Visa Inc. (NYSE:V) expects millions of consumers to use AI agents to complete purchases by the 2026 holiday season. The question for investors is what happens when those agents start deciding not only what to buy, but how to pay for it.

The Bear Case Visa Must Answer

In February, Citrini Research published a hypothetical scenario in which AI agents identified the cost attached to card transactions and shifted settlement toward cheaper stablecoin infrastructure. Visa, Mastercard Inc. (NYSE:MA), American Express Company (NYSE:AXP) and DoorDash Inc. (NYSE:DASH) fell after the report.

There is an important distinction in Visa’s case. The company does not earn the 2% to 3% interchange fee, which moves from acquirers to issuing banks. Visa says in its annual report that the fees it receives from issuers and acquirers are not derived from interchange or merchant discount rates.

The more serious bear case is that agents unbundle the payment stack. If an agent can choose a cheaper settlement rail while sourcing identity, authorization and fraud protection elsewhere, Visa loses volume without ever having collected the interchange fee.

Visa Wants To Make Its Trust Layer Harder To Remove

Visa launched Agentic Ready in Europe on March 17, then expanded it to Asia Pacific, Latin America and Canada. More than 50 issuers joined the Asia Pacific rollout alone. The program lets banks test enrollment, tokenization, authentication and transaction controls before agent-led purchases reach volume.

The strategic objective is to keep the credentials, permissions and protections surrounding an agent transaction attached to Visa’s infrastructure. Consumer caution helps that case. A Visa survey published in April found only 27% of Americans were comfortable letting an AI agent spend without limits, while 60% would not allow an agent to spend any amount without approval.

Early Tests And The Copyability Problem

In December, Visa said it had completed hundreds of real-world agent-initiated transactions with more than 100 partners. DBS Group Holdings Ltd (OTC:DBSDF) has piloted agent-initiated purchases with Visa in Singapore, and on Aug. 19 DBS Hong Kong announced a partnership with Visa and Preface to extend that work to consumers.

For Visa, copyability cuts both ways. If no single bank can build a lasting moat around agentic features, value may migrate to the infrastructure that lets many banks deploy them. The opposite outcome is also possible: large banks could build more of the stack themselves and connect it to non-card rails. Mastercard is already there, having introduced Agent Pay for Machines in June to settle across cards, accounts and stablecoins.

Stablecoins Make Citrini’s Scenario More Credible

On Sept. 1, 21 financial institutions including Bank of America, Citi, Goldman Sachs and Wells Fargo announced plans for a bank-issued dollar stablecoin targeting a first-half 2027 launch. The institutions on the issuing side of today’s card ecosystem are building another settlement option, and software can compare rails on every transaction in a way people do not.

Visa is hedging. It joined more than 140 companies behind the Open USD stablecoin in June and launched its own Stablecoin Platform in July. Its Intelligent Commerce Connect product, introduced in April, accepts both Visa and non-Visa cards across four agent protocols. The company appears to be preparing for a world in which the rail underneath a transaction changes while Visa still supplies the identity, tokenization and trust layer around it. That role may come with thinner economics.

What Investors Should Watch

Visa enters this transition from strength. Fiscal third-quarter net revenue rose 14% to $11.6 billion, and processed transactions rose 10% to 71.7 billion. Agentic payments are far too small to move those numbers yet.

Watch how many Agentic Ready issuers move from testing into production, how quickly banks copy DBS, and whether Visa starts disclosing agent-initiated volume. The decisive evidence will come when an agent picks something other than a card. If Visa still earns on identity, authentication and orchestration in that transaction, Citrini’s scenario changes the business without displacing it. If those functions can be bought more cheaply elsewhere, the bear case gets much harder to dismiss.

Benzinga Disclaimer: This article is from an unpaid external contributor. It does not represent Benzinga’s reporting and has not been edited for content or accuracy.