Con Permiso

Host (Courtney) [0:00]: Alright. Let’s hear it for our speakers. Come on up. Welcome. We have Simon Taylor from Brainfood, Pat Kinsel from Proof, Ines from Visa, and Haley from TD Bank. We’re excited for the panel today. And I’ll pass it over to Gagan from Circle. Gagan Mac, please. Let’s give them a round of applause.

Simon Taylor [0:21]: Alright. Thank you, Courtney, for bringing the energy. We appreciate you so much, and I appreciate Proof and the guys for having me here. I run Fintech Brainfood, the weekly newsletter, but it’s not about me. Today is about bigger than the buy button. You’ve probably heard a lot about agentic commerce this week, but now you can hear from the people at the coalface making it real. So very briefly, I want to start out with Pat. Do you want to explain who Proof is and what they do, very quickly, for people who don’t know?

Pat Kinsel [0:53]: Thank you all for coming, first off. So Proof has been a business for about ten years, and we’ve spent that time digitizing all sorts of transactions and proving who someone is and proving what they’ve done. And so in the agentic space, the hot question is: how do we actually know the identity behind the agent? How do we know what the person actually authorized? And so we’re solving that problem.

Simon Taylor [1:14]: Did anybody see that JPMorgan and Citi and US Bank said, we don’t want your stinking agents? Like, this looks no good. Probably for good reason. There’s probably a lot of risk out there, but what are we going to do? Are we going to block it? Are we going to allow them? Panel, I want to start very quickly. We’ll lay it out. Gagan, are you team block or team allow?

Gagan Mac [1:42]: Team authorize.

Simon Taylor [1:44]: Oh, we’ll get more into that. Ines, how about you?

Ines Clark [1:48]: I think I share the same perspective. And if we want to dive into it, I think that, ultimately, there’s such a huge payment volume opportunity that we should make sure that agentic transactions can create the productivity for consumers and businesses.

Simon Taylor [2:03]: Heck yeah. Haley?

Haley Stelzer-Monahan [2:05]: Yeah. I think it’s a bit nuanced. From a corporate side, software talking to banks isn’t a new thing. When you think about corporates, they have API connections. They have ERP connections, and those are onboarded, credentialed and entitled. But agents are new, and that requires a deeper level of understanding that we’re working with the corporates on. So if we can’t identify a connection that comes in, we can’t tie it to a corporate, and we can’t tie it to a verified human at the corporate, then it’s not coming through, and that’s to protect our clients.

Simon Taylor [2:35]: Got it. So we are team authorize. We are not team block. We’re not necessarily team allow-unfettered in a big way. But, Ines, I want to come to you. I want to know what agentic looks like on the network today. Visa has been doing an awful lot in this space. What led to Intelligent Commerce and Trusted Agent Protocol? What was the push, and what are you seeing?

Ines Clark [3:01]: Yeah. So if we start with your first question, what does agentic look like on the network today? I think what’s quite fascinating is if you had asked this question two months ago, it would have been a completely different discussion on the panel today. At the end of the day, agentic commerce volume just started coming in. We’re seeing it on the network. I think everybody’s played with Muse or with Instinct. We know that Instinct has passed $1 billion in payment volume in a very, very short amount of time.

Simon Taylor [3:35]: Really?

Ines Clark [3:36]: Yeah. And so, back when we were even just prepping for this panel, a lot of the discussion points were around: we know agentic is coming, it’s not at scale today. That’s changing literally in the blink of an eye.

And so to prepare for the time that we’ve now reached today, Visa was really focused on putting in the foundations to ensure that we can actually have the benefit of agentic commerce from a consumer perspective and from a business perspective, with the trust, with the security, and making sure that we have the right identity layer in place. So the past year has been really focused on creating those first products to ensure that that volume could be put through.

So VIC is Visa Intelligent Commerce. This is really our evolution of network tokens to ensure that, now that the consumer is not always clicking the purchase button, our tokens are able to provide the right guardrails: spend guardrails, intent guardrails, category guardrails, which are necessary for today’s world.

