Unscripted with Victor: Agentic AI, Fintech's Future, and the Death of the App Economy

Hello dear colleagues um friends of R136 Ventures. Um I'm here in studio Victor Arloski. This podcast Ventures from the Valley is brought to you by R136 Ventures and today we will reverse um engineer everything. So I will be a guest in the studio and my colleague Austin uh will be uh interviewing me.

Um there were a couple of uh important events in my life uh throughout this summer. I uh have um got through uh Stanford executive program which is a kind of an MBA for executive leaders, one of its kind. Um and um I'm excited to be now part of um formally a part of uh Stanford Alam. Uh so we may speak a little bit of that and obviously about AI evaluations and um everything around that.

Uh so Austin has prepared questions. We agreed that I will not see questions in advance. So to challenge uh me a little bit and um let's go ahead and um spend some time uh talking. Awesome.

Yeah, thanks for that intro Victor. So definitely want to spend some time talking about yeah your experience at with the Stanford Stanford program. Want to focus a lot on AI too. Get your perspective there.

But why don't we go ahead and dive right in. So with your background, you've worked as a developer, a bank tech executive, and now a growth investor. So which of those identities would you say has most shaped how you assess AI companies today? Well, it's a it's a tricky question indeed.

I think that everything has shaped it a little bit. Um and um I delivered a uh a TED talk a kind of a TED talk um while I was to uh SAP to the Stanford executive program. Uh and I referred to uh my uh past experience in building. Uh so when I built something uh it was like long ago and I built one of the first mobile banks in the world if not the first one.

Prior to iPhone era, we built almost a million uh consumers uh in uh what is what was then called featured phone or SMS banking. Um then we built um a super app um and uh reached almost um 60 million uh I think even more than 60 million consumers with almost 20 million daily active. Uh and uh that was uh uh largest uh deployment of digital bank back in 2014. Uh it was bigger back back then than revolute today.

uh I think uh and uh it was bigger than all the US bank combined uh in digital uh space back in 2014. So um uh we didn't call it uh super app though uh because there was no name for that likewise there was no name for mobile bank when I built a mobile bank. Uh but um I do have a lot uh from that time a lot of like technical specs and I was wondering um what if I u uh take this technical specs back to life um how long will it take to build what I was building back then. So I took some of the technical specs and I had uh like the standard procedure for them back in 2001 and then back in 2009.

So I asked um a few development teams to uh make a proposal. What if I'm going to build it now? How long it will will take? So just to compare apples to apples um uh back in 2001 uh text packs uh which we um uh built and uh we got like competitive offers for like building the entire product was say around $3 million and um around u 30 people um and three years to build.

So uh today I'm not exaggerating uh I got proposals for the same thing uh uh from bunch of teams to build it in less than a week and less than uh $10,000. So that uh what the acceleration of progress is all about. So from 2001 $3 million um uh three years and 30 engineers to less than a week um less than $10,000 altogether uh and one or two engineers um that how easy it is to build now. So my experience shows that building now is um ever uh less expensive and easier than uh any time before and you uh have all this like u quotes from Habis and San Malman and others that building today is easy uh that you have a PhD in your pocket.

Um I uh read this um as said by Mark Zuckerberg read it in full uh uh instead of just reading uh commence to that and um that what he is exploring right what that what he's saying this is a PhD in your pocket and that's right you have a PhD at least for software development like senior software developer in your pocket or bunch of those in your pocket so it's easy to build uh but uh there is something what is more and more difficult is to scale So now to go from 0 to one is easy uh while go to from one to 100 is very difficult.

Uh because there are so many people building uh so um my take away from this is uh you have to build we have to build and we have to invest into what is difficult to build instead of uh trying to um invest into something what is easy to build. Uh I think it is more important now to invest into what is difficult to build. So my perspective uh as investor um and that what uh has been shaping through AI uh transformation is that I increasingly more concerned about how easy it is it to build everything around yourself.

So I uh wonder what are the most difficult problems you can pick uh to build upon and that what uh is building up as investment thesis for our next fund. So we uh tackle four very specific areas um in uh business which we think is very difficult to build upon. Amazing. Yeah.

And that kind of answers the question I was going to ask next, which is what's the most important thing that kind of leads into the next question of what's the single most important thing an AI founder should know before taking let's say latest stage capital um that most of them find out too late. You know, given your given RO136's focus on on late stage, what do you think is the single most important thing an AI founder should know before taking late stage capital? Well um I don't think that there are that many companies now that got to a late stage uh in terms of revenue.

