Why AI Agents Are Eating Enterprise Software | Agentic EP 03 Founders Panel

Everyone probably knows what an LLM is at this point, but what really agentic is if you don't specify what the sequence of events they need to do, you just say solve this. And the agent reasons about how to actually get to that endpoint and then reasons about the skill sets that it has.

A lot of people think, oh, let's just use OpenAI tool calling and it'll work. It doesn't. Where you invest your millions of dollars and where you develop a plan for development is probably the most critical step towards building something of value towards customers.

What is really great about this platform shift, that AI agents them also very close to AI workers. And that's really easy for enterprise clients to understand.

The thing that works for us is that we solve a very specific problem in a very specific vertical. Turns out you can take that technology and apply it downstream to more general purpose problems.

The big value that you'll be able to give to customers over general agents is going to be in the outer tail of distribution of tasks.

It's so early and it's evolving so quickly that it's just really hard to figure out like where the puck is really going to go. Like we know it's like in some direction of night. 80% AI automation and like 10% human involvement. That's going to be the big first milestone.

The customer audience about AI. I find that generally people have not adopted it and are not familiar with it. Too many tools, failed projects, burnout. I don't want another AI tool. A big part of the success in Agentic is going to be based on building agents that actually work. 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.

Hey.

Hey.

Welcome to the Gentek Podcast. So if you enjoyed the star studded SF VCs in our last episode, you're going to love this one too. But this time we're shifting perspectives and diving into a founder panel featuring some of the brightest minds building innovative AI startups today. This was recorded live at our Agentic and Vertical AI event in sf. And this episode brings you the stories, insights and strategies of founders shaping the future of AI. So without further ado, let's jump into the episode. All right, welcome to our founder panel. Really excited. So let's Go ahead and kick right off. And so can you guys just quickly introduce yourselves? What's your name, company where you based and what's your uvp? And any initial thoughts and insights from what you heard on the VC panel?

Hello.

Hey.

So I'm Jackie, I'm the head of AI. And Nullify. And Nullify. We're building AI application security Engineer. So very much focused on agentic use cases for enterprise. And like when we're networking, we'd love to talk to any agent builders, anyone really technical. Love to chat and share notes. So yeah, on the VC plan, I actually thought it was like really, really. I actually really agree to a lot of the thesis that they stated. I think a lot of the issues that have come up with agentic AI is that people go too broad too quickly. You do need to narrow down, you do need to figure out like, you know, what those issues are. And I only found out a lot of the problems and a lot of assumptions that we made once we productionized, once we've tested it with customers, the design partners can be really good as well. But in areas like cyber security that are also green and AI is also green, there's a lot of, you know, reimagining from first principles about how you build agents for these use cases.

Hey guys, My name is Aria Atar. I'm the co founder and CEO of a company called tensorstacks. We're building autonomous data scientists. So they're able to help you build production scale data pipelines, ML pipelines and so on. We're based out of the Bay Area. I tend to agree with a lot of the stuff that the VC panel also had as well, especially the notion of like vertical agents. I think that's gonna be really, really important to focus down. Our unique value proposition is that we're gonna, we're basically horizontally scaling your data science team.

Right.

And integrating directly into whatever data stacks you're already using. Right. So if you're using something like Airflow and Snowflake, we're not gonna come in and, you know, rip and replace that. We can just drop into your existing environments, run them the exact same way that your existing engineering team would.

Hi, I'm Max. I'm the co founder and the CEO of Mayhem. We are building a automated testing platform for AI applications. So chatbots, voice bots, and now AI agents as we learned is yet another tool. I need to remind people of that. What we do is we simulate users or simulate interactions. So we simulate basically the human side to allow coming to the UVP developers to go 100 times fast about debugging their, their AI applications doing like a thousand times more rigorous search of where flaws or weaknesses could be in AI applications given that they are probabilistic. I have a PhD in well natural language processing. My co founder in building AI safety simulations for self driving agents. Someone mentioned the Waymo. So we're bringing that system of simulation and therefore going then through security, safety and quality assurance to the level to the, the world of AI. My hot take from the the the previous panel actually agreed with like most of what they said but like one, one thing is I, I think to sound a little controversial I think no one cares about AI. So if you were searching for a solution, any business, anyone including myself, I'm not hoping hopefully there's an AI solution for it. I just hope for a solution and someone takes it away. But where everyone thinks about AI is a lot of about the downside. So when you say hey I'm using AI people get anxious and get can I trust this? So I think we need to work a lot on how, how can we get and keep this controllable and safe and build solutions that work not just because they are, because they're useful for humanity.

