A card network renewing an issuing agreement with a fintech is routine. Integrating the fintech's technology into its own infrastructure is strategic. But using the fintech's product for its own corporate operations? That is something else.

Visa did all of it this week.

The company announced a renewed multi-year issuing partnership with Ramp, deeper integration of Visa Intelligent Commerce and the Visa Trusted Agent Protocol into Ramp's platform, and something that most coverage buried: Visa is becoming a Ramp customer, using Ramp's AI agents for targeted corporate service use cases.

Ramp has been building AI agents for corporate finance in stages, and the capabilities are now specific enough to evaluate.

The company launched its first AI agents in July 2025, designed to enforce expense policies, block unauthorised spending, and stop fraud in real time. In October 2025, it shipped Agents for AP: three specialised agents for accounts payable.

The Invoice Coding Agent automates invoice coding by applying learned logic. According to PYMNTS, it gets 85 percent of accounting fields right on the first attempt and improves with every cycle.

Fraud detection works differently. The Fraud Detection Agent analyses 63 data points per invoice, from payment history to vendor details, flagging suspicious invoices and vendor changes before the bill is even created. For early access customers, these agents flagged over $1 million in fraudulent invoices in 90 days.

The third agent, Card Payment Optimisation, finds card payment opportunities directly in vendor payment portals, capturing cashback that would otherwise be left on the table.

These are bounded, production agents with measurable outputs. They code invoices, flag fraud, and optimise payment routing. No chatbot features. No text generation. The distinction matters because it separates Ramp's approach from the broader agentic AI hype: specific financial tasks with clear accuracy metrics.

The partnership goes deeper than issuing. Ramp is integrating two of Visa's agentic commerce initiatives into its platform.

Visa Intelligent Commerce (VIC) is Visa's infrastructure layer for AI agent-initiated payments. It provides the framework for agents to initiate and complete transactions securely at Visa's global scale. Ramp is applying VIC specifically to streamline corporate bill pay operations, allowing customers to earn cashback on card payments through an automated process.

The Visa Trusted Agent Protocol, introduced in October 2025, is an open framework that helps merchants and payment processors distinguish between malicious bots and legitimate AI agents acting on behalf of consumers or businesses. More than 20 agents and agent enablers are integrating directly with VIC, including Akamai, Worldpay, and Cloudflare as co-developers of the Trusted Agent Protocol.

"Enterprises are looking for payment solutions that reduce friction, not add to it," said Chris Newkirk, President of Commercial and Money Movement Solutions at Visa. "Ramp's approach to automation and real-time controls aligns with Visa's mission to make commerce simpler and more secure."

For Ramp, the integration means its AI agents operate within Visa's trust and authentication framework rather than building a parallel one. For Visa, it means VIC and the Trusted Agent Protocol are being validated in production by a fintech processing over $100 billion in annual purchase volume.

Consumer agentic commerce, where AI agents shop and buy on behalf of individuals, gets the headlines. B2B payments automation is where the economics are more immediate.

Corporate bill pay is still overwhelmingly manual. A typical accounts payable workflow involves receiving invoices, matching them to purchase orders, coding them to the correct accounts, routing for approval, scheduling payment, and reconciling the results. Each step involves human judgement that is largely pattern-based. That makes it a strong fit for bounded agent automation in a way that open-ended consumer shopping is not.

Ramp now serves 50,000+ customers including CBRE, Shopify, Anduril, Figma, Notion, and Cursor. The company has generated $10 billion in cumulative savings for customers and saved 27.5 million hours. Its enterprise customer base grew 133 percent year-over-year in 2025, with more than 2,200 customers contributing $100,000 or more in annualised revenue. Ramp crossed $1 billion in annualised revenue on the back of a $32 billion valuation in November 2025.

Colin Kennedy, Ramp's Chief Business Officer, framed the philosophy: "The best financial systems don't add controls after the fact. They build them into every transaction."

That is the B2B thesis in one sentence. Consumer agentic commerce has to solve discovery and trust for open-ended purchasing decisions where preferences are subjective. B2B agent payments have to solve invoice processing and payment routing for structured, repeatable workflows where accuracy is measurable. The second problem is more tractable, and the ROI is easier to prove. As we have seen with Visa's tokenisation push and Santander's live agentic payments, the card networks are building infrastructure for both. But B2B is where the agents are already earning their keep.

Ramp is the flagship B2B implementation, but Visa's agentic commerce ambitions extend further.

The company has 100+ partners collaborating globally on agentic commerce. More than 30 are actively building in the VIC sandbox. Visa has completed hundreds of controlled, real-world agent-initiated transactions and expects millions of consumers to use AI agents to complete purchases by the 2026 holiday season.

On the consumer side, Visa cites research showing that 47 percent of U.S. shoppers now use AI tools for at least one shopping task. Partners include Skyfire for consumer product recommendations, Nekuda for fashion, and PayOS for payment infrastructure. In the Middle East, Visa partnered with Aldar for AI agent-managed property fee payments.

The geographic rollout is advancing in parallel: active pilots in the U.S., programmes anticipated in Asia Pacific and Europe in early 2026, readiness in Latin America within the year.

The Ramp deal answers a question that has hung over agentic payments: who will be the first major fintech to integrate a card network's agent trust framework into production B2B workflows? Ramp answered it.

The next questions are harder. Will the Trusted Agent Protocol become a standard that other fintechs and corporate payment platforms adopt? Will competing approaches from Mastercard (Agent Pay, Verifiable Intent) and Stripe (Machine Payments Protocol) converge or fragment the market? And will the 85 percent accuracy on invoice coding hold as agents handle more complex, multi-jurisdiction, multi-currency AP workflows?

Visa becoming a Ramp customer is the strongest signal of confidence in the category. Card networks are not known for adopting unproven technology for their own operations. If Visa's internal teams are using Ramp's agents for corporate payments, the bet is not on a partnership. It is on a thesis: that bounded AI agents, operating within defined trust frameworks, can automate B2B payments at scale.

If Visa trusts Ramp's AI agents enough to use them for its own corporate payments, what is stopping the rest of the enterprise market?