So I mean don't mix up valuations. I mean there are many companies which are billion dollars plus in valuations. It's not any any any more a unicorn that we should make a different name for that because uh you could not um have so many unicorns. So I would argue that there must be a different name uh for companies which are 1 billion plus maybe like 100 billion plus are still unicorns but don't mix up lane stage with valations.

Flame stage is where you are ready uh to either go public or be a matured company, right? You can stay private as long as you want but uh you have to be uh cash profitable. Uh you have to u grow profitably and continue to grow profitably. Um and that what my clarific classification of the late stage uh uh it was like that it is like that.

So u obviously um uh if you're talking about like late stage companies first of all if you are in AI native space and you are say three $400 million in revenue already and you um um uh your gross margin is above 60% which is healthy and uh uh you are close to break first of all congratulations with that. So I think it is uh like increasingly more important for you uh to uh stay uh AI native uh and it means that you have to capture uh to create not to capture to create new value every time.

Um so uh things are pivoting so fast. Um people are going to uh um catch up uh who is like behind you so fast that you have to constantly innovate and see what's like behind the corner. So I would not uh stop there innovating right. So five years years ago I would say you are done all good.

You got to half a billion dollars revenue. All good. you can now think of protecting yourself and at some point you have to start protecting defending your uh field your uh domain rather than just growing aggressively right because it is also important u so um I think that uh my advice for a late stage founder in that respect is to uh keep going keep innovating like you are a seedstage company so um think of like pivoting uh killing some of your products uh making um um bets which are more about conviction rather than consensus and keep treating yourself as a seedstage company.

So I would probably give one single advice to a late stage AI native company um you you have not won yet and we see it um on even companies which are way bigger than uh hundreds of millions of dollars in revenue. Think of Antropic, think of OpenAI. Uh these companies have got to uh dozen of billions uh close to maybe like 60 70 billion dollars in revenue. Yet these companies are not winners yet are not winners.

Why? Because the competition is increasing, right? So competitization of the product of technology is becoming uh faster is coming faster and faster. So my take is that either you keep going, keep innovating, keep building new stuff or you uh go out of the market quite um easily.

So that's my take. So you know you're a growth investor. You talk a good amount about entry price discipline, right? So in in Q1 of 2026, we saw that 80% of global venture dollars went to AI companies.

Open AI alone, which you alluded to, raised, you know, what what I think was $122 billion in one quarter. So do you think that AI can genuinely be an industrial revolution and still be a bad investment at current prices? Because those are kind of two different questions. It will and it is an industrial revolution for sure.

I think that's uh out of question whether uh it is a new u a huge opportunity that's out of question. It is right both in uh remote uh uh workloads u like open AI entropic or take like open uh source models u that is like still an infrastructure play uh and ending up with the real world right robotics and all this stuff so uh it's definitely going to transform the way we live the way we work and I think the way we uh learn and the way we think uh it will definitely going to be a huge transformation and this transformation is going crazy fast.

So uh uh is valuation right? That's a another question right. So uh I u believe that u what is now uh going on in infrastructure is definitely uh overhyped and uh overvalued. uh it's not because of there is no uh demand for infrastructure going forward.

There will be demand for infrastructure being it uh cheap manufacturing uh uh bandwidth energy there will be demand for that but we all know that infrastructure is commoditizing really fast and um uh we are going through the same cycle again and again since the beginning right since the beginning of industrial revolution if you think of um uh 1800s mid of 1800s uh there was nothing to compare fair with since then with uh um railroad uh boom and rail railroad hype. Uh did we stop traveling?

So we didn't stop traveling. We travel more and more. Goods travel more and more. People travel more and more.

Do we still use railroads? No. Right? So uh it's not like about travel is not important.

It is just about a pivot in technology. Right? Rail roads are not important anymore. Uh although they exist, right?

And there are still a lot of goods being and people being traveled in different parts of the world through train. But uh railroads is not a hype anymore. Why? Because infrastructure is like always like that.

So u being at uh computer infrastructure or transport infrastructure doesn't really matter. Infrastructure is not going to survive uh the downturn and the infrastructure is not going to be uh valued high when the cycle goes to an application layer and what will uh value high is uh applications is where businesses where uh uh consumers are using the tools applied to different specific domains uh and that where I think the most value will be created in the next decade. Uh so um we are investing in applications in enterprise applications.

Uh and we believe that the most value will be created in enterprise applications rather than in infrastructure. So uh from that perspectives I think um and it is like a question whether entropic openai and deepseek is an infrastructure or not right if it is just a model then it is an infrastructure. We built our own application called uh secretus. So secretus is an application um for R136.

