How AI agents will change payments, commerce, and fraud - Amir Sarhangi (CEO, Skyfire) — Transcript
Full transcript
- 0:00The projections are by 2030, a trillion
- 0:04dollars worth of [music] commerce will
- 0:06be driven by AI agents. First step is
- 0:09just to even understand how often agents
- 0:11are coming to their website,
- 0:13>> [music]
- 0:13>> right? Today, one of the challenges
- 0:15merchant merchants have is that they
- 0:17don't actually understand what that
- 0:19[music] is. If you look at the
- 0:20statistics, the amount of traffic on the
- 0:22internet has 10x [music]
- 0:24in the last couple of years, right? And
- 0:26over 50% of that is
- 0:28>> [music]
- 0:28>> essentially bot traffic. Those types of
- 0:30actions will create a lot of problems
- 0:33when it comes to fraud and when it comes
- 0:35to security.
- 0:35>> Okay, guys, machine economy. Well, it's
- 0:38complex, right? I I totally understand.
- 0:41Okay, let me simplify it. AI economy,
- 0:43all AIs making purchase on behalf of us.
- 0:47It is still um quite complex. I totally
- 0:49get it. That's why today I'm joined by
- 0:52Amir Sarhangi. I'm CEO of Skyfire, and I
- 0:57believe this is the right person to
- 0:58explain us what is um
- 1:01machine or AI economy is, how AI is
- 1:04going to change how we do purchases, how
- 1:06we do e-commerce, and how we can
- 1:09safeguard ourselves when AI agents start
- 1:12making the payments on behalf of behalf
- 1:14of us. Amir, welcome to the show.
- 1:17>> Sam, thanks for having me. Great to be
- 1:19here.
- 1:20>> Right, Amir.
- 1:21Just the first question straight. What
- 1:24is know your agent? If you could explain
- 1:26a couple of words.
- 1:27>> Yep. Know your agent uh is actually a
- 1:30term that we came up with a few years
- 1:31ago when we first started the business.
- 1:34And really is synonymous with uh when
- 1:37you think of KYC, know your customer.
- 1:39So, know your know your customer is
- 1:41about
- 1:42actually doing verified check, an ID
- 1:45check on on a human being when you open
- 1:47a bank account, for example.
- 1:49Uh similarly uh for KYA, which is know
- 1:53your agent, is understanding who really
- 1:55this agent is representing. So, it's a
- 1:58combination of verified information on
- 2:00the human
- 2:02that's behind the agent, the agent
- 2:04developer itself. That agent developer
- 2:06could be a company like uh
- 2:08ChatGPT or Anthropic. Uh or could be an
- 2:12individual developer um with like
- 2:14OpenClaw or something like that. Uh so,
- 2:16it's both of the those two information
- 2:19combined in a token we call KYAI token.
- 2:22Uh and that know your agent also has
- 2:24information about the intent of the
- 2:26user. So, for example, if Sam's trying
- 2:29to purchase uh tickets using an AI
- 2:32agent, uh has what what is Sam trying to
- 2:35do? And then, has Sam given permission
- 2:38to that agent to do that on his behalf?
- 2:41>> Right. And
- 2:42I mean, look, um I want to understand um
- 2:45when
- 2:47a company develops an AI agent, I
- 2:49understand like for example, OpenAI or
- 2:51some some other big companies. But when
- 2:53an individual developer, let's say, I
- 2:55sit at home, I develop my own agent, and
- 2:58it starts making um
- 3:00a purchase on my behalf, right? At
- 3:02Skyfire, how do you guys understand
- 3:06um that this is a legitimate bot? Uh you
- 3:09can trust it, you can trust the
- 3:10information it's sending you guys. What
- 3:13I'm trying to say that I'm trying to
- 3:14understand if there are certain rules, I
- 3:16mean, in the industry or you guys
- 3:18created perhaps that says, "Yes, you can
- 3:20trust this agent, you can trust this
- 3:22AI." Or no, you cannot try trust that.
- 3:25How do you do that at Skyfire?
- 3:27>> Yeah, great question, Sam. So, that's
- 3:29probably one of the biggest uh problems
- 3:31that exist today uh is that it's very
- 3:35difficult to understand who these agents
- 3:37are. If you look at the statistics, um
- 3:40the amount of traffic on the internet
- 3:42has, you know, 10x'd in the last couple
- 3:45of years, right? And a lot of that is
- 3:48over 50% of that is essentially uh bot
- 3:52traffic.
