How AI Agents Are Scaling in the New Agentic Economy | Alan, Bryan, Patrick, Lei

our moderator today, Lelay from Kite, and she is going to introduce the rest of the panel and start our discussion.

Awesome. Thank you guys so much. It's really excite exciting to be at Denver again. Um, I'm Lelay. I'm VP and head of BD at Kite.

We are focused on being an AI payments blockchain um and focused on the Agentic uh native uh infrastructure for the agent economy. So, I'm really excited to have some of the top investors in the AI space with me today. Um, so why don't we go down the list and uh if you guys could introduce yourself, where you work and also tell us a fun fact.

Oh, fun fact. Okay. Hey everyone, can you guys hear me? Cool. It's Brian D, a principal at Alumni Ventures.

Uh, we are one of the most active VC in the world. Invested over 1500 companies. Uh my focus is AI and blockchain. Um we're also one of uh investors backed Kite AI. Uh my fun fact is that I'm currently building a open cloud agent to help me document all the interactions I have with uh this week at East Denver.

So maybe by the end of this week, my AI agents knows more people than I do.

Hi, thanks for having me. Alan D. I'm a partner at M12, which is Microsoft's venture fund. Uh we're not a current investor in uh in Kite AI, but a little bit of history. I spent the last six years working at PayPal Ventures where I did lead uh Kite AI series A uh and uh it's great to see the growth of Kiteai uh over the years.

And fun fact, uh I used to work with work with Brian here at PayPal,

PayPal Mafia.

Hello everyone. Uh Patrick uh I run a fund called Dispersion Capital and it's a fund that's uh exclusively focused on decentralized infrastructure. So that being said, our biggest uh thesis that we had was that edge computing and blockchain were going to converge and empower decentralized AI. We were pitching this four five years ago and now it's really coming about. But uh background on me a 20-year VC exclusively always focused on infrastructure uh mostly developer facing um but yeah know was one of the very first investors in kite AI but you know before that predecessor Zeta block so excited to see the the growth and the the progress here.

Um, I guess fun fact about me, uh, and actually probably maybe actually relevant for all crypto is, um, I was subpoenaed by the DOJ to, um, um, kind of go against, uh, Google. I was supposed to testify against Google's antitrust case. So, it was f it was fun. And I actually, uh, went on the case after Satya Nadella. So, that was my little claim to fame.

Wow. Okay. I didn't know most of those fun facts. Um, all right. So, let's get started.

Um, as investors, you know, as you guys look to deploy capital, um, in the AI sector, are there cert specific verticals or parts of the stack that you're really excited about and increasingly focused on?

Yeah, sure.

Well, I can start. Um overall in the agentics uh agentic commerce space where you guys have been hear a lot of uh talking about this recently uh specifically in this space I really got excited about the humanto agent interface so more specifically how does human agent uh how does AI agent understand the dynamic human attention I will give you example so let's say I want to buy a ticket a flight ticket I tell my a I tell my AI agent to buy a econom class ticket to LA under $100 and my agents goes online, find a ticket for $98 and then bought it, drop down. But what if in reality in the world the the the first class ticket was actually sold at $15. So if I if I were browsing myself, I would definitely buy the first class because it's only $5 more over my budget. Great deal.

But my AI agent won't do that because it will follow my instruction under $100. So it get the job done but does it really uh fulfill my needs like my uh my intent? Not really. So I think the problem is that in the future when we have AI agents um doing emergency behaviors on on behalf of human we need a dynamic uh authorization. Um AI agents need to um operate autonomously within the programmable boundaries.

So that's a um area I'm really excited to see what companies are building around that. Yeah, I would say Microsoft given kind of, you know, we're a relatively large company. We do a lot of different stuff. Uh the investment horizon or spectrum if you will is quite wide. We can do things from, you know, down to the hardware level, you know, looking at new chips, uh and all the way up to, you know, things like quantum computing, but and and and also everything in between.

Uh what I would say where I'm spending most of my time uh on is uh you know as you know these days you can't really hear hear hear a conversation without hearing open claw agents this and you know AI that um as people are beginning to deploy more and more agents right right now you know Brian referred kind of you know referred to the the human agent interface but what we're uh you know where we're spending time is that in the future more and more agents going to be interacting with one another. Uh so how do you build trust between agents right in a trusted environment it's easy right it's really easy to have that agent trust handshake however if you have agents that are now going to be transacting with multiple parties or multiple relatively unknown parties how do you establish that trust so agent identity is an area that we're paying very close attention to and I do think the current identity infrastructure is very much human centric using a lot of delegated credentials however there is no agent native uh uh identity infra I guess until until kite came along introduced the agent passport. Um but uh but the other other part is well when the whole market seems to be kind of getting a little nervous around what agent might do to software companies in general. We're just talking about this offstage. Um at the end of the day it's still software that handles the data, handles the security.