We built it with the help of a bunch of engineers and ourselves. Um it's amazing tool right what is uh behind the scene like a lot of models we started using uh uh entropic and then we switched to uh open source models and now we're experimenting using some models for some workloads and other models for other workloads like intelligence is like layered now into different domains. You have um audio, you have text, you have uh uh um mathematical formulas, u you have u um uh drugs, you have something else, right?

And all this um like different uh intelligence is being used for different uh type of domains. But also different intelligence is used for different um complexities, right? So there is uh like fleet and uh like most advanced models and there are cheaper models. So what we are using like we are experimenting with those models and we apply the best model which is currently available to the particular workload optimizing the cost optimizing the quality optimizing uh time uh uh uh to uh delivery.

Uh so um we are quite agnostic to models and uh what I noticed is that by changing from one model to another we are not losing anything. So switching cost is amazingly easy. It is as easy as I would go with my phone from T-Mo to um AT&T, right? I mean, just I just need to spend like half an hour, maybe a little bit more and then I I'm with my number with everything, my contacts and stuff.

I'm now not um a T-Mobile user anymore, but AT&T user or Verizon user. So, it's I mean the switching cost for models is even easier because I do not need to do anything. uh I I just have an agent um which is kind of orchestrating managing other agents and then this agent is choosing which model to use. If tomorrow there will be a cheaper and better model it will just switch easily without me noticing it.

So um um for that reason I consider uh this foundational models LLMs to be an infrastructure rather than to be an end user application and obviously they can migrate to become end user applications and they want to be end user applications and they they are working hard to become end user applications being it enterprise application or consumer application. I think that u they both are doing a good job but it doesn't mean that they will win right because many others do also good job.

So um I believe that if you be if you stay as an infrastructure you're not like you losing big but um who knows who is your electricity provider or your water provider right I mean you don't care as soon as you have water uh in your sink as soon as you have electricity in your socket you don't care right what you care is an end user application whether you are using a Mac uh or whether you're using Windows uh computer or uh like iPhone or Google phone or whatever right that what is important important.

So and that what creates a mode. Uh so infrastructure doesn't create a mode. Uh and it could be a profitable business, right? Revenue generating business.

But we would be a business which is one of the top 10 largest companies in the world. No way. So you need to create an application to that. Uh and all seven companies which we know like magnificent seven they are application companies rather than infrastructure.

So they get to an end user right end user could not survive without those seven and that's the reason why they exist on a top uh ranked uh um top 10 uh companies in the world. So you've also said that the app economy is dying. So that AI agents are will select vendors through APIs based on speed, price, reliability, but not on brand or user interface. Right?

So if that's true, who owns the customer relationship when the customer's agent and the merchants agent are negotiating with each other with each other? Yeah. Yeah. I'm big I'm a big um uh believer in the end of the applications in a way we know them when talking say um uh banking uh like new era of banking um when I build banks there was like featured phone bank mobile only and then there was uh application in like like iOS Droid bank uh uh digital only uh now if I built something I would neither build anything related to an application, right?

You do not need an application to be a digital uh u financial advisor or a digital financial u assistant uh to your customer, right? You just need a fleet of agents, right? And this fleet of agents could be appless, right? You do not need an app.

Uh so I think that uh everything is like streamlining towards like no apps but agents and agents need no UI, right? So we were all like I mean take like five banks um um whatever random banks they will all do the same but they will have different UI right some UI is easier simpler some is not so this is all a UI play right there is no any single differentiation between banks for a consumer uh apart from like UI right so UI is not going to be differentiation in AI new AI native era What is going to be intelligence, right?

Intelligence and connections, right? How connected your agent is, how smart it is, and how well it performs your uh orders, right? Your tasks. Uh and obviously when we speak about um AI assistant shopping assistant, it doesn't really know your preferences, right?

you go Amazon or you go Expedia just because you got used to uh like your habits right to buy that very way right it's very challenging to change your habits right you already got like a good experience buying on Amazon so keep going Amazon I have got a good experience buying um um eBay or buying elsewhere I will keep going right so user experience is important um because we really value our habits, right? We want uh to spend less time, less energy on to dealing with UI, right?

That's the all UI is all about. Less steps uh uh higher convenience um uh better transparency, one to click go. Right? Now, imagine that AI agent doesn't have this habits, right?

It doesn't have habits. it goes like instantly to 10 websites, 10 applications or even 50 applications and it chooses whatever. So it's not like optimizing for time, optimizing for habits, for low energy. It is optimizing for best result.

