Top SF VCs Discuss Agentic & Vertical AI | Agentic EP 02

Now the period that we're in is in the agentic AI period on our way to whatever's next. I think that could be quantum, but it also could be something around the corner that none of us knows.

I don't actually think the LLM and foundation model layer is set yet, and this is coming from the firm that was the first check in OpenAI.

I don't think any real agentic products have worked in production environments in a relatively unconstrained way. When people start with pretty narrow use cases and there's a small set of actions that people can take consistently.

Part of the challenge is technical. For example, why do you do agentic solutions need to be fully automated or not at all? Why can't there be something in the middle where you have a multi step workflow and maybe for one step of it you provide the user with hey, this is a suggestion accept or not. And then meanwhile the system is learning about that.

I think agentic AI is the solution that is now being embedded inside of vertical AI at this juncture. Industry knowledge of your of your domain is an edge and speed of shipping is another edge. And if you do that fast, you'll probably dominate a market.

Welcome to the Agentic Podcast where we dive deep with the pioneer AI founders reimagining business and investors betting on them. We bring you the war stories, insights and perspectives you won't hear anywhere else. I'm your host George Van Darren, the general partner of Untapped Ventures, a VC fund that invests in early stage AI startups. Let's go. Hello and welcome to the Agentic Podcast. I am excited to bring you today's episode which was recorded live at our Big Agent Convertible AI event during SF Tech Week recently. This was an incredible panel featuring six top venture capitalists sharing their insights on the state of AI, the future of agentic AI and what they're looking to invest in. Right now you'll hear from some of the leading AI focused firms like Khosla Ventures, Conviction, Parvc, Gradient Ventures, Georgian and B Partners, six panelists where we dive into key questions like where are we at in the AI revolution today? What's exciting about agentic AI? How can you create differentiation and emote, and how can founders successfully attract investment in today's market? So without further ado, let's dive right in. Just go ahead and introduce yourself if you would please. Name, title, firm and what do you focus on and any thoughts on what I presented specifically? Like, do you strongly agree or disagree with anything and it's okay you Know if you, if you don't, if you disagree, go ahead and say it. I like a good debate when I moderate. I like to have fun and so with that said so.

Hey everyone. I'm Naheem Nasser, head of engineering at Georgian. You might be wondering what is a head of engineering doing being in vc. Our model is very different at Georgian so I work with our investment team on software infrastructure thesis as well as our portfolio companies by hopping into the trenches with them shipping code at growth stage scale. What was the other question there?

Any thoughts on what I shared and do you strongly agree or disagree?

So I agree with most of it except for the one part that I think agentic AI is the solution that is now being embedded inside of vertical AI and I totally believe that this is an application that's going to start to go and enter these different verticals.

Gotcha.

Hi everyone my name is Arash. My last name is unpronounceable so I'm from parvc. I lead a lot of our AI and infrastructure investments there within that AI modern data stacks, cloud tooling, cloud infrastructure security and developer tooling been investing I'm an accidental investor. I was a software engineer and founder before by the way of Microsoft and a few startup mic Apple few startups in our per VC we are early stage investors basically we are the first checks in most of the time our focus is pre seed and seed and some A's. We live and die by that stage. We have two offices one in San Francisco, one in Menlo Park San Francisco office we call it Studio 220 desk lots of founders work out of there. We would love to see more of you guys there and we investing our fund for this is a $430 million fund and our check sizes are between 250k to 6 million and reactions to that obviously you know it's a very exciting time to be alive and to be exposed to what's happening in the AI and agent world. We also have a similar view that the agent agents are amazing but what are the use cases and where they really add value. That's what we really focus on.

Thank you.

Hi everyone my name is Ale. I'm a principal with B Partners. We're a pre C tech firm based here in San Francisco. We also invest that inception stage so pre seed only pre no product, no revenue, no customers. No problem is what we like to say.

I like that.

And I in terms of you know agreeing or disagreeing I don't want to repeat it but essentially we it's. It's Agentic is a tool, it's the how, it's not the what. And broadly speaking, our whole thesis or the reason why we were started in 2009 was because we saw that machine capacity was outpacing human capacity and we kind of saw that in the presentation. And so our whole thing is machines will win and how they win is up to us. And that kind of breaks down into three main categories. I spend the majority of my time on human machine interaction and the way that I see it for this current wave in generative AI and sort of the platform shift is like for the first time the way that humans and machines interact is changing. And so it used to be that we as humans had to adapt to the way to basically like how machines operate and we had to learn how to use software systems and again just sort of like it's not necessarily intuitive, whereas now it's kind of the other way around. And so this thing that might seem very subtle is actually incredibly important because it means that we now can collaborate and not just like send these dumb tools kind of like tasks to go and execute on, but rather also having a companion. And so that means that there's kind of like a new way in which we collaborate inside of an organization, a new way of thinking about how we do our work and essentially too that can, you know, higher level transform entire organizations. So that's kind of the general picture.

Of what we believe.

Thanks Ali.

Hey, sorry for being late. I'm Andrew from Gradient Ventures and we are a seed stage AI fund backed by Google. We operate super independently, we've been around since 2017 investing in AI since 2017. So lucky enough to be involved in some amazing early companies in the AI world like Lambda Labs where we were the first investor and yeah and prior to Gradient I was a founder, built a healthcare company, healthcare software company and mostly focused on kind of application layer AI. And yeah, super excited to be here. In terms of agents I think echoing what other folks have said, it feels very early. I think we're interested in the, I think what we saw with LLMs a lot of cool infra built over the last few years that now critical to building great apps. I think the same process is happening on the agentic side where there's a bunch of new core companies building infrastructure for building and using agents. So I think there's some exciting stuff happening on the infra side and then on the application side I think still super early. A lot of kind of agentic plays are very bristle today. So I think we're going to see as the technology improves a lot. A lot of really interesting kind of virtual applications.

Yeah. Great. Thank you. I'm glad you made it right in the nick of time.

Told me 2pm so my bad.

Okay.

It's all right. I almost only showed up. Can you guys hear me?

Yes.

Yeah. So I'm John, one of the partners at Coastal Ventures. Do a lot of infrastructure and AI stuff. Basically anything I've written code on in the past. I'm a technical guy. Prior life was one of the OGs at Palantir and Open Door and started a company and sold it a while back. And let's see, for contrarian takes here, I'll say the areas I disagree with you, George. Yeah, I don't actually think the LLM and foundation model layer is set yet. And this is coming from the firm that was the first check in OpenAI. So the I do think like with the standard transformers things are more or less kind of cemented. But you will see new research from left field or potentially different architectures and that will actually create some kind of wave like whether or not it replaces transformers or is an augmentation or just like another form of thinking like neurosymbolic logic may actually show up as another model. So that's one area where I think there's interesting things to do and we have a couple bets there. And the other one is like most of the large AI labs are US centric or like there's going to be continued regional interest or areas of opportunity that just continue to crop up. Like one of my investments last year in the C stage, Sakana is OpenAI for Japan. Right. And like, you know, if you're the Japanese government probably don't want your defense tech on an American company that can get turned off tomorrow. So there's going to be continued regional or like sustaining kind of opportunities that exist in that foundation model layer.

Thanks, John.