And it's not going to be that different when agents get deployed to build you know very quick dashboard or build a very quick you know small thin software at the end of the day still needs to be able to access the same data sources securely. So what's happening within the database space uh is also really exciting to see because you need to take your data out of the current silos current you know your salesforce and be able to put it in a place where agents can easily access right so one interesting area it's relatively small but it's super interesting area is that you know people are now beginning to look at graph database again because you know I think someone really famous said you know the world is graph and not tables right anyways um those are the two areas where I'm uh where I'm spending time

yeah super interesting thing. I mean, you know, Allan brings up a good point, you know, areas on, you know, what enables AI to flourish. You know, that's something that we kind of look at from an infrastructure standpoint. We've been spending a lot of time looking at, you know, what is going to make this like autonomous agent ecosystem actually be autonomous? Uh Allan brought up a good point about like identity.

That's one area that we've been looking into but kind of realized maybe it's like the opposite is like we would not uh invest a dollar in identity because we just don't see money being made there but identity is going to become a big feature that is going to be necessary for agents to be able to transact to be able to be known and to be able to do enterprise uh level stuff. But that being said, we look at the full stack and kind of uh have to see where we think that uh technology uh pieces are going to need to be decentralized and other parts that don't need to be decentralized. So there's areas that we kind of see that is uh probably like more around like the middleware layer where you're going to focus around coordinate uh the coordination layer the uh uh program programmability uh component but uh that hopefully enables more uh uh pieces that enable agentto agent uh interactions um so that they can be fully autonomous. And then you know other components that we're really looking into is kind of that what we talked about on the edge side is ondevice uh AI where it's going to these smaller models going to be more efficient. They're going to leverage personal private data um and and be more a lot a lot more personalized.

So every everyone's potentially going to have their own agent on their phone or maybe even uh in on the edge. So you know areas that we've been pretty excited about. Awesome. Um, really, really interesting to hear where you guys are focused. Um, so I guess maybe let's shift gears a little because East Denver attracts some of the best developers and founders and I think many of you guys in the audience um are always you know looking to uh potentially raise capital.

Um so as a investor um what what kind of advice would you give to founders who are raising capital or what are the qualities that you that really stand out for you?

Yeah. Yeah. Happy to start. Uh you know so so I would say it doesn't even matter if it's like an AI founder or not. Um, you know, one of the key things that I personally look for for any founder is uh what what kind of problem are they are they solving and if they have some sort of unique insight or unfair advantage in solving that problem?

And the the key thing is being able to articulate that extremely well, but then also having some sort of unique insight into why. because you know it might sound very abstract but having like lived problems is one of the best ways to be that expert in solving that problem. So you know as a VC and I'm I don't know about these guys here but you know when we hear stories from entrepreneurs that are saying like hey I worked at this company I uh internally realized that you know something was very difficult and I built this in-house and I realized that you know more companies will need this product. I'm gonna leave this company and build a whole company solving this problem that I built inhouse. That's like music to my ears and that's something that you know we look for first and foremost uh before investing in anything.

My advice is be an efficient storyteller. Now with AI building things are getting easier than ever. So that means more competition, more pitch stacks, more noises. um the those founders who can be remembered by people are like are those who can tell good story who can um make who can just talk and highlight a unique advantages. So I think if a VC cannot uh explain what comp what does your company do during the IC maybe the peach doesn't land.

So my advice is talk talk short, highlight your unique advantage and try to make people remember you. And this the VC startup ecosystem is still based on attention economy. So you're going to try to grab that attention and make yourself on under the spotlight. That helps.

Yeah, I think uh the two points covered are really really great points. Maybe I'll add another one for uh companies that are a little bit further along. Uh within AI specifically, we've seen a lot of very good technical founders, you know, doing productled growth strategy. But the issue is you got to understand how to sell enterprise, right? And this is coming from someone who works at a large one of the largest enterprises in the world.

You need to be able to speak enterprise, understand what our problems are. You know, a lot of founders will come in and get super excited about the product they're building or the tech they're building. And you know they will get stopped midway by our folks whether it's on the product side or on the investment side and say hold on one second you haven't spent a single second asking me what my problems are right so focus on the problem side kind of relates back to Patrick's point understand how that process might work whether it's through hiring people who have worked uh within go to market uh organizations or talking to a lot of enterprises just understanding what that procurement cycle looks like

is important Awesome. Um, all right. Let's go deeper into each of the each of your firms. So, Brian, um, Alumni Ventures was recently ranked top 20 by time and one of the most active uh, VCs. Um, I would love to know like how do you filter out between the signal and the noise?