So um I think that that what is changing really in uh this AI native uh era is that there will be no like need for applications and it's a good question how this this agents are going to speak to each other right how I mean if I have an agent you have an agent uh so you are a seller I'm a buyer so two agents speak to each other um definitely there will be less bargaining power on consumer uh uh definitely um like advertising will change right from being irrational and that what the whole economy is all about like you should love a brand you should trust a brand there is no anymore any love and trust it's just a completely rational thing uh so I believe that one of uh u the directions is this rationalization of uh like interfaces u it's it's more AP API rather than anything else and it means that uh brands will lose uh bargaining power on consumers, right?

Uh on the other end, AI embeds a lot of emotional intelligence. Why I like uh Antropic more than uh uh uh I mean cloud more than uh chat GPT. I use more cloud than Chad GPT and some of my family members use more Chad GPT than cloud. I mean from intelligence perspectives it's just the same.

I mean you could not really distinguish. It's like both very intelligent. What I like more in Claude is just uh emotions like I mean it's it's emotionally it's more connected to me but my family me members argue they say oh we like more chhat GPT because it just emotionally different right so I think that emotions is the first time ever can be embedded into software so I call it a software with a soul and I think that uh there are two different ways you can treat AI so one it's like completely rational abstracted from user abstracted from any like I love it any or I I I trust it because you don't care right because that what AI agent will decide and there is another extreme where apps should become your friend like if I was like a shopping assistant I would just be not just a rational shopping assistant I would become like a superhum superhuman intelligent listening um uh and supporting yourself and encouraging yourself and being passion and being funny and being whatever you want it to be, right?

And um um I think if you want to differentiate in AI area, I would choose this differentiation. I mean either you are like very technical and like go this like high-end uh like shopping assistant with no soul uh and no appearance whatsoever or I become like super super emotional um uh so it's your choice basically as a vendor which one to follow but ideally you have both right ideally you have like a super intelligent uh being emotional being and also So underneath like a super uh intelligent and non-emotional chooser of like the best product or service for your um end consumer but I anticipate that there will be not one assistant.

So people think of like oh there will be like one single thing so the winner takes it all. Winner is not going to take it all. Uh in my opinion there will be bunch of different uh use cases and you will be surrounded by fleet. You will be me.

You will be surrounded by fleet of different agents uh which will uh do specific things better than others. Right? Some will act as your legal advisor, another will be your shopping assistant. Uh third will be your fund manager, your entertainment manager, uh your travel uh uh assistant and whatever, right?

Because specialization is natural, right? We think of ourselves to be like super specious. We are not. We are very specialized, very very specialized.

Although although we think of ourself like oh we are like human beings on top of this uh um uh value chain and uh food chain we are far from that right and we are not the biggest by any like mean uh uh species on the planet. Uh not by number, not by weight uh commitive weight. We are one of many and we are very specialized. Uh and there are millions and millions of others like starting from like microorganisms ending up with like u um species which live into like different uh like oceans and uh in the air and whatever right.

So I think of intelligence to be the same. So whether it's like it's a different form factor right it's a different form factor. We are uh made of something and super intelligence is made of something else. But the way evolution will work is the same.

you either survive through specialization or we go instinct. You could not be too generic. So that my kind of philosophical take from where AI is heading. So I'd be curious how you think this plays out in categories too where where taste matters more than maybe pure optimization because in areas like fashion, travel, media, the best result can be pretty subjective, right?

So how do you kind of think about that with with AI being in in different categories? Well, we I mean I don't see any problem with that. I mean we are subjective. Uh so um I think that uh if um there is a perfect match uh uh that uh I'm subjective I have some point of view I have some physicists and then my fleet of agents uh have a perfect match why not I mean why should that be a problem right um um I mean there is also such thing like hallucination uh which is I wouldn't call it like subjective view it's just something what uh uh basically AI models are missing right I mean we are hallucinating a lot more than AI frankly like I mean our brain is hallucinating all the way because it's projecting future it's reflecting on past so I mean our brain is a machine a predictive machine and if you think that it is a perfect predictive machine I mean so far it is absolutely not right it's completely irrational it's uh like I And it's it's I mean the level of predictions we make around like math statistics is amazingly bad.

Um we are much more intuitive right uh so we make really good intuitive predictions on emotional side right so I know that you are in a more or less a good mood now by looking at you right and you might think otherwise in looking at me and that what models could not do but models are a lot better give a lot better predictions on so many things so I think that from that perspective like a combination of a human with its own predictive model and an AI agent or fleet of agents with their own predictive models could just be a perfect match.

So I definitely want to push more into where the value of AI is actually landing. You know what's working, what's theater, those kinds of things. So here here's a stat I read that I found kind of striking. 88% of companies say they use AI but only about 6% are showing meaningful IBIDA impact on IBIDA.