Hey everyone. Sorry getting over an illness but hopefully not an issue for the people who are further away. You should also imagine me like three inches taller. I'm prenup. I'm a partner at Conviction. We're an early stage fund that leads a small number of investments every year in seed and series A. So somewhere between a million and 10 million. And we run this grant program called embed, which is 150k checks into early stage companies that are anywhere from like inception to somewhere between rounds and so look like a lot of interesting companies. We just launched a batch three months ago. We'd love to see you when we launch another batch in three months. Before this I was a kind of shitty founder, mostly shitty engineer. Did some neuroscience research for a while. So laundry list of random activities contrarian take I think a general principle that this is a weakly held strongly expressed belief but a general principle feels like it's a difficult time to build meaningfully at the infrastructure for other people. Layer some way of interpreting the combination of what John said about like maybe foundation models get a lot better at learning or we have new tool sets that are available is I don't think the API is yet set for how people will build agents. And the second component of this is that I think you can count on one hand the number of agents that do meaningful work in any meaningful context. And so I would consider it the first building block for the majority of people is build search applications. Things that look like traditional QA problems are probably good ways to start and making an agent work yourself is probably the best way to go about ecosystem rather than trying to build a framework for someone else to make them work.

Awesome. Great. So part of what I heard in all your intros is that a little bit of caution as we're transitioning into Agentix. So I want to double click on that. I guess maybe you guys can expand whoever wants to take it in any order. So like, so where do you feel we we're at with AI in general and, and, and horizontal and vertical and agentic and where is this like fitting in and where, where are we going with this?

I think one thing we're seeing on the commercial side is just a little bit of burnout from bigger companies in adopting new AI tools. So definitely seeing in health care some AI pauses from health systems seen on the legal side as well. And so I think there's still a lot to shake out and a lot of noise in the market. So kind of interested to see what.

Do you think is causing that.

I mean every large enterprise tech enterprise is offering AI tools, right? Some of them not very compelling and some of them pretty combative with early stage startups. And then there's a ton of new startups reaching out to companies and making a lot of noise in vertical markets. So in legal there's a number of players now selling to law firms and so I think law firms are a little bit burned out on all of the options. Some of the technology maybe is a little bit over promise under deliver today. And so I think it's just an early market and so the next couple of years we'll probably see the technology get better and A few companies kind of stand out above the. The noise.

Yeah. So Nahim, at Georgian, I know you guys did this massive research study and there was a lot of insights on that. Do you want to share what the study was, who it was for? You had two segments and then what were some of the takeaways as it relates to the point that Andrew was just making? Sure.

So we did a study, we interviewed and surveyed about 600 growth stage execs. There were a few interesting insights. I think you can download the report later if you're interested. But the data that firms are using right now, they're using dark data. This was the most predominant training data set that enterprises were using. When I say dark, I don't mean like off the dark web, I mean completely underutilized data that organizations have been collecting for years, but no one was using. So I thought that was quite interesting. And then when you go into the use case, the use cases side of things, predominantly we're seeing a lot of operational use cases. So customer, you know, call center, customer collaboration, customer support, and then also analytics and operations for the most part. So just on the last note, I think maybe I just had one hot take on where we're at today on agents. Has anyone used the O1 model? OpenAI? Yeah, you probably use that. Chain of thought in its present form is pretty bad. And if you think about all of the problems with it, we already have problems with hallucinations. Now you're compounding that risk. You're basically saying, hey, I'm going to generate a prompt that's going to generate chains of more prompts. And if something goes wrong along the way, that error is going to propagate throughout the whole system. So your overall error rate just drops significantly. And then you could say, shut up, Nahim. We have evals, but it turns out that that also is now getting very expensive. So to your point earlier, I think generally speaking, this cannot be it. This cannot be the foundation for what agents are. There has to be another breakthrough because the present state, it's just literally engineers that are saying, hey, you got this data science thing that's up here, let's try to hack the crap out of it to get high accuracy and high performance. Sorry if I went on a rant there.

That was good.

It gave me a chance to do a time check and I saw that, yeah, we actually are ahead of schedule. You weren't late. We just started super early and I went through my intro fast. So sorry for giving you a little bit of crap there. And I'M happy to hear that. I'm happy to hear that. You know, usually it's the other way around, but. All right, good. So we'll have more time for some good questions from the audience and also for networking. Was somebody going to continue off of what he said or.

I mean, I largely agree with that. One point I would make is that agents haven't worked so far. Hopefully that's not terribly controversial. I don't think any real agentic products have worked in production environments in a relatively unconstrained way.

Ooh, the founder panels might have some thoughts about that.

I think where it does work really well is when people start with pretty narrow use cases and there's a small set of actions that people and take and take consistently. And so to your point about like, these are largely engineering problems today, I think one of the strong beliefs the fund has is like, well, if you model these agents as like, we need to take some set of like let's say 10 actions and that every one of them has to be right in order to get to successful outcome. And each one of these actions has some, let's say like 95% probability of working. You look at it at the end and that's like a 50% chance the whole thing succeeds. And so one of the biggest benefits for companies that are trying to do sophisticated tasks is just improvements in base models. And so I think we should all be excited for like new anthropic release or new OpenAI release that demonstrates better base model capability because it just enables us to make agents work better and more easily.

I will just point out there is one agent that works very well that everyone here I hope has had a chance to try. But the waymos outside are effectively agentic tech.

Just one quick thing because I think part of the challenge is technical. But what, so you're just going to kind of sit around and wait for the base model to improve? I think there are, there's another kind of core side to this and it really has to do with again this, the interface and how you design your product. And I think there's a lot of things that you can do in the interim. Like for example, why do agentic solutions need to be fully automated or not at all? Why can't there be something in the middle where you have a multi step workflow and maybe for one step of it you provide the user with, hey, this is a suggestion, accept or not. And then meanwhile the system is learning about that and that way you kind of, you reduce the chances of, you know, Running and kind of like replicating there all the way down. So part of it is technical, part of it is non technical and there's also like just time needs to kind of happen so that we as users also get used to incorporating this technology in our day to day. And I'm talking more about like you know, not just engineers but really kind of every single other business user out there.

I just wanted to add as well that when you look at the use cases that the agents can be used for, there are certain requirements for each use case. There's SLAs, there's requirements for, you know, the, the reliability, security, all those aspects. There is a curve in the enterprise, you know, you either make things faster or you're adding more functionality. The more you can get through various tool, various methods to get the agents to do what the use case needs to do, the more use cases fall under that curve and you will be able to implement more and more of those agents. But while you do that, you have to be very conscious of what the limitations and you know, the, the shortcomings of the agents today are. One aspect of it is evaluation. If your use cases require high reliability, you may need to use a neuro symbolic, you know, way of executing your actions or evaluate in areas that are evaluations are appropriate for that. So it basically today we will, we will see Foundational Model and OpenAIS afterward, introduce more tools and agentic support. And if you're a founder you have to be very conscious of that not to be in that strike zone. But assuming that's happening, what is it that you're going to be building that adds value?

Yeah, just one last thing. I think we're kind of all talking about the sort of like really high level concept here is trust and how you create an experience for the user that makes them trust your application or trust whatever it is that you're building is incredibly important. Part of that is technical challenges, part of it is interface, more like front end and design challenges. And then part of it too is like again us as a society just getting used to engaging with this technology. If you think about, you know, with the last platform shift and using Maps even, I mean I remember 10 years ago if I knew exactly how to get from point A to point B. I wasn't using Google Maps or Waze, but then it wasn't just about the product getting better, it was about me being like, oh you know what, I use it, everyone else does. And so now I don't even think about it. I just get in my Car and I'm going to the office which is the same place every single day and I'm always requesting the phone to tell me which is the most. The optimal route essentially. But that took 10 years.

Okay, so why don't we continue and now maybe get a little more specific in terms of. So what are you guys looking to invest in as it relates to Gentec? You know, we've kind of been talking around it a little bit, but let's get specific. Where do you guys think the opportunity is? Is there any recent investments that you have made that you're excited about that you want to share? Is there any areas that are not good? You kind of started to touch on it like as the foundation models are going to develop competing solution. So kind of some caution for the founders and sorry that I kind of have my half back to you guys. It's a full panel and also I'm not tall enough to swivel by myself so I have to step down. Yeah. So I know that was a multi part question or multiple different questions, but feel free to attack any of those essentially. What are you investing in Agentic? Is there a recent investment you want to share and is there any areas that maybe wouldn't be smart for a new founder to go build and take any of those?