- 3:53>> Okay.
- 3:53>> Now, some of this bot traffic is
- 3:55legitimate, some of it is actually
- 3:57potential revenue
- 3:59for a lot of the websites that these
- 4:01agents are trying to go to, but a lot of
- 4:03it is also fraudulent as as you
- 4:05mentioned. Like a lot of times like you
- 4:08need to know who these bots are. So,
- 4:10really what we
- 4:11you know, what really we have been
- 4:13embracing is with KYA is this concept
- 4:16that
- 4:17bot developers need to go through a KYB
- 4:20process so that you actually know who
- 4:23this is. So, in in the case of an
- 4:25individual or an enterprise, they would
- 4:28go through a KYB which is know your
- 4:30business and that's proving that
- 4:33this is really who you say you are.
- 4:36And then additionally
- 4:37when it comes to the end user having
- 4:39verified information on who that end
- 4:42user is becomes pretty critical because
- 4:45if you're trying to deal with a merchant
- 4:47or a business, that business needs to
- 4:49know who you are
- 4:50and needs to be able to in order to be
- 4:53able to provide you the service that
- 4:54you're asking for.
- 4:56>> Yeah, true. And um
- 4:58before going deeper into this topic,
- 5:00could you briefly explain what you what
- 5:02you guys are doing at Skyfire? What
- 5:05problems you are solving?
- 5:06>> Yeah.
- 5:07So, at Skyfire really there are three
- 5:10areas that we're focused on. One is
- 5:13agent identity.
- 5:15Uh secondly is around access. So, how do
- 5:17agents access um
- 5:20basically human websites, right? Like
- 5:22just be able to go to directly to a to a
- 5:25website and be able to interact. And
- 5:27then thirdly, ability for to be able to
- 5:30check out. So, for example, if
- 5:33an agent goes to a website, discovers a
- 5:35product it likes, how can it actually
- 5:38purchase it like a human being does?
- 5:40>> Right.
- 5:41>> The challenge today is that most of
- 5:43these agents get blocked uh, different
- 5:45layers of web security. So, for example,
- 5:49um, most merchants have uh, bot blocking
- 5:51technology like Cloudflare, Akamai. Uh,
- 5:54today we in fact we announced with
- 5:56Akamai and Experian.
- 5:58Um, and then additionally you have fraud
- 5:59checks. Uh, this is folks like Forter
- 6:02who are at fraud
- 6:04basically inspecting to make sure this
- 6:06is a real human being. And then, you
- 6:09know, one of the other big things is
- 6:11that it when you're purchasing, uh, a
- 6:13lot of times you want to be able to use
- 6:15your loyalty. So, you know, let's say
- 6:17you go to Best Buy, you want to be able
- 6:18to make sure that uh, purchase that
- 6:20you're doing also counts for your
- 6:23loyalty. So, ability to log in becomes
- 6:26another big aspect and that's typically
- 6:28protected by folks like Okta.
- 6:31>> All right. And um, where do you sit in
- 6:33this flow? I'm trying to understand. Do
- 6:35you directly work with merchants or you
- 6:37work with payment service providers?
- 6:39Where do you sit? And um,
- 6:42should merchants go um, make the
- 6:44integrations? How does it work if you
- 6:45could explain as well?
- 6:47>> Yeah, great question. So, uh, what we do
- 6:49is we work directly with the demand
- 6:51side. So, this is like an agent
- 6:53developer. An agent developer uses our
- 6:55technology. In this case, uh, what we
- 6:58call it a KYA token. They create this
- 7:00KYA token. That token has the verified
- 7:03information on the human and the and the
- 7:06uh, agent itself. Uh, and then that's
- 7:09used in the header when an agent makes a
- 7:12request to a website. That's included as
- 7:14part of the header.
- 7:15Uh, we have gone out and done uh,
- 7:18basically partnerships with a host of
- 7:20web security companies.
- 7:22Uh, again, this is people like Akamai,
- 7:24um, DataDome,
- 7:26>> Right.
- 7:26>> Cloudflare, and they uh, these web
- 7:30security companies accept this token and
- 7:32then provide access to the websites.
- 7:35>> Right.
- 7:36>> of this model is that it actually scales
- 7:38across the whole internet without
- 7:41all of the different merchants and
- 7:42websites having to support
- 7:44>> Mhm. anything new. So, there's no new
- 7:47requirements other than their web
- 7:49security company accepting this token.