So that that gap doesn't close on its own, right? So where do you think the actual bottleneck is? Is it data infrastructure, governance, workflow redesign or or maybe something else? Well, the bottleneck uh is everywhere everywhere you mentioned.

I mean um uh there is no clear model like the best use case, right? Uh we all like see this uh uh token maxing, right? uh this phenomenon uh phenomenon uh which is u uh really like unexplainable right so u uh from what I remember mata has u um I'm not sure if that's like validated uh uh information but that what I've seen somewhere that matter has spent like almost a billion dollars in just one month of April on uh uh on um uh different type of tokens mostly anthropic topic tokens uh clo uh so it's in in um AR it means like 12 billion right a month uh of April was like a almost billion dollars spent u is it healthy or not right um uh how much value was it created there with this uh bird and um uh we see like companies are shifting from get more of AI and then get less of the of AI and then get more of AI and and fire all like lay off all the junior uh uh software uh developers orders and then take them back.

Right? So uh why is that? Uh so this technology is evolving so fast that there is no one clear uh good strategy. So uh companies are experimenting in different ways.

Um I think uh it is okay right? I mean you should we should we all should experiment right? experimentation is the way to find what's the optimum right and what was the optimum in April could not be an optimum in May anymore right because technology is evolving so fast uh I think that um one of the biggest bottlenecks is for sure um uh the way companies are organized uh and the [clears throat] way uh um people are organized within these companies uh I think that um a good deploy deployment of uh AI especially in large companies requires even more even high level of autonomy uh if we're talking about like two pizza teams now it's like half a pizza team uh with less hierarchy and with more abilities of experimenting on a lower level right you could not uh dictate like a CEO oh we should we should use more of AI or we should lose use less of AI.

Every team member, every small uh component of it should decide on how best to use uh different AI tools uh into whatever they do, right? And obviously uh it has to be measured but again uh you could not measure result on just few weeks of use right or even few months of use. You should give time like for anything else you should give time uh uh to like this new ways of working to mature in a way. So um uh what I would argue uh large companies should do is uh to ease up organizational structures uh to give more freedom for people to decide uh built on their own.

Uh I believe that uh a new type of organization is uh this um uh creators organization if you will. We were all witnessing uh creators economy right now. Creators economy is knocking a door of large enterprise. Now creators uh are not independent uh freelancers but uh freelancers are within the organization.

So each organization by itself should be kind of a freelance organization within itself. So uh I would say that uh the biggest obstacle also uh middle layer managers who neither do anything like by hands like doing a real stuff uh nor uh having enough power to make decisions, right? Uh there is a middle layer in every company like people who neither have a full authority to decide because they have a higher boss somebody whom they report to but they are not like doing things themselves right.

They are more like okay we prepare uh like um uh decision for a high management to decide and then high management decides and then we communicate whatever high manager decides towards like lower level to execute. So this level is broken. So I think that if you think of like doing something good for organization take this away. I would just take it completely away.

So you are either having a full authority to decide or you do things on your own. You build. If you're not a builder and if you're not a real decision maker, you don't we don't need you. Um so I think that by eliminating this lay layer uh organizations can uh uh achieve a lot.

Uh and I I would argue that this is must have for every organization to think uh uh of like delaying in a way. So you should like become as lean as possible from decision maker to uh a builder. So there should not be a intermediate layer in between and most companies have like multiple layers. Uh so that what I think should change in an organization of the future.

Yeah, I keep thinking that the hard part isn't proving that AI can do a task, right? But it's rebuilding the team and the process around it. So those gains have a so the gains that come from AI can show up at the P&L level. And they will.

They will. Mhm. So uh I'm sure they will. Uh I think that um it is there already.

Uh but the thing is that um uh there are still like early signs of that. I just I can only speak about my um Mhm. role in the company and my own company R136. There are like 13 14 people now in in the company.

Uh I'm going to continue with this team. I'm not going to lay off because there is no middle layer. There are either people who building or people who making decisions. But what I'm trying to embed is that people who make decisions now building.

I'm building myself. I have a lot of stuff and I actually I was thinking uh okay I mean I'm not like AI literate. Uh I mean I was not thinking of myself to be like AI expert but I see like how mistaken I am is I like think that I'm not doing much in AI but as we speak now we have like I have 20 agents doing something uh while we speak. So when we finish I will go and check what they're doing and give them new instructions uh and that what my other team members are doing as well.

Now, uh, am I losing my job? Is my, uh, senior analyst losing his job? No way. Am I working now harder than ever before?

Yes, I am. Uh, is my intensity has my intensity increased? I think in one year it increased like fivefold. Five times.