Our second product was really focused on the trust and identity layer, something that we know is a priority, something we share with you, Pat. And what the Trusted Agent Protocol does is it really is the opportunity for the network to verify who is this agent. Is this agent actually a trusted entity that should be enabled to transact with the merchants and should be let in to transact? We think those are the two biggest foundations for agentic commerce to actually work at scale.

Simon Taylor [5:32]: I’m so interested to see your worldview on it emerge and thrive in the next year. It feels like this Money20/20 has been the Money20/20 of agentic commerce arriving. I didn’t even know that you’ve done a billion already. I would be surprised if that doesn’t explode soon, but that’s mostly the consumer space, I assume.

Ines Clark [5:53]: Oh, actually, in the case of Instinct, in the case of that stat that I just shared, yes, it is mostly consumer. I think most of us have an Instinct account nowadays. And if you don’t, try to get an invite from somebody here in the crowd. It’s very, very helpful, though very creepy, as we just discussed. But on the business side of things, we actually see a really huge uplift of use cases in terms of accounts receivable, accounts payable. Procurement is also coming soon. So whilst we do not see the same number of transactions, the volumes per transaction are higher, and we do know that it will be a really good opportunity because intent is simpler to manage. If you have an invoice, there’s really less ambiguity about what the intent is.

Simon Taylor [6:41]: Yes. It makes sense. I want to bring Haley in here, because you deal a lot more with the corporate customer side. What are you hearing and seeing about what they’re asking agents to do on their behalf inside the corporate?

Haley Stelzer-Monahan [6:54]: Yeah. Definitely. And I guess to start out, when we think about our corporate clients, they’re living primarily in their ERPs, right? And so a lot of times, they’re actually deploying those agents in those ERPs, whether it’s treasury agents or AP agents. And right now, the conversation is not necessarily around how we can get these agents to move money or how we can get this agent to initiate a payment. It’s really about leveraging those agents to automate highly manual workflows. Some examples would be invoice matching, which you just talked about, reconciliation, cash flow forecasting, cash positioning, maybe even researching missing remittance data or exceptions, and preparing payments, but still with that human in the loop.

Simon Taylor [7:41]: Love that use case, because no finance team and no person on a finance team enjoys that manual work, but automating it was always incredibly hard, because every company does an invoice slightly differently. Every email address is slightly different, and traditional RPA could never get there. And this is low-hanging fruit that people are really starting to succeed at.

But rounding out the rail side, Gagan, there’s another flavor of agentic commerce. There’s consumer and corporate that Ines and Haley have talked about, but what is it looking like at Circle and with USDC?

Gagan Mac [8:15]: Yeah. What we see is very different from what everyone described here. We see a lot more software-to-software payments. What we see is x402, using x401 for authorization, and using USDC. Whenever there is a use case where the payment is sub-30 cents, the card minimum, or whenever the agent wants to find software, data, or a tool for use, generally, x402 and USDC work really well.

A very good example: you can try this today. Tell your Claude, “Build me an app that rings my phone when Money20/20 is done,” and your Claude will get stuck. It’ll say, go buy a phone number from Twilio. And when you do it, that’s when it’ll be able to plug in this code and number. Another flavor of that is to enable your Claude with an x402 endpoint and then ask it to do the same thing. It’ll ring your phone. So when agents try to orchestrate complex actions that span multiple vendors, that is when a human in the loop actually is more of a barrier than an enabler. That’s what we see.

Simon Taylor [9:21]: And so now we have corporate payments and financial institutions and networks and even machines that want to pay, that are starting to pay. But, Pat, we’ve also got some very large headlines about people blocking it and the desire to see things authorized. So how do you prove where the agent got delegated authority from in the first place? Because I imagine with a Visa network, there are pieces of it. With different networks, though, there are lots of payment rails. How am I actually proving who the human is, or the legal entity behind it?

Pat Kinsel [9:56]: Yeah. If I could hope that people walk away with one thing today, it’s that you aren’t going to need to actually understand all of this. And it’s because we’re all working together to bring these things together. If you read some of the protocols that have come out, they’ve all said for the past year that someone has to issue the trusted keys to the consumer. And so that’s really the problem that Proof has solved. With TAP, with x402 and x401, these things really can come together.