- 7:52>> But But, you just mentioned that, for
- 7:54example, yeah, this is also not
- 7:56something new. For example, let's say
- 7:58um agent developer needs to go and do
- 8:02KYB.
- 8:03Oh, yeah. Agent developer. Now I Now I
- 8:06get it. Yeah. Yes.
- 8:07>> [snorts]
- 8:08>> And um
- 8:09for agent developer to do the KYB, this
- 8:11was my question I was thinking about.
- 8:13Let's say I'm individual. I don't have a
- 8:14business. I'm just a developer who sits
- 8:16at my home and just develop it. How
- 8:19would I do a KYB? What would you require
- 8:21from me?
- 8:21>> that's a That's a good question, Sam.
- 8:23So, that basically turns into a KYC. So,
- 8:26in the case where there is no there is
- 8:29no entity,
- 8:31uh it becomes essentially a KYC,
- 8:33um and that becomes an individual
- 8:35developer who basically is accessing uh
- 8:38something on his on on his his or her
- 8:40behalf, right?
- 8:42>> Okay.
- 8:43>> So, yes. But, there's still a
- 8:44verification process.
- 8:46>> And you still have to go through it.
- 8:47>> Yes.
- 8:48Exactly.
- 8:48>> Yeah.
- 8:49And if you talk about
- 8:51machine economy, where do you see at the
- 8:54moment AI agents are being used the
- 8:56most? Is it like e-commerce or
- 8:58subscription economy? Where is it
- 9:00happening the most at the moment?
- 9:02>> Yeah. So, look. I mean,
- 9:05purchasing with bots has something
- 9:06that's been happening for many years
- 9:08already.
- 9:08>> Okay.
- 9:10>> So, for example, if you if you look at,
- 9:12you know, the reason why when you want
- 9:15to go to a really popular concert and
- 9:17within seconds uh most of it is sold out
- 9:20even though you're refreshing your web
- 9:22page, it's not just because there's a
- 9:24lot of fans trying to buy. It's also
- 9:26there's bots that are coming in and
- 9:28trying to purchase on behalf of
- 9:30resellers. Same thing is true with Nike
- 9:32drops and many other places in
- 9:35e-commerce. So,
- 9:37bots have been purchasing, you know, I
- 9:40wouldn't say legally, but
- 9:42you know, in a gray zone manner.
- 9:45So, that's something that's already been
- 9:46happening it and you have companies like
- 9:48now that are much more, you know, front
- 9:51and center about it like Rye is one of
- 9:53our partners, Card AI. There's a bunch
- 9:56of players that are already now
- 9:58purchasing
- 10:00either on behalf of businesses or
- 10:02consumers. That's already happening. On
- 10:05the machine economy side of things when
- 10:06it comes to micro payments, I'd say it's
- 10:08still early days, but we're starting to
- 10:10see
- 10:12basically ability for agents on on their
- 10:15own to be able to pay for content and
- 10:18also be able to pay for APIs or access
- 10:21to
- 10:22whatever it is that they're trying to
- 10:23get to. But, I'd say you know, from our
- 10:26perspective, we see that it's still at
- 10:27its early days. Part of the challenge
- 10:29there is that
- 10:31a lot of the companies on the supply
- 10:33side of this equation continue to want
- 10:35to be able to sell SaaS, right? It's
- 10:37it's much better to sell sell a SaaS
- 10:39package than a on-demand per
- 10:43request basis. But, that is going to
- 10:45change.
- 10:46And the reason that will change is
- 10:48because you have many companies who will
- 10:51not want to pay the big SaaS dollars,
- 10:55especially if they're not using the full
- 10:56package. They'll go and try to, you
- 10:59know, use upstarts or even create their
- 11:01own, you know, stack. So,
- 11:04for those reasons, those pressures are
- 11:06going to force
- 11:08the those SaaS bundles to
- 11:10be unbundled and be offered on a more
- 11:13ala carte basis.
- 11:14>> You just you just mentioned micro
- 11:17payments and I often hear that it is
- 11:19one of the key parts of machine economy.
- 11:21Why are micro payments so important in
- 11:24this machine economy? What the role they
- 11:25play?
- 11:26>> Yeah, I mean,
- 11:28basically if you want to now, let's say,
- 11:30let's say you use HubSpot or you or
- 11:33agent wants to have access to some
- 11:36particular content
- 11:38for learning or for just needs access to
- 11:41it and wants to be able to pay for that
- 11:42content on a per article basis or wants
- 11:46to have access to an API
- 11:49or
- 11:50for whatever it is it needs.