I mean, five times. uh I produce more at least five times more as we as organization I think produce 10x from what we were able to produce a year ago because we can now work on 50 projects simultaneously. Our fleet of agents do a lot of work on their own, right? But they come up faster for my decision than I can imagine can have can imagine like a year ago, right?

They bring me something big picture. Now you have to make this decision, that decision, that decision, that decision. So I'm deciding on much bigger things now than I and much more more things now than I decided a year ago. And same around like all others.

Right now iteration takes so little time. I'm talking to like my senior analyst uh five times a day today uh on something what I would have think of like taking like two weeks right the iteration is now one day we are building a project so we have like imagined something we're building a project in hours what will take what what would what what would have taken like two weeks to build or even more so my intent ensity has increased and I think that it was like always the same, right?

Every piece of technology has increased my intensity. Uh I started working when we barely had emails, barely had emails. And I remember coming to um u I was working for a bank um uh which had a subsidiary which was Dutch bank long ago in the '9s, mid '90s. uh a subsidiary uh in Switzerland and I was asked to come and they had like a problem with the printer uh in their headquarters.

It was a small boutique in Switzerland, an investment bank and I came to like as in as engineer I came to fix this problem like uh of this bank uh printer has been broken over to it was unable to print and then um it was a software bug in a potawa. So I fixed this bug. It was easy. And then a printer started printing this uh backlog queue uh of some tasks.

And I got all this paper which was not reports. Uh it was just bunch of emails, emails printed. And then I came to a manager saying look I mean there was like something being printed. Oh, he said, "Oh, we couldn't really work uh for like a week or so while printer was uh uh not fixed."

Thank you so much, Victor. And I asked, "What are you doing with the all this? Why are you printing emails?" He said, "How how else we can answer those emails that how are you doing it?"

I mean, I was curious. He said, "Well, we are meeting every week on Monday and we have these emails on the table and then we read those emails and put like instructions to our assistant in in handwriting and then assistant puts it on to like boom boom boom because we could not type. I mean all this like keyboards. I mean it was like new, right?

It was 1995 and Swiss bankers were not like using keyboard. It was too difficult. So it was they call it a mail party. A mail party.

Every morning two hours they were printing up all the emails in one week collected and then handwritten instructions how to answer and then assistant was doing the work. That what I what I started with back in 1995. It was still like okay I mean is it true? I mean because I was obviously literate typing blah blah blah.

It was already I already had a computer back then. But that how people used internet and that how people used emails and it reminds me how people use AI now. Why I'm telling this funny story like almost anecdotal story is just because that how people use AI now they just with this mentality it's like mentality of those people in 1995 uh but with now new technology. So and that's one of the reasons why it is not uh flying as high as it should just because the mentality is changing so slowly like I mean we are not adopted to change fast as a as a human beings.

So where do you think the best early sign, you know, what do you think the best early sign is that AI is creating value inside a team? Uh you can do more uh you can expand faster, you can build faster. I think that is the most important trigger. Uh it depends on what you're building, right?

You can build a crap faster. You can build amazing things faster. Uh but now you can do things faster. It's like you were riding a horse and then you were riding uh you were driving a car and then you are uh sitting in a super uh like rocket which can get you to wherever you want in seconds where like car would get you to like in days and horse will do the same in years.

So imagine it that way. So now like if you do go faster to the wrong direction, you'd get nowhere. But this thing is definitely going to build a lot faster. Whatever you want to build, it will do things 10 times faster than 3 years ago.

And we see some progress like I mean I recently I'm very interested in like biology uh this uh um um reading this um uh proteins and understanding uh uh protein uh uh molecules is like was a big thing right we only understood like 5% of that three years ago what LLM allowed us to do is to understand 95% of them now uh those proteins uh like what is the shape? What's like the molecular look like? How it interacts with other moleculars? Uh in three years without LLMs, the projection was like another 30 years.

Uh AI allows you to build faster. Now you have to cope with the speed because when it is so fast, you have to choose what to build. Uh and uh if you start building the crap, you will build a lot of crap, right? Very fast.

What you need to do is to choose what you exactly want to build and that's again the reason why we think uh it's time to invest into what was very difficult to build back back like few years ago and which is still remaining quite difficult to build because that's but that what creates the mode. You still do it like 10 times faster but um you you do something what others could not do right using this the same very tools and technology. [snorts] So, where have you seen people still hold on to the old way of working even when these AI tools are clearly better?

Where are you seeing people kind of have a hard time um adopting this new way of working with AI? I think that most of most of people most of people being it employees or for their personal use they use AI as a as a um basically question answer tool. Answer me this, answer me that. Uh it's like a chatbot on steroids.