I guess I can start but on the application layer and this is going to differ by every single market. But the thing that gets us excited is if there's a story for defensibility and that story has to be like OpenAI is not going to run us over or any other host of competition out there at, at scale. And then as you go further down the stack, like I.

So by the way, John, so what would that, what does that sound like?

Like what are the, the typical moats like can apply, it can be workflow, lock in, it can be data modes. We, we kind of leave it to the founder to determine it because in some of these markets, especially in very specialized markets, there are insights that lead to moats that we wouldn't be able to say a priori. Right. But I think people that actually really understand their market well will be able to figure these out or realize that they don't exist.

Right.

The other part here that's kind of interesting is like if you look at the tech of agents today, like you go Back to the AlphaGo paper, right. All agents are some form of Monte Carlo tree search. And like in Monte Carlo tree search, the interesting part of that is like you go down a path, it's wrong. You backtrack backtracking is the hard part because state actually changes when you go back and forth. So the whole infra layer that you build for agents to utilize needs to change to match their use case. One of our portfolio companies, I'll call it, but actually it's an incubation that my partner Nikita still runs as CEO at Neon, has started going in this direction of being a database that has branching and rewinding, which traditionally is very hard for SQL databases. And really that's because one of our investments is Replit and and he spends a lot of time with Replit and he's like, oh, this is very obvious that this is going to be necessary for agents and the replit agent really needed this, right? So that entire interfro layer is going to end up being rewritten or being customized because agents need to be able to operate at speed because they're doing thousands of these searches across a tree in a search space and then one level down. I think a little bit related to my comments on like foundation models are not set yet, but I do think you will see things that are potentially not purely LLM based in more agentic action model spaces. We have a couple of investments in that area from researchers from your usual suspect labs. But like this is essentially the. I'll say like this space is probably more interesting because this is like Adept's original hypothesis and I think people were afraid to try it because Adept had such a stacked team. But now that that story is over, it seems a little more greenfield for people to go approach it.

I think, I mean see the value in asking about defensibility. I think we think about this a lot less early on in large part because execution feels like defensibility. If we start from the premise that lot of agents just don't work, getting something to work is the starting point that we think matters for most companies. And so like I think in general large labs benefit a Bunch from Focus OpenAI Anthropic and like your usual suspects of large labs benefit a lot from like having a singular focus of we are the best research lab. I think they're unlikely to be the best stewards of product that requires specific integration into verticals. And so I think our focus tends to be a lot less on like do we think this is defensible? A litmus test that we use a lot that I would recommend strongly for people is if you were to discover that there is a new model coming out tomorrow that has better capabilities. Are you excited or apprehensive? And hopefully the answer is always excited because models will continue to get better and continue to get better over the next couple years. And so this is, I'd say this is a question we ask ourselves a lot. We like to be on the right side of this. If we project model improvement, we should feel good about all the investments that we make. One thought related to what Andrew was saying earlier about some commercial apprehensiveness on AI, I think, I mean you all are founding AI companies, but they don't need to market themselves as AI companies. And so some of the investments that we've made, like one of the benefits is that the service that you provide can actually just look like a service. It doesn't need to market itself or be an AI service. And so to Ali's point about it sometimes helps to portray yourself as like a traditional service user doesn't have to know that this is any different than whatever they've been doing before. And so I think at one extreme this looks like roll ups like you should purchase. There's a bunch of these CPA style roll ups that are like let's buy an accounting firm and then just make the accounting firm more efficient or have met businesses that are like leasing services. But I don't think it has to look like that. It can also just be a much more minor version which is like we'll offer a more efficient service, hire a couple of practitioners in that subsequent space today and then eventually it's AI in two years.

And for us, we also focus very much on the ROI right now as was mentioned, the enterprises have done a lot of pilots over the past two years and now they're saying okay, what is the benefit here? So using agents is a tool to solve some business solutions for us. The areas that we are excited about, we have divided into 16 areas for ourselves that we how we look at those investments. But there are many problems on the agentic side that has the foundational models haven't solved or they will get there, but it's not there yet. One component of it is the action models. There haven't been enough data on the actions. When people even then you take actions just to take a step back. The agents have existed forever, right? It's not a new concept in software engineering. What's changed is that now the agents have a much more broader domain on the input or interactions, much more broader domain on, I mean undefined number of actions that they can take or much broader number of actions and decisions and the logic that goes behind it. So each one of those areas right now is at early stages and you know, foundational model and that platform is going to give us much better reasoning. It's going to get us much better tool called give us evaluation, all those components. But there are white spaces there to get these agents working properly for the use cases. So we are excited about the use cases that we can start from top down. This is a use case that an agent would actually create an ri, not just because, because it's an agent, it's because it actually created different business proposition there. And I understand that you need to look at inflection points as well, but it gives us a grounding about what is the data. If you know some. We have companies that are doing broad automation in agent world, but they start from the use cases and the building foundational models for actions, for example, or evaluations or how they take action. They use the symbolic that was mentioned just to make it more reliable. But there is real ROI in what they're developing. Like you can look at or be one of our companies, they're doing finance use cases right now that their requirement for them is very high. But they had to solve every single one of those problems. So think about if you're using an agent framework, what is. Why. Why is it different than you doing it another way? And so that's one of the bees.

I think for me, after watching the last OpenAI dev day, it was really interesting to sort of hear this transition towards almost like there's going to be, at least for now, two separate lines of development, one which is like reasoning and then the other which is just the typical GPTs and we're going to continue in that line. And it made me think about, with everything about agents, they didn't necessarily say we're building a, we're building an AI agent builder, but essentially they're providing you with all the tools to make it really easy to create that first version, which likely is more horizontal. It can do like a lot of the simple tasks. And so I'm personally really cautious when it comes to things that are too general, too horizontal because I think again that sort of stuff is going to become very quickly commoditized. It's the ones that, you know, the, the use case that they're solving for is probably very well represented in the underlying data. And I know that Aria from TensorStax is going to talk about this in the founder panel, but that's why we're a lot more interested in those kind of like vertical use cases where there are certain tasks that are a lot more complex that have a lot more value for the organization and are not very well represented in the data set. So there's a lot more stuff that you have to build in order to get those right. And if you can create more value, then arguably you should be able to extract more value.

This is a perfect segue into my favorite question for this panel. We're all learning together, right? And my favorite question was exactly on this topic of vertical versus horizontal AI in general and then specifically for agentic. Vertical versus horizontal Agentic. And it was funny because when I was preparing for this part of my framework of how I think about it and how I explain this to people, especially other communities like investor LP communities and stuff is, you know, SAS went from SaaS to vertical SaaS. And I give the examples of Salesforce and HubSpot. And then. Right. And then after that you add vertical CRMs that you know was built for a specific function. And so I think I've been replaying this Talk track that HubSpot and Salesforce came and then there was vertical apps that came focused on specific use cases that started to take market share. But last night as I was reviewing my notes, I said, but wait, in the process, Salesforce and HubSpot still built massive businesses. It's not like they went out of business or saw some super sharp decline because of the vertical per se. So but I'm really curious to get you guys thoughts on where we've seen this play out historically, this idea about like horizontal versus vertical. We've seen it in sas, I'm sure we've seen it in other areas as well. And then so if we start from there and go into AI generally and then specifically Agentic and this is what Ali was just talking about. I think a few of you guys, I think what I'm hearing is that you, you, you're very cautious about horizontal agentic and you'd be more keen on vertical agentic for the reasons you've been stating. But let's just have like I'd love to just tackle that head on.

