AI agents spark a new battle in the payments industry! Bernstein: Visa (V.US), Mastercard (MA.US), and others usher in new opportunities.

Investment bank Bernstein noted in a research report released recently that Meta's (META.US) launch of Muse has opened a new era and a new round of competition for consumer AI agents. The bank believes that on the path of agent commercialization, payments may become more strategically important, and the next major battle in the payments industry may be about to begin. Bernstein pointed out that as agent commercialization evolves rapidly, payments are keeping pace. Muse has already been used to perform various tasks: shopping (which Meta says is rapidly becoming one of its largest usage drivers), tracking prices after booking flights and hotels, making reservations, cleaning up inboxes, and coordinating outfits. However, the payment experience for agents is still somewhat clumsy. At checkout, Muse fills in the merchant's own checkout page through Stripe's Link or Shop Pay. AI agent startup Instinct says 40% of its users share personal credit cards within three weeks, though its early users may be more tech-oriented and higher-income than ordinary consumers. The fundamental question surrounding consumer AI agents in the payments industry is: in an agentic world, who will control payments, liability, and dispute resolution frameworks? What role will payment networks and PSPs (payment service providers such as Stripe and Adyen) play? Will this create incremental revenue opportunities (for example, for V/MA, Adyen, and Stripe), or will payments processors face commoditization risk if payments are abstracted or tokenized? Trust remains a major bottleneck for agentic commerce only 23% of U.S. consumers say they trust generative AI to handle payments on their behalf. Traditional business models are designed around a "human in the loop." In a world where agents crawl the internet, fill out forms, and purchase goods, a series of questions arise around trust: How do people know an agent is a "good agent"? How is programmability of transactions achieved? How is identity verification performed (for example, two-factor authentication, or 2FA, in Europe)? How are user consent and user intent ensured? How should liability frameworks be designed? How are disputes handled? How is fraud risk reduced? How are chargebacks managed? How are audit trails established? How are different payment methods optimized? How is personalization achieved? And how are standards and technological fragmentation addressed? The battle for control over agent transactions A "tug-of-war" is about to unfold over control of agent transactions, as well as control of credentials/payment tokens. Agents bring far greater complexity to a transaction than before. In the era of agentic commerce, trust and governance are crucial (even more important than in e-commerce), and also highly challenging and entities that can provide agent credentials/tokens and the related governance mechanisms will reap enormous benefits. To be fair, agentic commerce still accounts for a tiny share of e-commerce, but it is developing quickly, and payments companies are preparing for an agentic world. Both major payment networks, Visa (V.US) and Mastercard (MA.US), have launched agent tokens (Visa Intelligent Commerce, Mastercard Agent Pay), embedding more data and contextual information into token credentials (for example, deeper insight into transaction type, where the token is used, and the identity of the payer), while adding safeguards (signals based on behavioral history and relevant configurations). At present, adoption of these products and whether issuers actively participate (for example, around agent control and liability frameworks) remain unclear. In any case, 100% of agent transactions will be tokenized (though not necessarily using agent-specific tokens). Meanwhile, Stripe launched Secured Payment Tokens (SPT) for agent transactions a year ago. It should be noted that Muse or Instinct services most likely do not yet use Visa and Mastercard's "agent-specific" tokens, but instead use Link or one-time virtual cards meaning payment networks can still capture these transactions, but cannot capture the lucrative "agent" token business, which in the future could become a platform for distributing agent-facing services. Muse transactions are also covered by Link's purchase protection mechanism. Interestingly, Visa noted in July that it had established a partnership with Meta to use agent tokens within Facebook and Instagram, but no specific information specifically about Muse could be found. Whether payment networks can seize a leading position in agent transactions through "agent tokens" may ultimately depend on two words liability shift. Even if other tokens are technically equally advanced, or even more advanced, payment networks still play a key role in dispute resolution and liability frameworks, and this could become extremely important in agentic commerce. Learning from history? The battle for control over agent transactions has similarities to the token wars of 2014. Twelve years ago, with the launch of Apple Pay, Visa and Mastercard also