So it answers your questions but again you can have a conversation right conversation is good but you should use this tool to build and we can you can build I can build right we can build bunch of things I uh was to SCP um uh SCP um um course I mean the Stanford executive program uh through summer so um We had a bunch of um challenges uh I mean um say uh a negotiation uh uh uh challenge where you are selling me something I'm buying and then there is a use case like the whole use case and then you should optimize for selling me or or hiring me and I should like get better um conditions.

Uh so uh what I was doing always and it's it was permitted. I was just building a tool every time when I was like I had only five minutes to prepare and then you and I meet and you sell me something I buy. I always came up prepared with a tool because I was building an app on a spot and I'm coming to negotiate with you with already prepared tool. Uh you have a list and I have a tool.

who is a better negotiator because I can now play with numbers um on a fly. So I so much got used to building that I built like few uh basically apps in a week. Uh if I had more time I would build a few in a month a few in a day. Uh uh recently um my wife and I just discussed like uh like Straa uh for wine lovers.

Uh imagine like I open Straa and um uh I see you biking or you you write your whatever running uh or whatever right doing whatever sport. Think of like the same for wine lovers. What have Austin uh uh got yesterday for his dinner? What kind of why?

And uh did he like it? Yes. No. And what was like his pairing and stuff like this?

So it like straa like it's a social interaction for wine lovers. And I quickly found out that there is no app for that. So I built an app like few hours and the app is there. Now we have a wine lovers app like Straa.

So um my friends can enjoy and post and then I see what they're drinking, what kind of wine and what they pairing it with and whether they like it or not. Right? Um it's like Straa but for wine lovers. Uh I I didn't know I didn't need anyone to build it.

I just built it myself in few hours. So that's the way I think about building, right? Uh and who knows this app could become a billion dollars app. I mean I I'm not I'm not uh chasing it.

I'm not hunting for that. But um if I just because I'm not a wine lover. I'm I I I'm fond of sports. I'm not fond of.

But if you may think of like somebody who really loves wine, he can build this uh uh at scale. And that what I'm calling everyone to do like keep keep start or keep building as much as you can. So I want to switch gears and maybe talk a little bit about fintech uh in the fintech side of side of things. So, [snorts] so Vint investment recovered 116 billion in 2025, but deal count fell to an 8-year low.

Do you see that as recovery or consolidation? [snorts] Well, it's definitely uh consolidation in a way, but uh it is amazingly difficult to uh raise money now for fintech. uh if it is not like AI play uh and I have bunch of friends who built really good uh fintech uh uh products uh people do not really invest into early stage fintech uh there is no appetite I don't want to name a few uh fintech one of the best fintech VCs on the planet and I spoke to founding partners of those they try to switch to something else because fintech is not and they predominantly were doing fintech only nothing than fintech and they're now doing something else apart from fintech uh and um I'm curious so u I think uh it's far from being uh a true statement but u I also like witnessed a dialogue like on a board of company which was in credit tech um that we should tell company.

Why? Because like investors do not believe that fintech like credit tech will uh be on demand uh in 5 years and the reason is that uh people will lose their jobs so they will not take loans anymore because there will be no need for loans uh and and uh like I mean people will not be able to repay those loans. So uh why should we invest into something what will not exist uh in 5 years from now? Um so I am not uh supporting this argument.

I think that people will continue to borrow um they will continue to uh have like to take mortgages uh to pay back mortgages. Uh and uh I'm not uh supporting this view that fintech will collapse because AI will take over. Uh I think that there are amazing opportunities in fintech u uh related to stable coins, crypto payments. Uh I think that um aic e-commerce will uh change the landscape of payments quite tremendously.

what was called uh back 10 years ago embedded finance and we see no good cases in embedded finance uh I mean no like real like big companies in embedded finance uh finance and uh payments remain to be silos so I think that embedded finance and embedded payments is uh must have in agentic uh uh e-commerce and and agentic quote uh nobody is ready now none of the infrastructure pre uh payment platforms are is ready to even like Visa and Mastercards and name it uh Stripe is ready to adopt um millions of agents or even billions of agents or maybe hundreds of billions of agents that constantly iterate, negotiate and pay each other, right?

Uh so we deal with completely different environment where people pay people companies or companies pay people. It's all slow like compared to I mean we think it's cutting edge. It's all slow compared to imagine trillion of agents. Each has it its own account.

Each has its own limits. Each has its own patterns. Each has to communicate and make decisions on the fly in microsconds and pay in microsconds. otherwise it doesn't fly.