I think what Pranav said earlier was pretty interesting and I really agree with it. You don't have to sell your company to a customer that you're an AI software solution.

Right?

Like you just need to solve that problem. And so I think if you're building an AI agent in a vertical industry and you're selling it this like AI solution, I think your customer probably won't understand it. There's probably a ton of noise. And so really you just need to figure out what's the technical solution and business model that will allow me to solve this problem for this customer. And so I think sometimes yeah, if they understand services, maybe you sell it as services. Right. Like figure out how you can solve this problem for this customer and help them understand it.

Right. Just sell the outcome. We've made a number of investments that are selling the outcome and increasing their profitability.

Some interesting, you know, business model innovations that that's actually what gets me most excited about. You know, unique moat is important and, and kind of like unique insight on a market. But also like what's the business model you figured out that you know the customer likes and is high margin makes you money. Right. And so I think there's some interesting given how much companies spend on services in certain industries there are interesting kind of unique hacks you can come up with that aren't just like you know, usage based software pricing or SaaS pricing.

Yeah, but so you're in the camp of vertical. You wouldn't invest in a horizontal agentic.

I mean the sales force of this industry is going to be OpenAI. Some of the other foundation model companies like clearly OpenAI is not going to make and Google etc. I'm going to make a ton of money just selling API access to models. They're going to need to build a lot of infrastructure, consumer products and infra. And so again the point earlier about like as models improve is that hurting or helping your business? I think that that definitely plays with the more horizontal players.

Okay, so we have one vertical.

Yeah, maybe one thing to just say so you know we're seeing a lot of things play out on vertical. I'll call software development a vertical. So we saw Copilot. There's cursor. Kurt took the basically saying I'm going to do reg for your code base and I'm going to have awareness of your code base locally and then everything's better than copilot. I think the same thing will probably play out in multiple industries like health tech like clinical notes. Having context and awareness of those types of things to produce less burnout for doctors or legal tech as in structured clauses and things like that. I think so the vertical companies are the ones that are applying the new infrastructure changes the fastest. And I think at this juncture industry knowledge of your of your domain is an edge and speed of shipping is another edge. And if you do that fast you'll probably dominate a market.

I think an objection that a lot of founders have to founding vertical companies is that they just seem Much smaller like horizontal companies are much more expansive in ambition. And so like a general piece of advice I would give you is that it is useful when you meet with VCs to communicate some version of the mission that is both its most ambitious but also focused. And so there's some notion of how strategically you plan on dividing up large landscape into problems that you can tackle early and well. The ambition is also useful to communicate. We want to invest in founders that are building really large giant companies. And large giant companies do more often tend to be multi vertical or horizontal. But like a good example of this is if you look at like the most recent Harpy, you know, redesign. The Harpy redesign, the Harpy page redesign now looks like, you know, we're not just a legal company. Actually lots of things involve legal workflows. We'll do like let's do your M and A for you. We should handle more of the transaction than just the legal component of the transaction that we were working on PwC with. And so the I think a general fear for people is that we start as this vertical company, then we will forever be painted as this tiny vertical company. We'll never get to expand and do other things. But actually I think this is untrue even if you take like HubSpot as an example. Like they started as landing page optimization and it became a CRM company many years later when it became clear that customers needed CRMs. So I would start with like you should win something and then find the way, find the strategic vision from the thing you win to a much larger horizontal company. And I think it's unclear if like OpenAI or any of the API providers will be that horizontal company. It's, it's hard to say like where value will accrue, but it's probably not.

The API.

For us vertical versus horizontal. We actually invest in both. But we have a very specific opinion about how the horizontal companies that we invest in, what we consider there completely agree. You should start, you know, even if you're building a, when you're building a horizontal company actually you should have a bench entry point as well. It's really hard to build platforms or horizontal things most of the times bottom up, you need to have some use cases that you start with and then you can move to HSM use cases and build the platform that way if you just starting bottom up, you have to have enough experimentation beforehand to be able to have the use cases defined at least to set for set of use cases. And most of the times when I see people start Start a company like that, they run out of money before they actually address any useful use case. So I think we invest in horizontal and there's horizontal pieces in the agent world that are not being addressed at least directly by the foundational model providers. Like when you're in the enterprise context, how do you get your data to your. To your agents? What would that look like? What are the aspects of. There is a whole lot of lineages, security, all those aspect that has to be worked out by the time the agent can use it in a context. So those horizontal plays we are okay with. But again we very much like a vertical, you know, use case at the beginning. And then also the verticals can be large verticals as well. That as as was mentioned, can you see a large enough market for that? And that's important for us. Let me stop you.

I feel like I don't want to repeat what everyone said because it's. It's kind of. When you sent us this question, the way I think about it is like by stages and to your agents, what.

Would that look like? What are the aspects of. There is a whole lot of.

Without again saying focus at the beginning because it's important from a product development perspective, from a go to market and.

Linear security, all those aspects that has to be worked out by the time the agent can use it in a context. So those horizontal plays we are okay with. But again we very much like a vertical use case at the beginning. And then also the verticals can be large verticals as well as was mentioned.

Marketing resource efficiency perspective. And also from like a storytelling both to your customers and investors perspective. I just think that two key things to pay attention to very early on is like what is that?

First, can you see a large enough market for that? And that's important for us.

Let me stop vertical that you go after just having a really solid story of why you're going to focus there. And then again it's all theoretical. I feel like I don't want to repeat what everyone said because it's kind of. When you sent us this question. The way I think about it is like by stages and at the pre seed. Right. But there has to be some kind of lot. If you're talking about expanding horizontally and becoming a platform eventually how much overlap is there between that first vertical without again saying focus at the beginning. Because it's important from a product development perspective from a go to market and the second one that you're going after ensuring that that first vertical is big enough on its own, frankly, because Chances are that while you're building for X, there's a million other companies that are building for the vertical that you're going after and they're going to have a head start. So how are you going to be able to kind of like compete? You need a strategy before you even get going in that sense.

Right? They have the one main vertical and then they put, you know, five others on the slide thinking like, oh, it's just going to automatically happen next year.

And in the next marketing resource efficiency perspective and also from like a storytelling both to your customers as an investor perspective. I, I just think that two key things to pay attention to very early on is like, what is that first vertical that you go after? Just having a really solid story of why you're going to focus there. And then again it's all theoretical three years.

But no, it's not going to be that easy.

I think there's a few useful things that have been said on this panel already, but I actually think about horizontal versus vertical is kind of a misnomer. Okay, like you, you use the example of coding agents and you call it a vertical. Well, is, is a coding agent over.

Well at the precede. Right. But there has to be some kind of lot if you're talking about expand.

Vertical, is AWS coding agents a vertical? Is Python coding agents a vertical. So like, what exactly is the vertical.

Thinking horizontally and becoming a platform eventually, how much overlap is there between that first vertical and the second one that you're going after, ensuring that that first vertical is big enough on its own, frankly.

Right. And the question is actually one of value less of what is the vertical versatile?

Because chances are that while you're building for X, there's a million other companies.

That are building for the horizontal thing that you're doing. Just in any other market, the smaller you constrain your user base to the more specific you can make your product and the better your product for that user. And this is the same way to think about horizontal versus vertical. Like obviously there's how easy is it.

To build the vertical that you're going after? And they're going to have a head start. So how are you going to be able to kind of like compete? You need a strategy before you even get going in that sense.

Right. They have the one main vertical, right?

Like a foundation model company is a lot harder than just using a foundation model company for a vertical. But there's also like, what is gold?

And then they put, you know, five others on the slide thinking like, oh, it's just gonna automatically happen next year and in the next is the delta.