What Proof is doing is issuing this credential to the consumer, and then they’re actually signing and binding their identity to the explicit thing that they’re authorizing at the time of transaction. This could be a wire transfer. It could be an account change. It could be buying insurance. There are so many use cases.

And I think what’s important for people to understand, and I think it’s the headline, is that it’s bigger than the buy button. A lot of the early examples of Muse and Instinct use cases, beyond commerce, are “go get me cheaper insurance” or “go manage my financial assets.” And as I think a lot of you understand, those industries have very real regulatory and compliance obligations to capture consent, to be able to address things like Reg E. And so we give people that evidence of the transaction so they can prove exactly what happened.

Simon Taylor [11:18]: I think that proof of what happened, proof of intent tied to a legal identity, has been missing from this conversation for so long. We saw Sierra and Muse and Instinct do the personal agent protocol, but that was just a piece of it. And honestly, you could play a game of bingo with protocols at this point. There are so many of them. It’s an absolute alphabet soup. So thank you for trying to simplify that a little bit.

But agent wallets also have this other unique feature: they can carry spending policies and limits and allow lists and all kinds of other unique programmable functionality. And I think a lot about stablecoins when I think about wallets. So, Gagan, are you seeing people use stablecoins to do that? I know Catena Labs is quite close to Circle, and you’ve done a lot there. Talk to me about what the future of wallets with rules around them for agents starts to look like.

Gagan Mac [12:17]: Yeah. So policies are setting the guardrails. There is one more component that sits between policies and transactions, and that is carrying the human intent with a mandate. The way I view it, users can define policies in wallets. A policy could be: spend up to $5,000, never more than $100 per transaction. A mandate would be: I authorize you to spend up to $50 for getting me this service. And that intent needs to be carried through along with the transaction. I think that’s probably the missing piece right now. Otherwise, we see wallets are very capable of carrying signatures. They are very capable of carrying policies. We already see that through hundreds of billions of dollars of transactions in trading. I think the explicit mandates are the missing piece, and standardizing on those.

Simon Taylor [13:08]: So once I’ve got the policy, how do I link that policy to the humans? Talk me through that a little bit closer, Pat.

Pat Kinsel [13:16]: Yeah. Again, we sign it. There have been a bunch of standards that have come out. One is Verifiable Intent: what is a data payload that can actually transfer exactly what Gagan described with the transaction? We support that today. And this is all coming together with ultimately needing a liability framework and an open set of frameworks to support this.

Simon Taylor [13:43]: And as Visa knows a little bit about liability and standards and frameworks for all of that kind of stuff, how are you thinking about where that fits with human mandates and what the agent did?

Ines Clark [13:56]: So today, with the liability framework that we have, our role is to ensure that transactions that are fraudulent do not get authorized at transaction time, so that we are able to stop fraud before it happens. That’s what we’re focused on, and this is why, linked to identity and guardrails, we have put in place the foundations to track, at every point, the trust and identity data that matters.

By this, I mean that when an agent is given the ability to carry agentic tokens, that agent actually needs to be verified. So this is what we do. Secondly, when an agent is given the actual credentials from the consumer, this also needs to be verified. That consent needs to be given. And then when the consumer is giving the intent for that transaction, this is where we actually need to have that consent done explicitly, back to what you were saying, and have that consent recorded with the transaction. And lastly, when the transaction actually happens, we also need to verify that the consumer...

Simon Taylor [15:20]: I love that about Visa Intelligent Commerce, and I think Mastercard has something similar, which is that you’ve connected the intent to the wallet side. And then when the transaction shows up in the Visa network from the merchant, you’re connecting those two. I’ve got my intent over here. I’ve got the transaction that showed up, and I’m matching those two. And only if they match, and only if the credentials match, do I let it through.

Ines Clark [15:44]: Well, we have two elements, right? First, we want to ensure that the right agents are able to transact, which is TAP. In terms of the intent, we attach those signals to our agentic tokens, so in case of disputes, we are able to provide this information to the merchants, to the issuer, and to the consumer.