- 11:51>> Yeah.
- 11:52>> That API call is a single call and a
- 11:54single access and it may come back a
- 11:56hundred more times and use it again, but
- 11:59each time it's going to pay a cent or a
- 12:01fraction of a cent for that access. And
- 12:04that's where that micro payment becomes
- 12:06important because
- 12:08in in this scenario the agent is not
- 12:10signing up to some big plan. It's saying
- 12:12I'm going to pay you every time I use
- 12:14your product, which means that the the
- 12:17provider has to be able to sell its
- 12:19product on a essentially on a single hit
- 12:22basis.
- 12:24>> Right. Right. And then as you know,
- 12:26credit rails are not best for micro
- 12:28payments. Those those are very
- 12:30expensive. Credit rails especially they
- 12:32are very expensive. That's why
- 12:34everyone's trying to avoid that. Do you
- 12:36think that's why
- 12:37stable coins would become like let's say
- 12:39default currency for machine economy or
- 12:42do you think
- 12:44the current rails adopting and they will
- 12:46still exist in one or another way?
- 12:50>> I think look, I think
- 12:52you're going to see both, but I this is
- 12:54where the sweet spot for stable coins is
- 12:57because it's not just the micro payment
- 12:59aspect of it. It's the fact that you're
- 13:02actually transferring value at the time
- 13:04of the service. So imagine a scenario
- 13:06where you hire another agent that then
- 13:10hires another agent to do something. In
- 13:12that chain, you know, you have an agent
- 13:15that's getting a commission or getting
- 13:16part of a transaction. Now imagine
- 13:19there's fraud or you know, this can't be
- 13:22done in credit because like trying to
- 13:24roll that back or figure out who caused
- 13:27what fraud becomes a big hassle, right?
- 13:30Whereas here with stablecoins, once that
- 13:33money is transferred, like
- 13:36you know, that that next agent owns
- 13:39that, you know,
- 13:41um
- 13:41that currency, which is that becomes
- 13:43really critical. That uh that value
- 13:46exchange uh at the time of service is is
- 13:49pretty critical when it comes to kind of
- 13:51this new agentic world that we're going
- 13:53to see.
- 13:54>> I see. And do you guys do transactions
- 13:56actually? Or you do not touch money at
- 13:59all? You are only working on, as you
- 14:01said, as a know-your-agent and basically
- 14:04assisting the transaction.
- 14:06>> Yeah, but we provide a wallet. In fact,
- 14:07we started with stablecoins.
- 14:09>> All right.
- 14:10>> we have with our partnerships with Visa
- 14:12and Mastercard, we've added uh both uh
- 14:16Visa tokenized cards and Mastercard, and
- 14:19we'll we'll be also uh adding others. Um
- 14:23so,
- 14:24our wallet basically will provide the
- 14:26agent all the different mechanisms uh
- 14:29for payment, essentially.
- 14:30>> Okay.
- 14:31Okay. And uh
- 14:33why should I use your wallet, for
- 14:34example? Why cannot I go um
- 14:38use only your KYA, let's say,
- 14:40know-your-agent, and continue using
- 14:42Stripe or any other service provider
- 14:44that I'm using at the moment?
- 14:46>> Yeah, I mean, an user really can. The
- 14:47big advantage of using us, uh other than
- 14:50being obviously a one-stop shop for you,
- 14:52is the fact that when an agent is trying
- 14:56to go to a website or to uh an API, one
- 15:01of the important things that that end
- 15:03destination wants to know is that
- 15:06there's real intent for purchasing.
- 15:08Meaning, we can say that like san- the
- 15:11same agent
- 15:12has a credit card on file and is a real
- 15:17buyer, right? And then at the time that
- 15:20you do try to make a purchase, um
- 15:23understanding not only that this is not
- 15:25a fraudulent card that you're using.
- 15:27That's something that we're we're
- 15:29working with Experian to unlock is this
- 15:31kind of real-time risk
- 15:33aspect.
- 15:35And then secondly is is the fact that
- 15:39we capture the intent of the user in
- 15:41that process, meaning that you've given
- 15:44authorization
- 15:46to the agent to make that purchase on
- 15:48your behalf. All that together becomes
- 15:50really critical for merchants who are
- 15:54trying to provide, you know, Sam AI, for
- 15:57example, AI agent,
- 16:00a service or a product.