Hey, I'm Saurabh Sharma, I'm the CPO at a company called u dot com. Our focus is building a productivity engine to drive AI based business transformation. You know one of our value propositions is that we started off as a consumer company so we've been doing this at scale for a long time now and we can show that, you know, provable results through our consumer site. One of the things that really caught my attention at the earlier panel was this folk this idea that we're in the early stages, right. A lot of us in this room have been tracking every development and working deep in this field for two years. It feels like a lifetime but for most businesses it's not a lifetime and they're still very early at understanding what this technology is, where the ROI is and they're not rushing into it, right? They're going to make ROI based decisions and some of them are starting to jump off the sidelines but there's still a lot more to go.

Hi, my name is Jonas, I'm one of the co founders of BIM AI and we build autonomous agents for all sorts of back office operations. So we essentially go to companies that have a BPO that have teams doing stuff and we build agents that essentially can do the same job and we do this for enterprise clients and a couple of mid sized companies. And I think what I like people and also our clients, they don't really care too much about AI. A lot of people are very excited about AI agents and they're looking for solutions. But what is really great about this platform shift, that AI agents, we see them also very close to AI workers, so very much resembling humans. And that's really easy for enterprise clients to understand. They're like, hey, look, I have a team right now of 80 people doing that stuff. It doesn't make a lot of sense. Nobody really wants to do the job. And maybe I also have problems actually hiring people for this job. And this makes us very easy also with all of the technical definitions, etc. To sell this, which is very beautiful on a go to market perspective.

Hi everybody, my name is Asher Risky. I'm the co founder and CEO of Bounty AI. We're building autonomous go to market agents is essentially what it is. I think the only distinction, obviously give us the work that you and your reps and your anybody on your go to market team doesn't want to do. We'll do that for you. The only distinction is that you're not just getting an AI teammate, you're also getting a human teammate that's involved in that experience as part of that. Because we're realizing that we're really, really far away from all of the things that we heard in the last panel around truly autonomous agents. So it's like, let's just not wait for that. Let's augment with humans and provide that through an experience. I'm still forming some hot takes, I guess. I think the only maybe slightly contrarian thing that I've thought about is the thing that you get with this agentic technology. So I think a lot of the scorecard that we've heard some of the VCs say, not just in here, but just generally is like, I feel like they're applying 2010, 2014 scorecards to measure companies in 2024. And really what that means is, yeah, it's obvious. The advice is obvious. Start narrow, start horizontal. But the thing that you get with this agentec technology is flexibility, right? So if like there's market pull that's happening in a bunch of different directions, that's where you're gonna go, hey, there's a problem, we're willing to spend money on it and suddenly you've gone from being a really, really vertical company, horizontal company, much, much faster, right? So maybe it's not like three years out, maybe it's like six months, right? And that's just. And so it's just that flexibility that you get with this technology. I think that's the big learning here. And yes, we'll see it. We'll see how it plays. There's so much, so much stuff that needs to get played out to kind of form strong opinions. And so the only thing that I think we can do as founders is just be in learning, in curiosity mode.

So, yeah, you kind of ended off on the vertical versus horizontal debate, and I'm eager to jump into that, but I'll save that juicy topic for a little later. That was good of you to be willing to take a hot take on that. So it'll be interesting to find out. But yeah, I mean, why don't we kind of back up and hear a little bit of your origin stories. We don't have to go in the same order. Just, you know, whoever wants to share. So, like, you know, what's. What does agentic AI mean to you and how did you land that, starting an agentic startup?

