agents powered by Large Action Models. Five9 crossed $100 million in AI ARR. RingCentral integrated OpenAI into live voice calls. NICE acquired Cognigy for $955 million.

Five of the largest CCaaS platforms shipped agentic AI products in the same window. Not pilots. Products.

These agents can update accounts, resolve service requests, verify identity, and orchestrate multi-step workflows autonomously. But the moment a customer says “I’d like to make a payment,” the architecture breaks.

PCI compliance. Secure card capture. PSP routing. Settlement. Refunds. None of that was designed for an AI agent to initiate. The payment moment is the gap between “almost autonomous” and “fully autonomous.”

And this is not unique to contact centres. Insurance platforms, travel systems, ERP tools, commerce engines. Every vertical where software orchestrates a workflow that ends in a transaction faces the same structural gap.

$1 Four protocols, one problem

The industry’s response so far has been to build proprietary standards. Google has one protocol. OpenAI has another. Visa and Mastercard have their own. That is four competing agent payment frameworks, and we are still in the first quarter of serious deployment.

Each assumes platforms will adopt its standard. Each assumes a relatively closed loop where the network or PSP controls the transaction end to end.

But the platforms actually deploying agents do not care which protocol wins. They care that their agents can transact. Across channels. Across geographies. Across providers.

An AI voice agent handling a payment for a UK insurance platform needs different PSP routing than the same agent processing a US e-commerce checkout. The agent does not get to pick one network standard and hope it covers every scenario.

This is the same tension that existed before agentic commerce. Enterprise platforms have always needed PSP optionality, not PSP lock-in. The difference now is that agents compress the decision cycle from minutes to seconds. There is no human in the loop to manually switch providers or re-route a failed transaction. The payment layer underneath has to handle that automatically.

$1 Distribution, not orchestration

The payments industry has spent the last five years debating orchestration. Routing. Failover. Tokenisation. These are useful capabilities, but they solve the wrong problem for what is coming.

The real question is not “how do I route between PSPs?” It is “how do I make every PSP available inside every software surface where a payment needs to happen?”

That is a distribution problem, not a routing problem.

PSPs need to be present wherever platforms are deploying agents and workflows. Platforms need the freedom to support whichever PSP their enterprise customers or regional requirements demand. And the infrastructure connecting those two sides needs to work across voice, chat, links, embedded checkout, and whatever surface ships next quarter.

The winners in agentic commerce will not be the companies that build the biggest walled garden. They will be the ones that make PSPs distributable into software, without forcing platforms to pick a side.

$1 The infrastructure layer matters again

Payments processing was commoditised years ago. The value moved up the stack to monetisation logic, platform economics, and distribution.

Agentic commerce is accelerating that shift. When AI agents are the payment surface, the infrastructure layer between platform and PSP becomes the most consequential piece of the architecture. Not the network. Not the individual PSP. The neutral layer that connects them.

The companies that figured out checkout pages twenty years ago won the last era of payments. The companies that figure out the multi-PSP, multi-channel infrastructure layer for AI-native payment surfaces will define the next one.

The agents are coming. The payment layer underneath them is the part most of the industry has not built yet