Simon Taylor [16:05]: I think what’s so powerful about this is how mature these frameworks already are when the volume is just beginning. So what would you at Visa need to see to let an agent initiate a payment? I guess you talked through a little bit of that, but is there anything else you’re looking for to let that payment through? What does a great payment look like when it hits you?

Ines Clark [16:27]: A great payment. So the ideal framework for us is a payment that uses agentic tokens to ensure that the guardrails that the consumer or the business has put into place are met: it’s the right category, it’s the right spend amount, and it matches the intent. Secondly, we expect that agent to have ideally been verified as a trusted agent that the merchant wants to let in. And with that framework, using the rest of our network products, we’re able to see a great, non-fraudulent payment go through.

Simon Taylor [17:09]: One last question, and I’ll let you go, because I’m fascinated by how the networks are reacting to this. I think about disputes, and I think about compelling evidence. What evidence would you need to see if somebody starts to dispute a transaction at this stage with, “Oh, my agent didn’t really buy it,” that sort of thing?

Ines Clark [17:29]: Yeah. So intent matching is really important. And through [unclear in captions], we can pass an additional consumer object on that. I can pass it back to Pat to share more about what that signed intent looks like and how we can use it. But from our perspective, it’s really about being able to add the additional information that is now needed, given that the consumer actually did not press purchase in this scenario.

Simon Taylor [17:58]: Got it. And Pat?

Pat Kinsel [18:00]: Yeah. I would actually, for a minute, pretend agents don’t exist and just talk about credit cards versus ecommerce. If you think about a credit card transaction, the person actually has a credential, and inside that credential is a certificate. It’s a PKI certificate. That’s what the EMV chip is. So when someone presents a physical card, you’re actually signing the transaction with their credential. When you go on the internet, people are typing in their credit card information. And now with agents, they’re not even doing the typing. They’re a further level removed.

So a lot of what we’ve spent time on with Visa is how you make online transactions look like traditional card-present commerce. And that’s where your face comes into play. Give someone a digital credential, and you present your face, which unlocks access to that certificate, and it actually signs that payment exactly like your credit card can. And so I think we should look at agentic commerce not only as this new challenge, but as a reason to solve online ecommerce fraud completely, and to issue credentials that can be used across all of these different use cases.

Simon Taylor [19:11]: As the accent may give away, I’m from the UK. We have Strong Customer Authentication, and this might be a good wedge to do something similar with the selfie to verify. Gagan, your thoughts on what happens in stablecoin land when a customer is disputing something, given that stablecoin payments are often irrevocable? How are you thinking about that in your own network world?

Gagan Mac [19:35]: Yeah. It’s true. Stablecoin payments are final and settled, especially on layer-one networks like Tempo, like Arc, and others. So there is no clean recourse for reversing a transaction. That concept does not exist in stablecoins. The recourse layer, we believe, should be built on top, and there are mechanisms to do so. For instance, for merchants, there’s merchant reputation based on some of the intent matching you described, so there are intent matching protocols that exist. Another form of recourse is escrow. Money goes into escrow before the service is delivered, and then the payment is released. And once released, it is final. So there are numerous mechanisms to provide recourse to customers. It really depends on the use case. The way I think about stablecoins is as open protocols for fast, final settlement, and then everything else is composable protocols on top.

Simon Taylor [20:28]: You can build that on top. Yeah. And then you see the card networks like Visa doing a lot today with MeshPay and others to instantly issue card credentials, so maybe the user never even feels like they had a card. So it’s exciting times.

I’m actually going to start with Haley on this one, because when disputes are made, there’s a very clear understanding that there was either a human or a business. What gets harder when there’s an actual agent in the loop of that dispute?

Haley Stelzer-Monahan [21:00]: Yeah. It definitely becomes more nuanced on the corporate side. When you think about corporates and disputes, if one of our corporate clients came and said, “Hey, I didn’t authorize this payment,” we’d go and look to see whether or not the payment was authorized: that there was an instruction that came in that we could tie to a corporate, and that it was from a verified human at that corporate.

