Written by:IreneDu
This is the 2.5th installment in the Stripe AI strategy breakdown series.
This series originated from the release of 288 products at Stripe Sessions 2026 on April 30th. I observed that Stripe is trying to become the economic infrastructure of the AI Agent era.
The first article, "Stripe is not a payment company," attempts to answer the question, "Why Stripe?"—its very DNA dictates that it can do this.
The second article argues that KYC is dead and the agent economy is rewriting the foundation of financial regulation. I want to dismantle Stripe's bet on the future—what the agent economy actually looks like and why traditional payment infrastructure becomes completely ineffective in the face of it.
But when I wrote the second article, I received a message from a colleague:
I completely agree with the first part. Whether it's AB 316 or the laws of any sovereign state, they won't recognize "Agent as a legal subject" in the short term—the ultimate defendant will always be a specific person. This principle of Know Your Agent cannot and should not be changed.
However, I have reservations about the second half of the statement—"The only change is in payment and clearing efficiency." The problem with this statement isn't the conclusion, but the framework it assumes: it views KYA as an upgrade to the existing payment system.
This is what I think deserves to be discussed in more detail.
Let's first go back to the muscle memory of a former payment industry professional:
Payment methods are driven by specific scenarios, not designed from within the payment system itself.
Every real leap in payment—online banking, mobile wallets, QR code payments—is not because someone has created a better product at the payment layer, but because a new transaction scenario has emerged that has broken through the underlying assumptions of the original payment system.
New payment methods "grow" from the infrastructure required by that scenario, rather than being "optimized".
I worked at Ant Financial for a period of time on payment innovation. At a platform that was once an absolute industry leader, having created "Quick Payment," "Mobile Payment," and "QR Code Payment," the greatest joy and pain came from pondering: What will be the next generation of payment methods?
We developed watch payments (and heartbeat identity verification to replace facial recognition), NFC payments (the original technology of "tap to pay"), participated in and wrote many "next-generation" payment protocols, and even tried to get my boss's support to explore metaverse payments.
Most of these projects failed to materialize.
Looking back, the reason is the same: we tried to define new payments at the payment layer, but the scenarios that drive payment transformation have not yet arrived. Without the right scenarios, the necessary infrastructure cannot grow, and no matter how clever your design is at the payment layer, it won't be able to keep up.
The agent economy is the new scenario I've been eagerly awaiting.
KYA is that layer of infrastructure that is growing out.
KYA is not a payment layer product; it is the infrastructure layer of the Agent Economy.
In my previous article, I defined five layers of KYA: Agent identity, scope of authorization, intent signature, chain of responsibility audit, and credit rating. Of these, only the scope of authorization and chain of responsibility audit fall within the payment chain; the other three layers (identity, intent, and credit) are not even in the payment process.
- The identity layer serves all scenarios that require agent identification: cross-platform calls, regulatory filings, and internal corporate audits—payment is just one of them.
- The intent layer serves the larger issue of AI alignment—payment is just one of its many verification scenarios.
- Credit layers serve any system that needs to assign permissions and limits to agents—payments are just one such user.
Therefore, the colleague's judgment that "the only change is in payment and clearing efficiency" can be translated into infrastructure terms as: KYA is considered a subsystem of payments.
My assessment is the opposite: payment is a subsystem of KYA.
This reversal is the core of what this article will discuss.
Stripe's investment activities at the forefront of industry are a perfect example of this.
At Sessions 2026, Patrick Collison didn't use the term "AI payments," but rather "economic infrastructure for AI." This isn't marketing rhetoric; it's a strategic positioning choice. It indicates that Stripe doesn't intend to confine itself to the identity of a "payments company," but rather it's betting on building the foundation for the agent economy.
Specifically regarding product portfolio:
The Agentic Commerce Protocol (ACP), co-developed by Stripe and OpenAI, is now used by Microsoft Copilot, Meta, and Google Gemini, which joined in April of this year—it is essentially an identity and session protocol, not a payment protocol.
Shared Payment Token separates the Agent from the actual card number, handling authorization layer functions, not clearing layer functions.
Stripe's acquisition of Bridge to obtain stablecoin infrastructure, its acquisition of Privy to obtain embedded wallet capabilities, and its self-built Tempo blockchain as a settlement pipeline—this entire strategy is not within the framework of "payment efficiency optimization".
This investment portfolio only holds true under the premise that "KYA is an infrastructure layer." If the Agent economy is merely a matter of payment efficiency, Stripe doesn't need to develop stablecoins, embedded wallets, or build its own L1 layer. What it's doing is securing its place layer by layer within the five layers of KYA.
Several figures presented by Emily Glassberg Sands, Stripe's head of data, in an interview with Every in April of this year, further confirm the same point: a large AI client had 250,000 fraudulent free trials blocked every week; she had seen an AI company burn through $25 in computing power per free trial with a 4% conversion rate, meaning that it lost $625 for every paid user acquired; and the abuse of free trials has increased fourfold overall in the past six months.
These figures collectively illustrate one point: in the AI economy, the true determinants of whether a transaction will succeed and whether it's worthwhile no longer occur at the moment of checkout—they arise upstream in questions like "Who is this, what do they want to do, and is it worthwhile to allocate resources?" This is why Stripe is shifting its risk control radar from the "moment of the transaction" to the "entire user lifecycle": not by making traditional risk control faster, but by changing the focus from "Is there a problem with this payment?" to "Are there any problems with the entire behavior of this user/agent?" The former pertains to the payment layer, while the latter falls under the KYA (Know Your Customer) layer.
Returning to the question posed by that colleague: who ultimately bears the responsibility?
He's right—ultimately, the legal subject remains a specific individual. AB 316 has already legally sealed this matter.
But this is precisely the real problem that KYA needs to solve: when the chain of responsibility becomes distributed, finding "which link specifically falls on which person" is something that the KYC era did not need to do, but the KYA era must do.
In the KYC era, the chain of responsibility is linear (user → payment/bank → merchant). If a transaction goes wrong, you instinctively know who to contact.
In the KYA era, the chain of responsibility is a network (user → agent platform → model supplier → payment agreement → bank → merchant, and other agents may be involved in the process). Even if the law tells you to "find the person, not the agent", you still don't know who to look for - because the responsibility is already distributed among 5-7 entities.
KYA cannot change the ultimate legal jurisdiction. However, it can use cryptography to solidify the role and actions of each entity within a complex network—who authorized what, who executed what, who settled what, and who fulfilled what. It transforms "no evidence found" into "evidence found"; and "which link went wrong and cannot be verified" into "verifiable."
This is not an improvement in payment efficiency.
This is the first time that accountability has been traced within an agent network.
Therefore, I believe the statement "the only change is in payment and clearing efficiency" reverses the concepts of infrastructure and function.
What actually happened was:
- Because a new type of economic agent has emerged, a new layer of infrastructure (KYA) has been forced to grow.
- This infrastructure layer redefines "who is on the other end, what can be done, and who to contact if something goes wrong"; on top of this infrastructure layer, payment will reorganize itself in a form that we cannot yet fully comprehend today.
What exactly is the next generation of payment methods? What Stripe is trying to define is precisely the new species that it still doesn't understand.
But in a world of uncertainty, one thing I'm certain of is that it won't be designed into the payment layer.
It will emerge from the scene after the KYA infrastructure is laid out.