In value that you can provide when you constrain your market to. Some call it vertical. And that, that's really the, the question that's there. And that's the same thing that happened in SAS too is like three years.

But no, it's not going to be that easy.

I think there's a few useful things that. Why do people have vertical sasses? Well, because like I can cut out some smaller market that I think is still large enough to build a competence set on this panel already. But I, I actually think about horizontal versus vertical as kind of a misnomer. Okay. Like you, you use the example of coding agents and you cut it a vertical. Well, is, is a coding agent over me. I can build a better product for that smaller market. And voila, there's a company here, right. It's like the same thing in AI. The only difference is the way the technology works. The other thing that people here have said that I think is quite value vertical is AWS coding agents a vertical. Is Python coding agents a vertical. So like what exactly is the vertical? Is like you don't have to start large. The vision needs to be. Well, if you're building a company, the vision doesn't need to be large. If you're building a venture scale company for large tier one fund vertical. Right. And the question is actually one of value less of what is the vertical versus horizontal thing that you're doing? Just in any other market, the smaller you constrain, the vision has to be large because otherwise we can't return our funds. Right. But like your initial MVP or your first starting point doesn't have to be large. So Peter Thiel funded my company back in the day and like the conversation I had within your user base too, the more specific you can make your product and the better your product for that user. And this is the same way to think about horizontal versus vertical. Like obviously there's how easy is it to build him over and over was like what you're looking for is a beachhead and you want to own the beachhead. And that's like your base of strength. But you want to be able to expand from there and take over the continent. But you always have to start at a beach head. And that's kind of what everyone here is saying.

So I want to continue off that. So right.

Like a foundation model company is a lot harder than just using a foundation model for a vertical. But there's also like, what is the delta in Value that you can provide when you constrain your market to some call it.

I think we've established that point like pretty clearly. I love that. So then besides that and maybe we could start with Ali since you know you had that famous line of the day so far. What was it?

Vertical. And that's really the question that's there. And that's the same thing that happened in in SaaS too is like no.

Customers, no revenue, no problem. I think you had a couple other no's in there. So then, you know, what are you guys all looking for starting with Ali then you know, assuming it's.

Why do people have vertical sasses? Well because like I can cut out some smaller market that I think is.

Still large enough to build a company vertical agentic AI. But what, what are you looking for in. In the business and the team I.

Can build a better product for that smaller market and voila, there's a company here. Right? It's like the same thing in AI. The only difference is the way the technology works. The other thing that people here have said that I think is quite valuable.

And anything else that for your decision making process.

Is like you don't have to start large. The vision needs to be. Well if you're building a company, the vision doesn't need to be large. If you're building a venture scale company for large tier one fund.

I think and this might just come from recency because I'm thinking about the last company that I talked to in sort of like the vertical space. One of like the bigger questions for us and putting kind of the team aside, because that's a separate conversation was there's kind of this chapter one. We can arguably say that the incumbents have won. The ones that you know, kind of existed pre Genai.

The vision has to be large because otherwise we can't return our funds. Right. But like your initial MVP or your first starting point doesn't have to be large. So Peter Thiel funded my company back in the day and like the conversation.

I had with and now they've like sprinkled some stuff on top and now they're AI enhanced. But now there's kind of like the second wave coming of AI first companies. And I guess one of the things.

That I him over and over was like what you're looking for is a beachhead and you want to own the beachhead.

And that's like I'm always trying to figure out is what is the delta? Because again what matters is not how.

Your base of strength but you Want to be able to expand from there and take over the continent. But you always have to stop.

You get to that outcome. But it's, it's. I mean it's, it's not the underlying technology necessarily but it's like all the other things that it can unlock. So for example, because I think one.

Of the questions at a beachhead and that's kind of what everyone here is saying.

So I want to continue off that. So I think we've established that point like pretty clearly. I love that. So then besides that and maybe we could.

That you asked before was about spaces that are a little bit crowded and it's like that kind of top of funnel SDR bdr. You know, let me generate leads, qualify them and do that first meeting for you. Like that was pretty much already figured.

Out with start with Ali since you know you had that famous line of the day so far. What was it?

Previous technology with just your basic rpa. And so with AI there is somewhat.

No customers, no revenue, no problem. I think you had a couple other no's in there. So then you know, kind of an.

Improvement in the experience. But it's not real, it's only incremental. But what happens when the human gets involved and then when you have someone that is engaging with you know, multiple tools that has arguably you can talk about sales as having kind of like a similar.

What are you guys all looking for starting with Ali then you know, assuming it's vertical agentic AI but what are you looking for in the business and.

The team similar type of workflow but there's so many nuances that come there and so there's like different integrations and. And just there's a lot more complexity and multiple steps that you have to take. So.

And anything else that for your decision making process.

I think and this might just. I think I'm more excited currently about like last mile delivery of in the agentic space of not just doing a piece of the whole task but how do we get all come from recency? Because I'm thinking about the last company that I talked to in sort of like the vertical space all the way to the end. That's something that I'm particularly looking for. And then if it is vertical then I think it is incredibly important to have someone on the team that has that domain expertise. One of like the bigger questions for us and putting kind of the team aside because that's a separate paired with someone that has the technical capabilities. I think I can't remember who said.

This at the Beginning is that kind of what you're looking for? A2 co founder combo like a business and a technical. The business is the domain next.

Not necessarily conversation was. There's kind of this chapter one. We can arguably say that the incumbents have won the ones that you know, kind of existed pre Genai. But I, I think, I think it, it helps in the verticals, in, in verticals. And the reason for that is like you have an advantage in T0 because of the experience that you've captured and then so long as you can move fast. And now they've like sprinkled some stuff on top and now they're AI enhanced but now there's kind of like the second wave coming of AI first companies. And I guess one of the things that I know if you should be able to get keep that up in T plus one. But I'm just generalizing. We have multiple examples that are not like that at all.

So yeah, so, so you, you took the team part, you didn't address that part. Can you also I'm always trying to.

Figure out is what is the delta? Because again what matters is not how.

Just address that because I want to go around, I want to make sure you guys have a chance to just share very clearly like what, what are you looking for? How do you evaluate? Sure there's going to be people that after the next panel are going to want to come and talk to you. So it'd be helpful for them to get a better sense of like, you.

Know, you get to that outcome. But I mean it's not the underlying technology necessarily.

Am I a better fit for Allie versus Andrew? Like who should I talk to? So maybe we can come back around to you.

But it's like all the other things that it can unlock. So for example, because I think one of the questions that you asked before was about spaces that are a little bit crowded and it's like that kind of top.

Let's hear from Andrew. But yeah, just, just clearly if you guys have like hard and fast rules or just generally what you're looking for as it relates to the founding team and the startup and how you like to be pitched and what you're looking for, I think that'd be helpful because then I want to, I want to go around everybody on that and then we'll have time for a few audience questions, questions which we weren't planning on but you know, now that we made.

Up time, we'll be up a funnel sdr bdr. You know, let me generate leads, qualify them and do that first meeting for you like that was pretty much already figured out with previous technology with just your basic rpa. And so with AI there is somewhat of an improvement in the experience but it's not real, it's only incremental. But what happens when people to do that?

Yeah, yeah, I think I, I'm super biased toward intellectual curiosity from founders so I think people that have a contrarian view in their market maybe sounds, sounds, you know, non contrarian but I love to learn so I love when I meet a founder and they have like a super unique view on a market or unique insight with a business and it like changes my perspective on a market or I learn something and so I think the companies I've enjoyed working with a gradient and I think the approach I tried to take as a founder was that to have a unique view on my market and a unique insight for building the business.

Human gets involved and then when you have someone that is engaging with multiple tools that has arguably you can talk about sales as having kind of like a similar, similar type of workflow but there's so many nuances that come there and so there's like different integrations and.