With agents, it does get a little bit more nuanced. It’s not as black and white. We have to think about the upstream impact that it carries. With an agent in the loop, if we do have a corporate client coming in saying, “Hey, I didn’t authorize this payment,” then: OK, what agent was registered? How was it acting? What authority was it given? What information was it relying on? What events triggered that?

When you think about that, it could be that the agent actually acted within the control framework that was put into place. And so then the question becomes: should the agent have been given that decision-making power to begin with? And when you peel that layer back: was the human who delegated it authorized to give that decision-making power? So it is more nuanced. It’s not as black and white as just looking to see whether or not a payment was authorized.

Simon Taylor [22:23]: Did it match the intent or not? That could be different from whether it was inside policy, which can be different from whether it should have been given that policy in the first place. Ines, your thoughts on how disputes get harder?

Ines Clark [22:38]: How disputes get harder? Well, I’ll remind the audience of what Pat just shared about the transition from in-person transactions to ecommerce transactions. At the end of the day, we ended up having more information about the transaction than we used to have before. And here, when we’re thinking about the move from human transactions done online to actual agentic or human-in-the-loop transactions, we end up gathering even more information in order to build the trust and identity attached to, first, the agent and, second, the transaction. So we could actually be evolving toward a world where disputes can be done against a much richer background of information.

Simon Taylor [23:37]: Maybe it’s an opportunity. I love this framing that maybe agents are an opportunity to fix the thing that’s been going on all along. That came out really loud. Sorry about that. But we do need to fix it, so it was emphasis. It worked.

Cards give us a lot of superpowers that wires and ACH don’t necessarily have out of the box. Haley, I’m interested in your views on when the agent’s not authorized to send it, or the vendor was fraudulent. A lot of the controls inside organizations that manage that start to break. So how are you helping people think about that?

Haley Stelzer-Monahan [24:16]: Yeah. And I’m going to start out by saying I don’t view it as, quote unquote, breaking. Agent or no agent, as the industry evolves and as technology evolves, there are always going to be weaknesses that we have to build solutions against. So I view it as weaknesses. When we think about the agentic framework in this use case, with wires and ACH, the corporates and the banks have robust controls in place to verify human credentials. Think about entitlements, approval limits, dual approvals. All those things have been operating since before agents even became a topic of discussion. And now agents are here, we’re talking about it, and it is more complex. We need to understand more of that workflow to make sure that we have the right safeguards in place, just like with human verification.

Simon Taylor [25:10]: And I think when you’re understanding more of that workflow, the thing that you’re all talking about and saying time and time again is evidence. I need that evidence. Evidence. Evidence. Pat, maybe if I can briefly bring you in. How does the evidence travel with the payment? How do I see all of the upstream things Haley was talking about, like, this is where the invoice came from, and this is what happened before that, and before that?

Pat Kinsel [25:37]: Yeah. What I would say is that we can’t live in a world where everyone has to integrate into everyone else’s services to get data. I think we would all agree with that. So what’s required is that the specific authorization the person is making needs to be signed. It’s one data object, which includes a cryptographic signature, which has the data of who the person is and the actual mandate they’re allowing, that can travel to the next party, and that party can verify that data. And that’s why we spent several years becoming audited to be what’s called a certificate authority, which means we’re a trusted root issuer of credentials on the internet that people can verify even offline. That’s very critical. That’s how credit cards work, again.

I think the other thing that’s happening is that identity and payments are becoming cybersecurity.

Simon Taylor [26:33]: Payments are becoming cybersecurity. 100%.

Pat Kinsel [26:35]: And it’s all becoming the same thing. People’s IT infrastructure can already validate all these types of signatures and authorizations, and it’s all coming together.

Simon Taylor [26:44]: Fascinating. I’m going to go through the panel, because we’ve only got a minute or so left. What’s one thing that has to be true before your institution lets an agent do more? Haley, what do you think that thing is, and are we close to it?