- 16:02>> Does it mean that Well, basically, if I
- 16:04understand correctly, if I take the
- 16:06bundle from you, meaning like your
- 16:08wallet and other services, it will be
- 16:10more secure than combining it with,
- 16:13let's say, other providers that I'm
- 16:15already using. Do I understand it
- 16:16correctly?
- 16:17>> Not only more secure, but also you get
- 16:20ability to go to probably more
- 16:25destinations and websites because now we
- 16:27have a complete
- 16:29understanding of who this agent and the
- 16:33user behind it is and the intent of that
- 16:35user. So, it's really becomes a
- 16:37combination of access and security in
- 16:39that scenario.
- 16:40>> And
- 16:41what are some of the
- 16:43industries that are adopting AI agents
- 16:45in the first place?
- 16:47And here's an interesting question, how
- 16:49do you guys make money out of this?
- 16:51>> [laughter]
- 16:52>> Well, we we make money in a few
- 16:55different areas, but really
- 16:58initially really it's about adoption.
- 17:00So,
- 17:02where we make money is KY A tokens being
- 17:04created,
- 17:06and that's something that, you know, we
- 17:08work directly with agent developers
- 17:11with,
- 17:12but also in the future, you know, we'll
- 17:14hopefully also be able to uh provide
- 17:18more uh risk signals to the merchants.
- 17:21That will be another area. Um in terms
- 17:23of, you know, we're seeing quite a few
- 17:26different areas in terms of um
- 17:28adoptions. Uh we're working uh with uh
- 17:32not only kind of AI-first vertical um
- 17:37developers who are essentially focused
- 17:39on things like, whether it's fashion,
- 17:41shopping, uh travel, you're going to see
- 17:44a lot more personalized AI assistant
- 17:47type experiences
- 17:49um that are going to be um essentially
- 17:52uh providing a really high-end
- 17:54experience to uh to users.
- 17:57Uh and then the other area that we're uh
- 18:00getting some great traction around is
- 18:01financial institutions.
- 18:03Um this is card issuers, this is um
- 18:06folks who are essentially uh providing
- 18:10uh you today a wallet. Uh
- 18:13and and that wallet could be either
- 18:15through uh
- 18:16your your card or your bank. Uh so
- 18:19that's another area that we're starting
- 18:21to see
- 18:22uh quite a bit of interest.
- 18:23>> I mean, what do you think? Which
- 18:25industries are adopting AI agents the
- 18:27first? Where do you see the real
- 18:29adoption curve basically going up?
- 18:32>> Yeah, so in the consumer space, uh I
- 18:34mean, there's consumer and B2B. On the
- 18:36consumer side, what we're seeing is that
- 18:38you have uh AI-first companies that are
- 18:42being created now. Those companies are
- 18:44starting really with a completely new
- 18:47user experience. It's more of a, you
- 18:49know, kind of a chat commerce
- 18:50experience.
- 18:52That's uh you know, we're starting to
- 18:54see that in fashion, we're seeing that
- 18:56in travel, um and you know, many kind of
- 18:59different categories in the consumer
- 19:01space.
- 19:02You're also starting to see
- 19:04existing companies uh who previously are
- 19:08like a marketplace. Uh this is where you
- 19:10go to discover new things like think
- 19:12Pinterest or others like that who are
- 19:15also wanting to take the friction out of
- 19:18the buying experience. So, anywhere
- 19:20where you have a referral link where
- 19:23that referral link takes you to another
- 19:25site, that can be now more of a buy for
- 19:27me button. It's kind of like that
- 19:28instant purchasing experience where you
- 19:31now don't have to go to these
- 19:34steps and websites to to purchase. So,
- 19:36that's more on the consumer side. Um
- 19:38>> Okay.
- 19:41>> companies, you know, whether it's your
- 19:43card network, anyone anywhere where you
- 19:46have a wallet like think Chase Preferred
- 19:48or
- 19:50you know, Capital One.
- 19:52Those types of experiences will evolve
- 19:54in the future and you know, you will now
- 19:57just be able to say, "Hey, this is what
- 19:59I'm looking for." It'll find you the
- 20:01best price, apply the coupons, and be
- 20:04able to make that
- 20:05a much more of a concierge
- 20:07uh kind of a VIP experience for you.
- 20:11>> Do you do do you think
- 20:13I mean, we are in agent or AI economy or
- 20:17it's very very early stage. Why I'm
- 20:18asking this question is that you must
- 20:20have read that OpenAI and Stripe they
- 20:23launched a partnership so that they
- 20:25enable purchasing directly inside the
- 20:27chat. But but then they scaled back this
- 20:31partnership.