So I think outside of that there aren't that many hard and fast rules. Like I think great businesses usually outliers so they're not going to look exactly like the last great business you backed.

Or and just there's a lot more complexity and multiple steps that you have to take take.

So succeeded in a market. I started a healthcare business with no healthcare experience but so I you know I could have a rule that like you need experience in the market but I, I think I tend to just back people who move quickly.

Well I think I'm more excited currently about like last mile delivery of in. In agenda in the agentic space of like not just doing a piece of the whole task but how do we get all the way to the end? That's something that I'm particularly looking for.

And then if I have a ton of curiosity and excitement about building about the market and kind of insight on the market.

Okay.

Maybe I can tell a little bit about Georgian. So we typically do growth stage deals so we're kind of in a sweet spot. We're not the first check, we're not the last check series.

If it is vertical then I think it is incredibly important to have some someone on the team that has that domain expertise.

B series C Usually you're starting to put numbers up on the board. Your numbers are speaking for themselves. But we do deviate from that from time to time. And we had this one investment earlier this year basically going off of what you're saying. He had a unique insight into how.

Data paired with someone that has the technical capabilities. I think can't remember who set this at the beginning.

Is that kind of what would be.

Structured moving forward for training and and. And how the AI landscape would would.

But you're looking for a two co founder combo like a business and a technical. The business is the domain expert.

Not necessarily but I, I think. I think it. It helps in the verticals in. In verticals and the reason for that is like you have a change.

So he created this. It's the company's called Surreal DB and they're. They're very early but unique database founded on the idea that point in time.

Data an advantage in T0 because of the experience that you've captured and then so long as you can move fast.

Basically is required for training as well as multimodality being a thing Anyway growth stage typically is what we would do series B 25 mil to enough.

You should be able to keep that up in T +1 1. But I'm just generalizing. We have multiple examples that are not like that at all. So.

Yeah, so. So you. You took the team part. You didn't address that part. Can you also.

50 mil is usually our. Our strikes strike zone.

We are both preced and seed. So at precede they have slightly different criteria. Obviously when we look at that what we don't have is the luxury of data. We don't have data. So the first thing I'll just address.

That because I want to go around. I want to make sure you guys have a chance to just share very clearly.

Like for us is the founders and the founder team. We have this concept for pre seed. Is it a minimum fundable team? Once you pass that bar it's just.

Are you able to share what that bar is?

Well if you're building a technical company and you come and you're just two business people and you need to complete that with a you know, very technical person. Do you have leads on that? Who, what.

What are you looking for? How do you evaluate? I'm sure there's going to be people that after the next panel are going to want to come and talk to you. So it'd be helpful for them to get a better sense of like you.

Know, have you talked to people who are interested and at least some aspect of that. So. Or if you're building in a very specific vertical. How are you going to gain the knowledge of that?

Am I a better fit for Ali vs Andrew? Like who should I talk to? So, so maybe we can come back around to you. Let's hear from Andrew. But yeah, just, just clearly if you guys have like hard and fast rules or just generally what you're looking for as it relates to the founding team and the startup and how you like.

To don't necessarily require to you have that knowledge. I built a company software defined networking when I started out. No background in that either. But have they done the customer discovery?

Have they done talk be pitched and what you're looking for? I think that'd be helpful because then I want to go around everybody on.

That and then people discovered a pain point that people you know have that part of the work has to be done beforehand. Then we have also you know various other metrics on the founders what we what to just standardize within our firm. But one key component is that when I meet with you I should not be able to sleep at night until I contact you again.

Right.

And so that that comes from and.

We'Ll have time for a few audience questions which we weren't planning on but you know now that we made up time we'll be able to do that. Yeah, yeah.

I think I, I'm super biased toward intellectual curiosity from founders so. So I think people that have a contrarian view in their market maybe sounds.

Not only wanting to solve a problem which is why, why in the core. Why is it a core thing for you that you're doing this startup? It's going to be hard, it's going to be lots of ups and downs and you know what would get you.

Through that is you know, non contrarian but I love to learn so I.

Love when I ups and downs in addition to the monetary risk, you know, returns and all that. Why are you, why are you building this company? So that's one component that we look at. Obviously the market has to be perhaps the potential to be big for be.

A founder and they have like a super unique view on a market or unique insight with a business and it like changes my perspective on a market or I learned something and so I think the companies I've enjoyed well are.

Kind of firms again a lot, a lot of good companies can be built and they don't have have to be venture backable for tier one ventures. You need to have some version of that and the third comes the product.

And working with a gradient and you know I think the approach I tried.

To take is a vision One thing I want to mention about AI is that we go in, it happens a lot that you go from something that's being impossible and then becomes possible as a founder. When you look at it and you want to build in that area, did it become easy or is it still very difficult?

Yeah. Was that to have a unique view of my market and a unique insight.

For building the business and just become. Become possible. I usually gravitate towards it was impossible, it's very difficult to build still. But now it just became possible. This is my vision, this is why I'm building it, that's why I understand my customer. So some of the things we look at.

So, so for a firm like Pear and you guys are agnostic, right? You have cons.

And so I think outside of that there aren't that many hard and fast rules. Like I think great businesses are usually outliers so they're not going to look exactly like the last great business you backed or that succeeded in a market. I started a healthcare business with no healthcare experience.

But so sumer you do like health like lots of non AI. And so for the non AI, for the agnostic VCs on the panel, do you look at AI, do you have.

You know I could have a rule that like you need experience in the market but I, I think I tend.

To just back people who move quickly some different ways. You look at AI startups specifically agentic different than the rest of your portfolio. If it's a founder building an e comm or B2C. Is there different layers or levels or.

I have a ton of curiosity and excitement about building about the market, kind of insight on the market.

Okay.

Maybe I can tell a little bit about Georgian criteria.

We are not agnostic, we just broad investors. So we have a specialist on the team for each area of the, of the stack of, of the. I mean each area we have a bio partner, we have a healthcare bio partner but as it happens a lot of the companies have an AI company.

So we typically do growth stage deals. So we're kind of in the sweet spot. We're not the first check, we're not the last check series right now.

So, so what we look at for.

AI company B series C usually you're starting to put numbers up on the board. Your numbers are speaking for themselves again.

It's already been mentioned a few times. What, why are you building? What's the value you're going to be at?

But we do deviate from that from time to time and we had this one investment earlier this Year basically going off of what you're saying. He had a unique insight into how data would be structured moving forward for training and how the AI landscape would.

I think what inflection point are you seeing? If you're a very technical founder, what is the invention that opens up something that was impossible now becomes possible and you're going to be spending the next five, 10 years change.

So he created this. It's the company's called Surreal DB and.

They'Re, they're very like building that impossibility into a possibility in a world that didn't exist. So all of those typical things, you know, it depends on, you know, really depends on the opportunity and what they're looking at.

Early but unique database founded on the idea that point in time data basically is required for training as well as multimodality being a thing anyway, growth stage typically is what we do series B.

25 mil to what the founder market fit is.

Gotcha.

Yeah, I was trying to think of like characteristics that we look for in teams and I really liked the point on kind of infinite curiosity and just wanting to learn.

50 mil is usually our strike zone.

We are I think something that we look for and it's very much part of our diligence process is like the deltas over time.

So precede and seed. So at pre seed they have slightly different criteria. Obviously when we look at that, what we don't have is the luxury of data. We don't have data. So the first thing for us is the founders and the founder team. We have this concept for pre seed. Is it a minimum fundable, like okay.

We spoke on Monday, it's now Thursday. What's happened? And we're not looking for like everything is up and to the right. Like we're pre seed investors. We know. And then there's you know, multiple investments on the wall and many of the majority of them have failed and truly like the most important things or the most like.