Haley Stelzer-Monahan [27:00]: Asking a bank to pick just one thing is a bit tough. But if I were to sum it up, we would have to make sure that our corporate clients feel comfortable. And in order for them to feel comfortable, they need to understand the use cases, understand which use cases work best for them without impacting their risk, and lastly, that there’s transparency and trust between the corporate and the bank. From a bank perspective, that means we need to see that there’s a registered agent at the corporate, that there’s a verified human who actually set that up, and we need to see the mandate. We need to see the limits. We need to see the guardrails. And back to auditability, we need to make sure that we can track that, so we know the agent is actually acting within the authority it was given.

Simon Taylor [27:44]: Yeah. The corporates feel like they need a lot of education. Ines, how about yourself? What’s that one thing you think we need more of, and are we getting close to it?

Ines Clark [27:52]: I think maybe I’ll do two things.

Simon Taylor [27:54]: Oh, OK. Why not?

Ines Clark [27:57]: I think we all agree, and this has probably been mentioned on every agentic panel, that the ecosystem really needs to work all together to ensure that the different trust and identity elements are in place for those transactions to happen in a way that is safe for the entire ecosystem, and that agents can actually perform. Agents can actually access merchant sites, they can effectively transact, and consumers can continue to use those agents for productivity. So that’s one thing.

Separately, I think we need to see more volume. At the end of the day, in order to keep evolving and setting up the right products and the right frameworks for this space, we need really significant volume. This just started a month and a half, two months ago. So I think we’ll be able to have a much richer conversation in six months’ time, when we can talk about risk scoring and how different actors are performing in this space. How do we want to rate different agents? What do we want to actually put in place?

Simon Taylor [29:16]: Alright. Gagan, rounding us out on what needs to be true. What would it be for you?

Gagan Mac [29:22]: Yeah. I’ll package it as the trust stack for agents. Think about it. If you think about agents as software workers, we’ve all learned to trust strangers on the internet. We get into strangers’ cars. We stay in strangers’ homes on vacation. What is different about a software worker? It’s hard to verify that they are who they say they are and that they do what they say they do. And that, to me, is the trust stack for agents that we believe is evolving over time. It needs to be immutable. It needs to be verifiable. It needs to be open. It should not be able to be spoofed or overwritten. These are the requirements.

Simon Taylor [29:57]: Fascinating. And Pat, to close out. And just before Pat answers, if you’re in the room and you’ve got questions for our panel, we have a couple of minutes for Q&A. So do show me your hands when Pat’s finished his answer. I’d love to get some questions from the room.

Pat Kinsel [30:13]: Well, I’m going to be the guy who says you already can. At this point, our MCP is live in Muse, Grok, ChatGPT, and Claude. And x401, you can deploy in about an hour. So if you want to put x401 on your website and require human presentation and explicit mandate authorization, you could do it today. Now it’s about coming together around standards and policy and driving an ecosystem, and we’re really excited. With Circle today, we announced that we’re collaborating on this so that agents not only can carry identity, but they can buy and do more. And with partners like Circle, we’re going to make this happen. So if folks want to do this, you can do it right now.

Simon Taylor [30:56]: Heck yeah. Alright, I’m going to come to the audience. I’m hoping the microphone doesn’t kill the speakers. Let’s see if there are any questions out here. Now I can see you all. Alright, question here.

Audience member [31:08]: Thank you so much for the panel. I have a question. These all assume that the agent is doing the transaction and the merchant knows the agent. What if the merchant doesn’t know the agent?

Pat Kinsel [31:21]: Yeah, I can take that. That’s actually not an assumption. One of the things we’ve really designed for is the idea that a lot of what AI is doing is going out and procuring new services. So by definition, you do not have a relationship with the entity you’re transacting with. So how can you provide sufficient data that, even if they know nothing about you, they can transact?

In financial services in particular, I don’t know how closely you’re all tracking it, but FinCEN came out recently saying that verifiable credentials issued by a trusted issuer can also be useful in KYC and CIP. So, literal account opening. There’s still more change that needs to happen to promulgate that, and this is where things like TAP are really important. If we’re carrying identity with validated information, and we’re using a framework like TAP to say this is coming from a platform and a system you should trust, then as a merchant, you shouldn’t have to worry about this stuff. If I’m giving the honest answer about what’s required to drive adoption, frankly, merchants shouldn’t have to think about any of this [expletive]. They should just turn it on, and it should work.