launched their own token services. At the time, there was also a great battle over who would control tokens payment networks, issuers, and PSPs competed. Today, although multiple types of tokens coexist, payment network tokens have gained widespread adoption and now account for 60% of e-commerce transactions. Tokenization has also provided payment networks with a genuine leading position, enabling them to distribute their own services through this channel, such as providing fraud scores, automatically updating payment credentials, and push provisioning issuer credentials to merchants. Today, they are finally beginning to price these services as well, and there is still a long runway for pricing growth in the future. The potential market opportunity could be enormous The potential revenue scale could be very substantial. For example, Stripe currently charges 0.15 cents for SPT. Once payment networks' agent tokens gain widespread adoption, they can distribute their own services through these tokens. In an optimistic scenario, if agent transactions ultimately account for around a mid-single-digit to high-single-digit percentage of overall consumer transactions, this could contribute several percentage points of growth for payment networks. What is worth watching? Factors to watch include: issuer and merchant system integration (rather than simply defaulting to on); the holiday shopping season (and agent adoption); new platform partnerships; whether Alphabet Inc. Class C or Meta's Muse will implement agent tokens; and agent experiences launched by banks (for example, Amex, Chase). By the way (and this is also instructive for platforms), none of these recently launched agent products currently let agents control merchant payments (or take on the Merchant of Record role) at least, that model is quite complex. Instead, these products mainly provide consumers with a frictionless way to complete tasks and shop (for example, Muse completes payments through Link or one-time virtual cards). Merchant system integration faces serious difficulties, and controlling the checkout experience, which typically belongs to merchants, also faces many challenges, including managing conversion rates, real-time pricing, post-purchase services, merchant catalog and inventory synchronization, applying loyalty and discounts, fulfillment, and multiple shopping baskets. Investment implications Bernstein believes payments are a beneficiary, not a loser, and that in a world of consumer AI agents, payments may become even more strategically important. One possible scenario is that in the future consumers conduct more transactions through agents (smaller basket sizes, more long-tail product discovery), which would benefit payments companies, including card payment networks. In addition, agents can also drive further increases in online penetration for complex services and some retail categories. This may be less intuitive, but higher e-commerce penetration is generally very favorable for cards, because cards have higher online penetration (as well as higher attach rates for value-added services, or VAS). Consumer AI agents may also mean more complexity in areas such as agent identity verification/validation, consumer intent, dispute handling, and agent-specific cards/limits therefore, this also creates opportunities for V/MA and payment processors (such as Stripe, Adyen, and Checkout) to differentiate themselves and provide value-added services. It should be noted that 100% of agent transactions will be tokenized (meaning V/MA may earn higher revenue per transaction, while also gaining a channel to distribute more services). Agent-to-agent commercial transactions may also emerge later, which could require a richer set of services and therefore would also be a positive factor. Bernstein has been thinking deeply about this issue and believes digital wallets (such as PayPal) are in an interesting position in the consumer AI agent space and have strategic optionality (for example, if agents struggle to find an effective commercialization model). On the one hand, if agents handle people's payments, will people really still care about the friction of "guest checkout"? On the other hand, two-sided wallets like PayPal (PYPL.US) can significantly expand their value proposition, including trust, protection, fraud prevention, and dispute handling, while also helping consumers dynamically find cost-saving opportunities, rewards, and personalized services, and optimize payment methods (for example, choosing between BNPL, credit/debit cards). Buy now, pay later (BNPL) providers such as Affirm (AFRM.US) and Klarna (KLAR.US) may also be well positioned, especially in areas such as: interest-bearing loans for near-prime and subprime customers (including costs such as late fees), dynamic credit underwriting (can they negotiate with agents?), 0% APR loans (for example, for prime customers), and terms transparency. In the research report, Bernstein assigned "Outperform" ratings to Visa, Mastercard, Affirm, Adyen (ADYEY.US), Block (XYZ.US), and Toast (TOST.US); and "Market Perform" ratings to PayPal, FIS (FIS.US), Fiserv (FISV.US), Global Payments (GPN.US), and Klarna.