Um, and you know, you you mentioned that you've reviewed tons and tons of pitch decks. Um, so what stands out for you? Yeah, it's a good question. Uh, and I I want to double down uh Patrick's point. So, we review thousand of deals every months at Alumni Ventures.

Uh, one filter I really use is called vitamin versus painkiller. Vitamin is nice to have and painkiller solves the real problem. Um, there are in the agentic AI world, there are a lot of AI agent companies, early stage startups, what they are building is actually just a wrapper on the foundational model. they create marginal benefit but that doesn't solve like the fundamental problems and there are other um those are vitamin companies and there are some other um painkiller companies they are solving the problems like who can never been solved before let's take kite as example um they are building a gentic payment um infrastructure because nowadays if you are looking forward to see a gentic commerce world like today's internet is not going to work it's not ready for the aentic commerce. So that's why Kite is building a next generation payment rails and network for for the agents.

I think to me that is a perfect example of a painkiller. So yeah, that's that's my take.

Awesome. Thank you. Um Patrick, I think you mentioned infrastructure before and I I really think you're one of the most knowledgeable infrastructure investors out there. Um so if you look at the intersection between blockchain infrastructure and the Agentic space um you know where do you see the natural fit um what kind of excites you and what are some of the you know uh proudest investments besides Kite uh that you've invested in?

Um yeah I mean for for us like when we when we look at infrastructure I think history repeats itself. So when we first started looking at blockchain, we actually uh looked at looked at it as a uh analog to how the internet was built and how traditional uh stack was kind of created. And we looked at how all the all all the missing pieces that that were there and you know simple things like you know DNS like the basic networking uh phases that blockchain suffered through uh and and still suffering through you know uh that is now finally being solved with some scalability issues um but now when we look at AI and uh the agentic side um you know it's almost also uh repeating itself as well because the agent uh side is uh you know pretty pretty light uh at the moment. Um and it's it's almost you know it sure there's a lot of infrastructure around it but it's almost as if it's u its own it acts as an application. So you you need other components of infrastructure that um that it's going to have to leverage.

So, you know, we kind of I don't want to like talk about the same things, but you need you need certain components from like the compute, the data, the the identity side. There there's so many like basic needs that it's still piggy back piggybacking off of. Um but yeah, that that's something that we we kind of looked at. It's like how do these um products uh kind of relate to something that has already happened and how do you uh make these more efficient? So when you when you quantize these models and make them super efficient and put them on device, how are they going to manage like potentially keys and how you going to do like very simple basic stuff that has already happened from like the TLS's uh as well.

So um that's kind of how how we view it. Um Ellen, uh Microsoft has a uh extensive end toend strategy in AI um and has made a ton of announcements and um a lot of uh updates on Copilot, Azour and many other areas. Um so I guess uh sitting from M12, how do you think about some of the synergies or you know how do you think about where to deploy capital? Yeah, I I would say, you know, this is sort of an advantage to corporate VCs or strategics as we call them ourselves. Uh you sit you have this massive commercial muscle, right?

This massive machine that you're very close to and you're witnessing firsthand uh what's being built, what's being used and you also hear firsthand all the complaints, right? Uh so we're we're very aware where the gaps are uh and I would say that allows us to focus uh be laser focused on solving immediate problems uh that are that are very very tough for developers. For example, we're spending a lot of time looking at uh coding agent platforms. uh you know a lot of folks always point to sort of you know clock code as the gold standard but there's also a growing community around you know open- source platforms like you know open hands augment uh open code and what etc right so it's important to be aware of why developers are choosing uh to go with one platform over the other and how we can improve ourselves uh and it's also allows us to really kind of deploy capital early on right to be able to get that signal very early on uh So another one is uh you know as we witness a lot of folks whether it's within Microsoft or or the broader tech community begin to really tinker around this uh open claw uh thing we're like okay while agent economy or agent workflow is coming faster we're getting here sooner than we all thought well we better figure out security around that stuff we better figure out identity infrastructure around that stuff so that allows us to really stay focused on okay who in the current startup community are building toward that one of the things that you I mentioned earlier a lot of the agentic identity solutions are built around human centric identity infrastructure and kind of building an extension of that. But as we think about what really constitutes an agent identity agent identity if it's purely native it shouldn't be attached to an email and a login that's just not going to scale and work in the long run.