I mean none of the infrastructure is ready for that and somebody has to build it and it will be times bigger than Visa, Mastercard and Stripe combined. Uh so we will see a winner there, right? We will see someone who is going to build it. Uh so I'm a big believer in fintech on that front.

I think that there is a huge opportunity now uh to disrupt uh what was called new banks, what was called digital banks, to disrupt those banks and to disrupt payments uh in a huge way. I think that's the way to go in fintech. So, do you see Agentic Commerce strengthening a Stripe or a Visa or do you think it'll allow agents to kind of route around them? It depends.

and depends on whether uh uh those companies you mentioned or others will catch up and build. Uh I think well I think that again I mean think of like Visa and Mastercard. You need a person why Visa and Mastercard are so successful because they built on banks. they uh uh made alliances with banks and banks were distribution channels for Visa and Mastercard across the globe.

The beauty of Visa and Mastercard is the infrastructure they built. Uh you can come to remote place in uh uh South Africa in like South America, Asia and still pay Visa, Mastercard because there will be one single bank there with its US or ATM or whatever. you can either withdraw money or pay, right? It's just an infrastructure being built.

Now, with the agents, it could just be all different, right? You don't need the same infrastructure. You just need a different infrastructure. Uh and um uh whether Visa and Mastercard are going to be this different in infrastructure uh they will manage to build or there will be a new company whatever the name is that will build this infrastructure.

I don't know. Right? But uh opportunities is equal. So I think that uh to build another infrastructure for people is difficult.

We see China has built it with all this QR codes uh without a card. You have a QR code. What a QR code. Why it was built there?

Because they were completely underbanked. So there were no banking cards, right? And that was the reason why build this why they build this QR codes. Now yes sorry I have to run to another U session.

Um anyways yeah I have I have another call. Um so I will I will I will we I think that we have to end up in like five minutes. Okay. I I had one really one last question.

Absolutely. So let me let me finish here. So um yeah. So um and that's the reason why uh one country has adopted QR codes and maybe some other Asian countries and others did not because banking card was in every pocket, right?

POS was in every shop uh and and ATM was on every corner. So it was really hard to beat. Now when you have a switch to agents uh opportunities are equal uh for those who built from scratch and those who built upon visa masterard or strike. So that's the reason why I think uh the winner is not decided there.

So last question just to wrap everything up here. Let's say it's August 2028 you know two years from now. What do you want the record to show about the decisions R136 made in 2026? Uh we are extremely price disciplined.

Uh nowadays uh we have learned through 2021 that uh this uh hypes end and uh I think that this hype will end. Uh I mean hype of um um uh every price is good and every price is justifiable. uh we are quite disciplined in uh uh choosing application over infrastructure and I think that um I would wish this thesis to win in uh 2028 that being price disciplined uh price sensitive and uh investing in application rather than infrastructure will get us to uh a success. um through 2026 and 2027.

Uh I think that um we will be 10x more efficient than we are now. Even we are 10x more efficient than we were a year ago. Uh and we will uh keep building our own stuff um uh experimenting and uh uh improving our uh efficiency uh doing more with less resource, doing more faster and adopting uh and um I think that uh our um conviction with people apart from everything we invest in people and we built trusted relationship with both our LPs and uh our uh uh managers, our uh startup founders uh will continue to be uh I'm always like people are asking me I'm always asked uh whether machine is making uh decisions uh in R136 many decisions are made by machine.

The only decision we make ourself is all about people. Is this the right team to invest in? Are this the right people to trust? And we are not outsourcing our relationship with LPS as well.

So our relationship with LPS and relationship within the team stay uh and uh I am a big uh uh supporter of uh long-term and deep relationship uh and you are part of the team. I hope you feel it. Uh so uh uh we try to uh to build it um as a human company uh where human beings are essential most important part of the exercise not AI and it will continue to be and I think it is the same for our founders. Uh people are more important than ever before.

technology is important but people who are building behind those computers behind those LLMs and infrastructure and what have you is even more important than it was like few years ago and in 2028 it will be even more important than today. uh so uh uh we will continue to invest in people more than into anything else being as I said very disciplined in our choice. Awesome. Amazing.

Well, thank you so much for sharing your perspectives and your insights on AI on latestage investing, growth stage investing and um fintech and and everything in between. Thank you very much Austin being part of R136. Um and we will continue to have amazing talks with a bunch of founders whom we are planning to speak with uh in coming uh weeks. So you will uh see a lot of interesting uh uh podcasts uh ventures uh from the valley will continue to focus on people who are building and thanks for being with us.

Um um this was Victor Alosski and uh Austin Leglair. Uh and this podcast Ventures from the Valley was brought to you by R136 Ventures. Stay with us.