- 20:33Um there was a news. So, why do you
- 20:36think that happened in your opinion?
- 20:38>> Um
- 20:40There there there are many different
- 20:41reasons
- 20:42for that. You know, I obviously I'm not
- 20:45inside OpenAI. I cannot give you
- 20:47>> Sure.
- 20:47>> anything but from a outsider's
- 20:49perspective, you know, I think
- 20:51um the experience was not there yet, you
- 20:54know, from a from a consumer side. Uh
- 20:57and really, you know, if you think about
- 20:58it from a consumer perspective,
- 21:00when you
- 21:02want to use a
- 21:03AI agent,
- 21:04part of the reason you do this is
- 21:06because you wanted you expect that agent
- 21:08to do a better job than you
- 21:10Okay.
- 21:18Okay.
- 21:35Mhm.
- 21:35>> [clears throat]
- 21:40>> Yeah.
- 21:43Yeah.
- 21:50And I think that's where you know, at
- 21:52the end of the day, you know, any of
- 21:54these experiences for them to work
- 21:56really well, they need to be
- 22:00scaled across the internet. And then
- 22:02more importantly, there are other
- 22:04implementation aspects that have to be
- 22:06considered around fraud
- 22:08and risk that, you know,
- 22:12become really important in any
- 22:14implementation. So,
- 22:16you know,
- 22:17your question is a is a really good one,
- 22:20but I think a big part of the reason
- 22:24there was a pause in that particular
- 22:27instance is primarily because I think
- 22:29they rushed into it very quickly to to
- 22:31launch something.
- 22:32>> So, you you think we are very early
- 22:35stage and there is no trust because
- 22:37again, I have another example is
- 22:39Walmart. They also rolled out basically
- 22:42this similar types of solutions, but
- 22:45they rolled back and the news said that
- 22:47basically customers were not converting
- 22:50from chat to their website, you know, at
- 22:53all. But it was like the conversion
- 22:54rates were very very low. So, they said
- 22:58like it
- 22:59we are all in this back. Do you think
- 23:00this is like trust issue? People do not
- 23:03trust AI agents or there is something
- 23:05else?
- 23:06>> I think it is Trust definitely is a
- 23:08factor in terms of adoption of any new
- 23:11technology. Even if you look at the
- 23:13early days of uh
- 23:15you know, commerce on the web, people
- 23:18would not trust using their credit card
- 23:21on a website. That was not too long ago.
- 23:24That That was just, you know, three
- 23:25decades ago when, you know, Amazon was
- 23:28trying to push people to use their card,
- 23:30right? And
- 23:32that evolution is one where you have to
- 23:35ease the user into that experience. Uh
- 23:38you know,
- 23:39the way we see it is like three
- 23:41different steps. The first step is this
- 23:43experience where you see something, you
- 23:46like it, and you say just buy it for me,
- 23:49right? So, that's a place where you've
- 23:50already discovered what you want, and
- 23:53you make that purchase. So, that's
- 23:54that's the first way of easing the user
- 23:56into the experience. The second step is
- 23:59more of a commerce chat experience where
- 24:01you don't know what you want yet, um and
- 24:04you're trying to figure out
- 24:06uh what's the best option for you. So,
- 24:08you start chatting
- 24:09um kind of like what you do search, but
- 24:12here it's a much better experience
- 24:14because you can actually
- 24:15um do it in natural language,
- 24:18be able to quickly find what you want um
- 24:21just like you do today in Open AI. I
- 24:23mean, the statistics are kind of
- 24:25astounding how much
- 24:27uh now you know, uh ref- referrals are
- 24:30coming from
- 24:31uh AI companies. The numbers are pretty
- 24:34astounding, and the projections are by
- 24:382030, a trillion dollars worth of
- 24:40commerce will be driven by AI agents.
- 24:43And And those numbers will happen very
- 24:46quickly. That's just 3 years from now.