And once you pass that bar, it's just.

Are you able to share what that bar is?

Well, if you're building a technical company and you come and you're just two business people and you need to come.

Useful data that we've captured is like when things haven't worked out and so we want to hear about the progress.

Please that with a, you know, very technical person. Do you have leads on that? Who would that be? Have you talked to people who are interested and at least some aspect of that. So. Or if you're building in a very specific vertical, how are you Going to gain the knowledge of that. We don't really necessarily require to you have that knowledge. I built a company software defined networking when I started that. No, no background in that either. But have they done the customer discovery?

Have they done talk about how fast you're moving? Because to the point I mentioned earlier of like yes it does help to be come to the table with some level of expertise especially if you're going after a vertical. But that only matters in T0. From then on it's all about how quickly you can move. So t +1 like it completely changes.

People discovered a pain point that people you know have that part of the work has to be done before.

Something that we talk about a lot internally is a founder's ability to zoom.

In and out and this then we have also various other metrics on the founders to just standardize within our firm. But one key component is that when I meet with you, I should not be able to sleep at night until I contact you again. So that comes from not only wanting to solve a problem, just why in the core. Why is it a core thing for you that you're doing this startup? It's to be hard, it's going to be lots of ups and downs and you know what would get you through.

That kind of goes back to the idea of thinking about that beach head and that the very tactical, you know, this is what I'm going to be doing in the next three months. And then also this is the big vision that I have and this is how the docs connect. They might not, but at least you have like that ability to like put your head down and get stuff done but then every now and then lift it back up to see wait, are we even going in the right direction finally because again ups and downs in.

Addition to the monetary risk, you know, returns and all that. Why are you, why are you building this company? So that's one component that we look at. Obviously the market has to be perhaps the potential to be big for our kind of firms. Again a lot of good companies can be built and they don't have to be venture backable for tier one ventures. You need to have some version of that and the third company, the product.

And it's all about potential at our stage and there's really nothing there. A lot of our process has to do with brain division.

One thing I want to mention about AI is that we go in. It happens a lot that you go from something that's being impossible and then.

Becomes possible storming sessions which not only will Include us. But maybe if there's someone in the network that's a go to market expert in whatever.

As a founder, when you look at it and you want to build in that area, did it become easy or is it still very difficult and just become. Become possible? I usually gravitate towards it was impossible. It's very difficult to build still.

Whatever it is that you're doing, we'll bring that person on and it's a working session because we want to see how.

But now it just became possible. This is my vision, this is why I'm building it. That's why I understand my customer.

How you work, how you think, how you learn. Because that's, that's all there is at that point.

Yeah, so some of the things we look at.

So. So for a firm like Pear, you guys are agnostic, right? You have.

Yeah, I honestly don't really know how to answer this question because like there's.

Sumer, you do like health like lots of non AI. And so for the non AI, for the agnostic VCs on the panel, do you look at AI? Do you have some different ways you look at AI startups specifically agentic different than the rest of your portfolio?

Like I can tell you a bunch of traits. I could be like yeah, they're hyper, curious, like flexible on the detail, stubborn, the vision, hyper ambitious and whatnot. But like at the end of the day it really comes down to feel Olio.

If it's a founder building an e comm or B2C. Is there different layers or levels or.

Of like individual themselves and what they actually want to accomplish and do you think they'll actually do it? And I guess like for, for us.

Like criteria we are not agnostic, we just broad investors. So we have specialists on the KV.

Is kind of a weird firm. So we actually don't really care about returns. Right. Like if you ask a node why he started kv, he'd be like if I wanted the best firm, I would have just stayed at Kleiner and taken over. That was on the table.

The team for each area of the, of the stack of, of the. I mean each area we have a bio partner, we have a healthcare partner. But as it happens a lot of the companies have an AI.

But like I want to have fun and the most fun is working with generational founders building generational companies. And then like yeah, you have to be in like the top five funds or whatever because then you have to work with LPs if you're not. And LPs are boring to work with. But like that's really kind of the crossover point is like do we think.

This is someone that components right now? So what we look at for AI.

Can be like a Sam Altman or like a Patrick Collison and not like now. Right.

Because like again it's already been mentioned a few times. Why are you building what's the value you're going to be adding?

When we met Patrick or when Keith met him back in the day, it's like wasn't the Patrick you know today. Right.

I think what inflection point are you seeing? If you're a very technical founder? What. What is the invention that opens up some thing that was impossible. Now it becomes possible and you're going to be spending the next five, 10 years.

But like do we think they have the traits that can get them there through trial by fire and and a lot of that is like I don't know, you go to an IC and.

Like six people say like building that impossibility into a possibility in a world that didn't exist.

So all of yeah, no way. This person's like terrible. And then one person's like no, I think it actually work and then they're right, right. So it's like kind of hard to actually say but a lot of it is by feel and by like what is you actually want to accomplish? Is it big?

Those typical things? You know it depends on. You know really depends on the opportunity and what they're looking at and what the founder market fit is.

Gotcha.

Yeah, I was trying to think of like characteristics that we look for in teams and I really liked the point on kind of infinite curiosity and just wanting to learn and do.

Does somebody believe in you and think you will actually run through walls to get there?

Yeah, I honestly don't have a lot to add on top of that.

That's good. That'll save some extra time for questions. But I do want to hear you. You. No, no, I do want to. I wasn't trying to cut you off but I was going to encourage you to go quick especially maybe you could talk about it from your program like.

Something that would we look for. And it's very much part of our diligence process is like the deltas over time. So like okay, we spoke on Monday, it's now Thursday. What's happened? And we're not looking for like everything is up and to the right. Like we're pre seed investors. We know and then there's multiple investments on the wall and many of the majority of them have failed. And truly like the most important things are the most.

What do you look for for the program now? You guys have such a great program and a high bar actually.

I mean we think of this as not substantially different. We want to work, I think as John said, like we want to work with the best people that we can find. We try actually. I would love feedback on this. So we like launched this program embed. It's $150,000 on an uncapped MFN note. So it converts to whatever your next round is. Our goal with this is to try and make it like as generous for people as possible. So if you look at these useful.

Data that we've captured is like when things haven't worked out. And so we want to hear about the progress, about how fast you're moving. Because to the point I mentioned earlier of like, yes, it does help to come to the table with some level of expertise, especially if you're going after a vertical. But that only matters in T0. From then on it's all about how quickly you can move. So t +1 like it completely changes.

Like I would never apply to this. Would love that, would love to understand why that is and how we can make this more friendly. But the basic premise is like the motivation was actually the same. Like we're a year and a half old, two year old venture fund with the same ambitious mission of we want to work with the best people we can and we recognize that as a relatively smaller fund with like a couple hundred million dollars, it's still a challenge to do that. As a.

Something that we talk about a lot internally is a founder's ability to zoom in and out. And this kind of goes back to the idea of thinking about that being beach head and that the very tactical, you know, this is what I'm going to be doing in the next three months. And then also this is the big vision that I have and this is how the docs connect. They might not, but at least you have like that ability to like put your head down and get stuff done. But then every now and then coverage.

And scale of number of people that we have working here. And so the way that we try and do that is more program based investing, smaller checks, but we get to work with really great people people. And then more broadly I think the general goal is still work with best people that we can find. We try and make bets early on, people that we think will be great. I think it's easy to get lost in specificity on market. It's Pretty hard to predict what markets will be or what products will evolve into. And so we try and focus. I think we're largely people driven investors with like secondary theses.

Awesome. So we have time for one or two quick questions. If you have something that'd be, you know, not super personal that be widely applicable to the audience, you could ask one on one questions. All right, from our founder panel.