Simon Taylor [32:25]: Heck yeah. Alright. Any more questions? Yep.

Audience member [32:31]: Thanks for a great panel. Outside the United States, a lot of this is driven by governance and regulation, specifically in the open banking space. What are your thoughts on the role of governance and regulation in making this all happen?

Pat Kinsel [32:47]: I can take that. Look, what we’ve basically built is PSD2 dynamic linking in the United States, for validated identity, and supporting it with agents. I don’t use that term a lot in the United States, because people don’t know what it means and because we don’t like policy from abroad, but we think it’s the exact same opportunity. As a cryptographic root, we can delegate enrollment. So for us, we’re less focused on enrollment. We’re more focused on the credential and the transactions, and that could plug exactly into a framework like the one in Europe. We’ve got a lot to do, but we see that as completely compatible with the standards in the United States. And I think PSD2 is a great model, frankly.

Simon Taylor [33:29]: Any more questions from the room? Seeing no arms go up. Alright. Well, my final question to the panel, then: if we’re doing this in twelve months’ time, are we saying that volume has doubled or tripled because we’ve nailed bigger than the buy button, because we have that identity? Or do you think it’s going to take a little longer than that? What’s your over-under? I’ll start with Pat, and we’ll go down.

Pat Kinsel [33:57]: It’s more than exponential. It’s orders of magnitude in twelve months.

Simon Taylor [34:01]: Haley, your thoughts?

Haley Stelzer-Monahan [34:03]: From a banking perspective, we definitely take more time to sit with our clients and really understand the processes and the safeguards we need to put in place. So we expect there to be a lot more conversation about that, and an actual focus on use cases, to be able to do it properly.

Simon Taylor [34:20]: Yeah. Understandable.

Ines Clark [34:22]: When Pat said it would be great for merchants not to have to do anything, that would definitely be quite ideal. Now, coming from retail, I will say that commerce tends to be a bit slow on technology adoption. So whilst in twelve months consumers and businesses will be extremely comfortable handing over trust to their agents, I think we will still be working through merchant enablement in the ecosystem, and how merchants can deal with the key strategic questions that this shift in the consumer surface has brought them. This panel is very much focused on identity, but at the end of the day, a lot of those businesses have to think about disintermediation, and about how they keep their relationship with their customers. And that reckoning will take more than twelve months to resolve.

Simon Taylor [35:27]: Wise words, Ines. Gagan, your thoughts?

Gagan Mac [35:30]: If you think about end users, consumers, retail payments, I think it’s going to be more linear, because just because you can do things differently doesn’t mean people will suddenly start spending exponentially more. But if you think about the composition of the transaction itself, the number of transactions, I agree with Pat that it’s going to go exponential, because you’ll see novel forms of transactions happen and novel forms of services get composed. That is where we are really excited about software-to-software transactions. And with software to software, we don’t know yet what that could be.

Simon Taylor [35:58]: Alright. Ladies and gentlemen... Oh, Haley, go ahead.

Haley Stelzer-Monahan [36:00]: I was just going to add to it. When I think about the question, I think a lot of people’s minds go directly to commerce. But going back to the use cases, and why we want to spend so much time on them, I think we are going to see broader adoption across banking. It’s really about agentic authority across banking. It’s not just commerce. It includes commerce, but it’s payments, it’s liquidity, it’s all of that scope. So in terms of commerce, definitely, like what you said, but there’s going to be an increase in adoption across all the other areas as well.

Simon Taylor [36:39]: And if I may double-underline that point, I think corporates and B2B are not explored enough. Everybody defaults to commerce, and corporates and B2B might be the most interesting space I’ve looked at. So, Haley, thank you for that. Well, ladies and gentlemen, what a wonderful panel. Please put your hands together for them.

Host (Courtney) [37:03]: And thank you, Simon. Thank you, everyone.