- 24:48And the reason is that you're going to
- 24:49have this progression where you start
- 24:51from this kind of buy for me experience
- 24:52to
- 24:54uh agentic kind of
- 24:56um
- 24:56chat experience, and that's still in the
- 24:59loop. You're You're still approving
- 25:01everything. So, when the agent goes and
- 25:03finds you what it wants, it'll ask Sam,
- 25:05"Hey Sam,
- 25:06do I have approval to buy this pair of
- 25:08Nike shoes for $110?" You approve it
- 25:11with your passkey and then goes and
- 25:13purchases for you. So, as you can see,
- 25:15it's just slowly building trust and then
- 25:18kind of the third phase of this is more
- 25:20autonomous where
- 25:22you want to buy a pair
- 25:24you know, you want to buy some concert
- 25:26tickets. You don't want to
- 25:28stay awake till midnight when they go
- 25:29live. You tell your agent, "Hey,
- 25:33if if you can find me these tickets
- 25:35under $200, just buy it. I approve it
- 25:38like beforehand with you.
- 25:41>> [clears throat]
- 25:41>> That becomes kind of the more autonomous
- 25:43experience and that's what we're
- 25:46starting to see and yes, you are going
- 25:48to have implementations that are not
- 25:51going to be the best ideal user
- 25:52experience and they will need to be
- 25:55rethought, but there's absolutely no
- 25:57question in my in my mind now that
- 26:00agentic commerce is going to happen
- 26:02because we're starting to see it already
- 26:04working
- 26:05and more importantly, it
- 26:07just gives
- 26:08you know, in in some cases, people want
- 26:10to shop. In a lot of cases, people don't
- 26:12want to shop and they much rather take
- 26:15that friction out and that's where
- 26:17agentic commerce will help
- 26:20free up time for people, make it much
- 26:22easier for them to do their daily tasks.
- 26:26>> I mean, I understand for end consumer,
- 26:28okay.
- 26:29But, I I will not go that far to the
- 26:32future. You are saying to certain sense,
- 26:34we are already in machine economy.
- 26:36Machines are already making purchases.
- 26:38Then, what do you think merchants should
- 26:40understand about machine economy? How
- 26:43they should be prepared for that? What
- 26:46should they think about it and consider?
- 26:49>> First step is just to even understand
- 26:51how often agents are coming to their
- 26:54website, right? Today, one of the
- 26:56challenges merchant merchants have is
- 26:58that they don't actually understand what
- 27:01that is. And and the reason I say this
- 27:03is that when you look at you go because
- 27:06obviously merchants are driven by data,
- 27:07right? Like data is king for them. And
- 27:10one of the things they care a lot about
- 27:12is understanding how often humans are
- 27:14coming to their website, what are they
- 27:16searching for, how often they leave a
- 27:18basket uh with you know, goods in it. Um
- 27:22and now they're starting to see a big
- 27:25part of that pie chart of humans being
- 27:28gray. That gray is basically a
- 27:31combination of fraudulent agents and
- 27:33good agents that are coming to their
- 27:35website to try to understand their
- 27:37product or try to purchase something.
- 27:40>> Right.
- 27:40>> They need to understand that that gray
- 27:42slice much better than they do today.
- 27:44And that's that's the first thing they
- 27:46need to be able to do is know who what
- 27:49type of agents are coming to them on
- 27:51behalf of what type of humans.
- 27:53And then being able to essentially
- 27:55unlock
- 27:56agents being able to purchase directly
- 27:58from the human website. No APIs, we're
- 28:01just talking about the regular old
- 28:02website because agents can do that now
- 28:05very effectively.
- 28:06>> Okay.
- 28:07>> And being able to unlock that, which
- 28:09then gives them much better
- 28:10understanding of uh what agents are
- 28:15driving revenue for them
- 28:17>> Yeah.
- 28:17>> and being able to kind of foster that
- 28:19and um and then being able to build
- 28:22essentially a a strategy around how they
- 28:25support agents going forward.
- 28:28>> What about financial institutions? We
- 28:31see you explained what should merchants
- 28:33do. What about financial institutions?
- 28:35Because they also need to be prepared,
- 28:37right?
- 28:38>> Yeah, so on the on the financial
- 28:39institution side, I think there's a
- 28:40couple of things. One is
- 28:42um you know,
- 28:44you if you're a card issuer, um if today
- 28:48you are offering as a financial
- 28:50institution you are offering an
- 28:51experience to end use end consumers,
- 28:54the thing what that's really
- 28:57like this AI
- 28:59you know, wave does for you is that it
- 29:01kind of evens the playing field. It
- 29:03allows you to provide a much better
- 29:05experience to your users
- 29:07that is very differentiated in the sense
- 29:09that now you can provide
- 29:11you know, let's say
- 29:13you're a card issuer you have a wallet.
