On April 14, American Express launched the Agentic Commerce Experiences (ACE) Developer Kit, a protocol framework for registering AI agents, enabling payments, and extending purchase protection to autonomous transactions. Two days later, on April 16, Amex announced it would acquire Hyper, an agentic expense management startup backed by OpenAI CEO Sam Altman.

Protocol on Monday. Product on Wednesday.

That is not coincidence. That is a company executing a strategy with two distinct layers: the rails that agents run on and the agents that run on the rails. Amex just told the market it plans to own both.

When a card network stops building protocols and starts acquiring the companies that build agents, the buildout phase is over. The land-grab phase has begun.

What Hyper Actually Does

Hyper, also known as Hypercard, was founded in 2022 by Marc Baghadjian and Nikolas Ioannou. Both made the Forbes 30 Under 30 list in Finance. The company raised $4.5 million, with roughly half coming directly from Sam Altman. Other backers include Gene Lockhart, former CEO of Mastercard, and Marc Randolph, co-founder of Netflix.

The product is straightforward but genuinely useful. Hyper's AI agents handle the grunt work of corporate expense management: auto-categorizing charges, filing reports, checking submissions against internal budgets and company policies, and nudging employees when filing deadlines approach. Anyone who has spent a Friday afternoon reconciling receipts understands the appeal.

Hyper and Amex already had a relationship. In 2024, the two partnered to launch the Hypercard Rewards American Express card, which embedded AI-powered expense agents directly into the card experience via Hyper's Agile Partner Platform.

"Our customers want smarter, more efficient ways to manage expenses," Raymond Joabar, group president of Global Commercial Services at American Express, said in the announcement. "We're thrilled to welcome Hyper, a team with deep expertise in designing and deploying AI agents, as we build next-generation AI capabilities."

The deal is expected to close in Q2 2026. Financial terms were not disclosed. Hyper's team will fold into Amex's commercial services division, where they will build an AI-powered expense management platform slated to launch before the end of the year.

Protocol Plus Product: The Two-Layer Play

Here is why this acquisition matters more than its modest deal size suggests.

Card networks have spent the past year publishing agentic commerce protocols. Visa built Intelligent Commerce Connect and the Trusted Agent Protocol. Amex built ACE. Each is a framework that tells AI agents how to register, authenticate, and transact on a given network. We mapped the full landscape in our agentic commerce protocol guide. The protocol layer is genuinely crowded.

But protocols are specifications. They describe how agents should behave. They do not build the agents themselves.

Amex appears to have concluded that publishing a protocol is necessary but not sufficient. You also need the product layer, the actual AI capabilities that sit on top of the protocol and do useful work. The ACE Developer Kit handles the plumbing: agent registration, account enablement, intent analysis, payment credentials, cart context as a risk signal. Hyper handles the application: agents that categorize expenses, enforce policies, and generate reports.

Protocol without product means you built a highway and hope someone else puts cars on it. Product without protocol means you built cars with no road. Amex now has both.

That is a fundamentally different bet than what Visa is running. Visa's Intelligent Commerce Connect launched on April 8, supports over 30 partners in its sandbox, and integrates with four different agentic protocols. It is an infrastructure play. Visa is positioning itself as the neutral settlement layer that any agent can use, regardless of who built it. The Featurespace acquisition for $946 million in late 2024 added fraud detection capabilities, not agent-building capabilities.

The contrast is worth watching. Visa says: we will be the platform. Amex says: we will be the platform and the product.

The M&A Signal

This is, by our count, the first acquisition of an agentic AI company by a major card network specifically for agent-building capability. Visa's Featurespace deal was about fraud intelligence. FIS has been assembling agentic commerce capabilities through partnerships rather than acquisitions. Nobody else has bought a company whose core output is AI agents designed for financial workflows.

That matters because M&A tells you where a company sees defensible value. Partnerships tell you where it sees opportunity. Protocols tell you where it sees table stakes. When Amex published ACE, it was signaling that agentic commerce infrastructure is becoming a baseline requirement, something every network will need. When it acquired Hyper two days later, it was signaling that the differentiation lives above the protocol layer.

The timing makes sense. McKinsey projects that agentic commerce could drive up to $1 trillion in orchestrated U.S. retail revenue by 2030, with global projections reaching $3 trillion to $5 trillion. At those numbers, the question shifts from "should we participate" to "where in the stack do we capture value?"

Amex is answering that question by moving up the stack. Not just enabling agents. Building them.

Why Expense Management Is the Beachhead

Expense management might sound unglamorous next to autonomous shopping agents and agentic checkout. But there is a reason Amex chose this entry point.

Corporate cards are the core of Amex's commercial services business. Every corporate card generates expense data. Every expense report is a workflow that someone, usually an exhausted finance team, has to process manually. The pain is real, the data is already flowing through Amex's rails, and the buyer is the same enterprise customer Amex already serves.

This is a classic wedge strategy. Start with the workflow where you already own the data relationship, prove the agent works, then expand. Today the agent categorizes your expenses. Next quarter it flags policy violations in real time. Next year it negotiates supplier terms or routes payments to optimize rebates.

We covered this pattern in the agentic commerce standards race. The companies that will win are the ones that control a specific, high-value workflow end to end, not the ones that build the broadest horizontal platform. Amex is starting narrow and deep. Expense management for its existing commercial card base. That is roughly 10 million corporate cards globally.

The existing Hypercard partnership validates the approach. The 2024 card launch means Amex has at least 18 months of data on how agents perform in the expense workflow. This acquisition is not a bet on an untested idea. It is a scale-up of something already working.

What This Means for the Agentic Commerce Stack

Pull back and look at the stack that is forming across the card networks.

At the protocol layer, the picture is increasingly clear. Visa has the Trusted Agent Protocol. Amex has ACE. Stripe and Tempo have the Machine Payments Protocol. OpenAI and Google have their own agent commerce protocols. FIS is connecting issuers to multiple protocols simultaneously. The infrastructure layer is being built by many hands, and it is progressing fast.

At the product layer, the picture is thinner. Most card networks are still relying on partners and developers to build the agents that use their protocols. Amex just became the first to internalize that capability through M&A. That gives it a vertically integrated position that nobody else in the network layer currently holds.

The risk for Amex is that it picked one narrow use case, expense management, while the market develops broader agent capabilities elsewhere. If autonomous shopping agents or procurement agents become the dominant category, Hyper's team may need to pivot fast.

The risk for everyone else is that Amex proves the model works, and the acquisition race kicks off in earnest. If owning the product layer above your protocol creates measurably better outcomes for customers, every network will need its own agent-building capability. Buy or build, the clock starts now.

The Bigger Question

Two years ago, the conversation about AI in payments was theoretical. One year ago, it was about protocols and standards. Today, it is about acquisitions and control.

That progression tells a story. When an industry moves from publishing specifications to buying companies, the early infrastructure phase is ending and the competition for market position is beginning. Amex just fired the starting gun.

The deal is small. Hyper raised $4.5 million. This is not a $946 million Featurespace-scale acquisition. But the strategic signal is loud. Amex believes that owning agent capabilities, not just agent infrastructure, is the play. And it moved faster than anyone else in the network layer to secure those capabilities.

McKinsey calls it "agentic commerce." The card networks call it the future of their business. Whatever you call it, the phase where everyone was building protocols and waiting to see what happens is over. Someone blinked. Someone bought.

Now we find out who is next.

Sources

Amex bought the agents. Visa is building the highway. Which strategy wins when every AI needs a wallet?

Charlie Major is a Product Development Manager at Mastercard. The views and opinions expressed in Major Matters are his own and do not represent those of Mastercard.