Hey guys. So one thing that was really interesting was the AGI timeline and I was curious to ask which markets you guys.

Think will be created or destroyed.

The post AGI lift it back up.

To see, wait, are we even going in the right direction finally? Because again, it's all about potential at our stage and there's really nothing there. A lot of our process has to do with brainstorming sessions which not only will include us, but maybe if there's someone in the network that's a go to market expert in whatever it is that you're doing, we'll bring that person on and it's a working session because we want to see how.

I don't know if that was a.

Quick how you work, how you think, how you learn, because that's all there is at that point.

Yeah.

Yeah. I honestly don't really know how to answer this question because like there's like I can tell you a bunch of traits. I could be like, yeah, they're hyper, curious, like flexible on the detail, stubborn the vision, hyper ambitious and whatnot. But like at the end of the day it really comes down to feel one.

But that was a good one I guess.

Like. Okay, one quick comment. I generally find it really hard to read reason about what post AGI means and I think part of this is just there's like a conflation of a bunch of different words. Like when people say take off they mean like a bunch of different things and there is not like a consistent definition here. I would bet money. Although who knows. I guess if I'm wrong, it doesn't matter. I would bet money that there was still like this of like the individual.

Themselves and what they actually want to accomplish. And do you think they'll actually do.

Businesses to be built post AGI and that problems like delivery likely still exist. Like the, the OpenAI definition of AGI is like a, a model that you would have do any work that any. That any human would do, including like physical labor. I have two beliefs. One is like that seems pretty far.

Away and I guess like for, for us like KV is kind of a weird firm so we actually don't really care about returns, right? Like, if you asked Benode why he started kv, he'd be like, if I wanted the best firm, I would have just stayed at Kleiner and taken over. That was on the table. But, like, I want to have fun. And the most fun.

Robotics seems terribly hard. It seems terribly hard to get data in lots of different dimensions. And then two, there's probably still a business to be built after. Like, if intelligence becomes much cheaper. There are businesses other than just like, energy and GPUs. So I just find it really hard to reason. I assume it's really far away and I'm like, kind of just mental blinders and just do my job.

Is working with generational founders building generational companies. And then like, yeah, you have to be in, like the top five funds or whatever, because then you have to work with LPs if you're not. And LPs are boring to work with. But, like, that's really kind of the crossover point is like, do we think this is someone that can be like a Sam Altman or like a Patrick Collison and not like, now? Right? Because, like, I'll just say I don't really have an opinion or thoughts. It's like, to me, this is like Silicon Valley soap opera stuff. I just don't want to get involved.

I haven't thought about it personally. I think what I think about the most is like, what does AGI even mean to go back to? Like, you know, how do we define it? Likely it means a lot of different things to people, even here in this room. And I just think there's a lot of work to do in order to.

Even at least when we met Patrick or when Keith met him back in the day, it's like, wasn't the Patrick, you know, today, Right. But, like, do we think they have the traits that can get them there through trial by fire? And a lot of that is like, I don't know, you go to an IC and like six people say, yeah, no way. This person's like, terrible. And then one person's like, no, I think it actually worked. And then they're right, right. So it's like, kind of hard to actually say, but a lot of.

To get to what. How I think about it, which, you know, there's a lot of data issues still happening. We need to capture infinitely more context. Part of it is new infrastructures. I just. I haven't thought about what happens after yet, so.

I don't have any good thoughts on it either. But one thing that I would Say is that it depends if we are. We are the owners or the AGI is the owner of the world.

If they don't, it is by feel and by like what is you actually want to accomplish? Is it big?

I mean who is going to be buying anything?

Strong thoughts on it or.

And one thing you see between now and Ag Idol is, is that everything is getting faster. The enterprises become more efficient. There are technology that you cannot afford not to have. I don't know what does that to the workforce but.

And does somebody believe in you and think you will actually run through walls to get there?

Yeah, I honestly don't have a lot to add on top of that.

That's good. That'll save some extra time for questions. But I do want to hear. No, no, I do want to. I wasn't trying to cut you off.

My guess is that at least in the short term it actually may expand the workforce because things get faster. You still need the same number of people to compete but doing different tasks.

I think the post AGI world is going to have a lot of tech debt basically.

All right on that. We'll end on that. I want a round of applause for our VC panel but I was going to encourage you to go quick especially maybe you could talk about it from your program. Like what do you look for for the program now? You guys have such a great program and a high bar actually, I mean.

We think of this as not substantially different. We want to work, I think as John said, like we want to work with the best people that we can find. We try actually. I would love feedback on this. So we launched this program embed. It's $150,000 on an uncapped MFN note. So it converts at whatever your next round is. Our goal with this is to try and make it like as generous for people as possible. So if you look at these terms and you're like, I would never apply to this. Would love that. Would love to understand why that is and how we can make this more friendly. But the basic premise is like the motivation was actually the same. Like we're a year and a half old, two year old venture fund with the same ambitious mission of we want to work with the best people we can. And we recognize that like as a relatively smaller fund with like a couple hundred million dollars, it's still a challenge to do that at as a coverage and scale of number of people that we have working here. And so the way that we try and do that is more program based investing, smaller checks. But we get to work with really great people. And then more broadly, I think the general goal is still work with best people that we can find. We try and make bets early on, people that we think will be great. I think it's easy to get lost in specificity on market. It's pretty hard to predict what markets will be or what products will evolve into. And so we try and focus. I think we're largely people driven investors with like secondary theses.

Awesome. So we have time for one or two quick questions. If you have something that'd be, you know, not super personal, that be widely applicable to the audience, you could ask one on one questions. All right, from our founder panel.

Hey guys. So one thing that was really interesting was the AGI timeline. And I was curious to ask which markets you guys think will be created or destroyed in the post AGI era.

I don't know if that was a quick one, but that was a good.

One, I guess like. Okay, one quick comment. I generally find it really hard to reason about what post AGI means. And I think part of this is just there's like a conflation of a bunch of different words. Like when people say take off, they mean like a bunch of different things. And there is not like a consistent definition here. I would bet money, although, who knows? I guess if I'm wrong, it doesn't matter. I would bet money that there is still like businesses to be built post AGI and that problems like delivery likely still exist. Like the OpenAI definition of AGI is like a model that you would have do any work that any, that any human would do, including like physical labor. I have two beliefs. One is like, that seems pretty far away. Robotics seems durably hard. It seems terribly hard to get data in lots of different dimensions. And then two, there's probably still a business to be built after. Like if intelligence becomes much cheaper. There are businesses other than just like energy and GPUs. So I just find it really hard to reason. I assume it's really far away and I'm like kind of just mental blinders and just do my job.

I'll just say I don't really have an opinion or thoughts. It's like to me this is like Silicon Valley soap opera stuff. I just don't want to get involved.

I haven't thought about it. Personally. I think what I think about the most is like, what does AGI even mean? To go back to, how do we define it likely? It means a lot of different things to people even here in this room. And I just think there's a lot of work to do in order to even at least to get to how I think about it, which there's a lot of data issues still happening. We need to capture infinitely more context. Part of it is new infrastructures. I haven't thought about what happens after yet.

I don't have any good thoughts on it either. But one thing that I would say is that it depends if we are. We are the owners or the AGI is the owner of the world. If they don't, I mean, who is going to be buying anything?

Strong thoughts on it or.

And one thing we see between now and AGI, though is that everything is getting faster. The enterprises become more efficient. There are technology that you cannot afford not to have have. I don't know what does that to the workforce, but my guess is that at least in the short term, it actually may expand the workforce because things get faster. You still need the same number of people to compete but doing different tasks.

I think the post AGI world is going to have a lot of tech debt, basically.

All right on that. We'll end on that. I want a round of applause for our VC panel.