- 29:15Now you can provide kind of this like
- 29:17really
- 29:19great concierge experience for your
- 29:21users to be able to use your card on the
- 29:24purchases that they want to do. So, it's
- 29:27you know, think about it as kind of
- 29:28Amazon Prime where like once you're
- 29:31locked in and starting to use Amazon
- 29:34Prime for one thing you'll start adding
- 29:36more and then it's just on automatic.
- 29:38It's the same story here. Once you start
- 29:41using an AI agent to do purchasing and
- 29:43you and that your card issuer is
- 29:46basically that's the default card you're
- 29:48going to be using
- 29:49then you are starting to really get your
- 29:53consumers to use your card for a lot of
- 29:56their purchases that are happening with
- 29:57it with agents. So, that's one part of
- 30:00the static strategy is really leaning
- 30:02into
- 30:04this new trend that will happen
- 30:08and and the first step is just you know,
- 30:10kind of that buy for me experience and
- 30:13then
- 30:14from there more of a concierge
- 30:15experience.
- 30:16The second area where financial
- 30:18institutions are starting to need and
- 30:21and needing to evolve very quickly is
- 30:23that their customers want to be able to
- 30:26access their accounts using agents. So,
- 30:29for example,
- 30:30you know, being able to
- 30:33access your bank account or access your
- 30:35financials even in a read-only fashion
- 30:39using Claude or ChatGPT or Gemini so
- 30:42that you can do analysis on your account
- 30:45for various right? So
- 30:48and similarly enterprises who are
- 30:50automating they want access to their own
- 30:53accounts using agents. So this becomes
- 30:55another big area for financial
- 30:57institutions. This is not so much
- 30:59agentic commerce but it's more like
- 31:01agentic access. So how does it
- 31:04How do I access my own account using
- 31:06agents essentially? That's the other
- 31:08area for financial institutions.
- 31:10>> I mean this is my last question. What do
- 31:13you think the biggest risk in the near
- 31:15term in machine economy? What should
- 31:18both merchants and financial
- 31:19institutions
- 31:21be ready?
- 31:23>> Yeah look you know I think in some ways
- 31:26if done right we can actually reduce
- 31:29fraud believe it or not
- 31:32in agent with agentic commerce because
- 31:35you have a much more secure way of
- 31:38onboarding the user and the card and
- 31:42then
- 31:43the
- 31:43essentially the way
- 31:45um
- 31:47everything moves from there is can be
- 31:50done in a very secure way. I think the
- 31:53challenge here and the risk is going to
- 31:55be if
- 31:56um we don't do that and we you know we
- 31:59have things like
- 32:01you know people using their you know raw
- 32:03pan their credit card number directly
- 32:06with agents or
- 32:09>> [clears throat]
- 32:09>> users trying to use an agent with their
- 32:11own username password um trying to
- 32:14access their bank and doing like two off
- 32:17with their own phone. That those types
- 32:20of actions will um
- 32:22will create a lot of problems when it
- 32:25comes to fraud and when it comes to
- 32:26security.
- 32:28So it it's more about actually
- 32:32you know the industry moving quickly
- 32:33enough to be able to provide the right
- 32:37tools
- 32:38for both agent developers
- 32:41and also enterprises to be able
- 32:44to do this much more securely.
- 32:47>> Right. I mean, I really enjoyed that
- 32:49conversation. Thank you for coming on
- 32:50the show and sharing these insights. I
- 32:52hope it was useful for my audience as
- 32:54well. For everyone listening to this
- 32:55episode, don't forget to share it with
- 32:57your friends and everyone interested in
- 32:59this and don't forget to subscribe.
- 33:01Thank you very much.
- 33:03>> Thanks, Sam.
About this transcript
This page contains the full transcript of How AI agents will change payments, commerce, and fraud - Amir Sarhangi (CEO, Skyfire) by Wrap Up, generated from the public captions YouTube serves with the video. The transcript has 5,371 words across 878 segments, with the original timestamps preserved so you can click any line to jump to that moment in the embedded player.
What you can do with it
Use the transcript to take notes, quote the speaker, build a study guide, generate a summary with ChatGPT or Claude via the YouTube Summary tool, or export it as a timed subtitle file with YouTube to SRT. You can also re-open it in the transcriber to choose a caption track available on the video.
Free YouTube transcript tool
YouTube2Text is a free YouTube transcript generator — no signup, no daily limit. Paste any YouTube link and get the full transcript instantly, with timestamps, click-to-jump, caption language selection, AI prompts for ChatGPT, Claude, and Gemini, and exports to TXT, SRT, VTT, or Markdown.