So agent identity needs to come from something else and we do believe that comes from persistent memory and the context that the agent accumulates and it has to be a living thing which is why you know that also leads us to look at blockchain based solutions that actually offers a dis distinct advantages to actually maintaining that immutable identity history and as and not only something they can read but also write as the agent continues to perform work and accumulate more context and memory. So that's some some things that we're able to see just being part of a larger organization that touches a lot of products.

This is why edge is going to be important too because you're going to need dedicated resources for these agents.

Right on.

Um Ellen, you mentioned agent security open cloud. Um you know I think that's a huge uh huge focus and topic out there. Recently um the Pentagon uh there was news at the Pentagon called Claude a national security threat. Um and then anthropics AI security researcher also recently resigned. Um and of course like OpenCloud has gone viral.

Uh but there's been a ton of security breaches. So as we um scale the you know Agentic economy I think agent security is a huge focus. Uh so as investors I guess how do you how do you think about the the companies that you're looking uh looking for to uh invest in that can really help build um AI in a in a safe and secure manner. That was that

oh any anyone? Yeah.

Yeah, sure. I think this is a fair concern. Um not only on AI agent but overall AI my Gemini and cloud already knows knows knows about me much more than my most of my of my friends do. So the question I think is not about whether to have AI know more about you. They already do.

They already have a lot of personal context. The question is how to have AI agent get a good control over the context. who can control the context and who can control who have access to it. So I think this is a governance problem. This is a audit and observability problems.

So I'm really looking forward to see what are startups are building AI observability like observability layer so that they can enable a good control in the overall AI agents.

Yeah. Something we're also looking for is uh you know if you're building AI for the enterprise, we hope that you have worked at an enterprise before or at least being close enough to it to understand you know it's more than just deploy an agent on your Mac Mini

and be able to just go shop for birthday party supplies, right? You're touching very sensitive data. You're touching things that can cause catastrophic uh failures, right? So, so having worked at these places where having been close enough to these products really elevates you as a founder above all else.

Yeah. I think one of the biggest issues right now uh especially is is most people around the enterprise side because with AI um it's all about data and you know individual user data your data my data individually it's all worthless but as an aggregate it's worth a significant amount of money and then for enterprise it this is their their most crown jewel this is like their golden goose and this is why a lot of times uh enterprises can't work with you know um you know cloud-based AI models So they have to host their own they have to uh contain all their data. Uh but this is the biggest risk is you know where where does all this data go? Who's training on this data? Um and especially with the consumerization of AI where uh the the you sorry like the employees of these enterprises are you know going around systems leveraging AI putting in their own uh company data into the AIS and and then you know being fed into uh generalized AI.

This is like a huge risk. this is like some of those security vulnerabilities and um you know that's why privacy is becoming a major major major issue uh and and a major talking point u at the moment but it's kind of funny because uh I used to work at Samsung and uh they banned um uh the chat GPT like when they first came out because the semiconductor teams were putting in proprietary chip designs into chat GBT and asking it you know for for advice stuff like that and all of that got trained on you know the master uh AI and then the information that it learned from there was being distributed out to competitors to open you know everything so uh it was a massive giant data leak so yeah it's it's a it's a huge security issue right now

all right I think uh we're pretty much at time so maybe last last point um what what would you like to leave the audience with like where do you see the AI economy in two I sure um this is a good question because in the past two decades we have been living in attention economy for a while and now we do see a a rise of intention economy. So what does it mean? Uttension economy is basically all the brands try to interrupt you, influence you, get in front of you. You see the ad seven times, you will buy the product. But this is not exactly what you want.

This is what what what was marketed for you. Um now with aentic commerce coming there's a intention economy this is completely different. It starts with your agents try to understand what you want understand your intent and then it goes outbound talk to all the vendors in the world and try to compare all the offers then find the best product for you then bring it back. I think this is completely different kind of commerce logic and maybe advertising will not disappear but when we look at advertising and discovery layers of all the e-commerce and and branding and search they will be rebuilt for this new kind of economy uh to optimize for agent to agent negotiation instead of um human attention capture. So this is very exciting.

Yeah, I would say, you know, given the the audience and the conference we're at, uh, you know, web two, web 3 has always been kind of two separate worlds before, uh, maybe what I hope to see is that sort of the blending of the two worlds, right? You know, instead of thinking like if you're builder in web 3, you know, doing something related to blockchain, maybe kind of figure out where that fits in the rest of that world, right? Fits into the broader AI wave rather than trying to build in a in a sort of a separate environment. Um, I would say, yeah, build build cool stuff that that needs to be solved. Um, and take my money.

Nice.

All right. Cool. Thank you guys.