Maybe I'll just start. So we actually founded the company in 2022 before ChatGPT was launched. And I founded a couple of companies before the last one turned into unicorn and got quite big. And then afterwards I did some fun stuff like yoga and meditation. And we're like, okay, what are we going to build next? And then my previous companies, we saw that AI is growing exponentially, it's getting better rapidly. And we also saw that humans are not developing that fast. And then we also saw that essentially AI can do all of the things that human can do. And then we said, well, if that's the case, probably it makes a lot of sense to build AI that kind of is very similar to humans. And then we started building these entities, which back then weren't really called AI agents. And we started building them and testing them with enterprise clients. And we saw, well, demos are really cool, but they don't really work in production. They're notoriously unreliable. Sometimes they do this, sometimes do that. So we spent really 23 on making sure that it works for enterprise clients. It was also a lot of, like, taking a lot of the flexibility out of the system, because enterprise clients want an agent or they want a software or whatever, they want to do it in a specific way. So we had to train our agents to behave in that very specific patterns. And this is mostly what we did in 23. And then we actually started to see that we can build these agents on very narrow things and then we can broaden them up. And then we can actually fully put them on autonomous mode. In the beginning you have a human supervisor that does that, but then after a while you kind of build trust with the system and then you can put it fully on autonomous mode, which is then great to see. And this is what we now saw. So a lot of like the hype, take it out, very basic and then expand, which massively helps with enterprise clients.

Yeah, I can go ahead and follow on with that. So we initially started the company building traditional MLOPs and infrastructure. And one of the main problems that we were personally running into during onboarding of a lot of our customers was we're building custom data pipelines for every single company that we're onboarding. And this was unbelievably time consuming for us to do. So over one weekend we just went ahead and built like a very basic autonomous agent that would help us clean up some, some data. And the value that we saw from this internal tool was something that we, that we had personally never seen before with any technology. And as we started showing this internal tool to more and more companies out there, it became very obvious that this is a problem space that is very, very broad and there's a lot more interest in this than we initially expected. So kind of moved on from that and made the autonomous agents the core product. So it started off initially as an internal tool for ourselves and became the core product.

There's a number of famous stories that start like that. So hopefully yours will be one of those as well.

All right, yeah.

Why don't we keep going and hearing from everybody else?

Yeah, sure. So first question on, like how do we actually find agents and how do we think about this? I think what, me coming sort of from academia and then going into startups, probably I'm coming up with a super academic definition now, but for me it's like everyone probably knows what an LLM is at this point. It's a machine that you give it some input, it just draws from its parametric knowledge of what it's been trained on, gives you an output. And then I think the next wave were chatbots, like a customer support bot, I would want to say, just detached to some other knowledge base. So you ask it again a question. But it could not only go into its own parametric knowledge that's been trained on, but also the company specific knowledge that's stored in this database and give you a response. But that was sort of it.

Right?

And then we gave further functionality to these things. They can do function calling if you ask them what the weather is they can actually go to a weather app and retrieve that information, give that to you. But I think what really is agentic is if you don't specify what the sequences of events they need to do or what they actually do is you just say, hey, solve this. And the agent reasons about how to actually get to that endpoint and then reasons about the skill sets that it has that could be function calling, could go to the RAC database and applies the first skill and then it reasons about its own results and then applies the next skill according to what has happened. And that does that autonomously until it reasons that the task has been accomplished, which is an order of magnitude more flexible, but also more sort of chaotic workflow.

Thank you.

But that is, in my mind an agent and how we got into this. Again, I came out of academia, my co founder as well, but we saw that that's actually true for chatbots or agentic systems. There is a really fairly complex workflow that everyone needs to build. And I did this as a researcher before. Tinkering around with these things is incredibly time consuming. Probably any one of you will agree that figuring out the right prompt, figuring out the right sort of engineering for these systems is super time consuming. And at some point you can't actually test that deeply because as a human you only have so much time in your, in your day or in your week. So this is why we started actually helping other AI builders and developers, giving them a tool to automate this and like, you know, simulate, so to speak, the users to have that, you know, orders of magnitude faster, have an AI basically simulate the human that ultimately would use, use that tool to see where it automatically breaks, where then the human can go in and improve that tool. So this is how we got into our space.

Yeah.

So the way I think about agentic AI is there's typically. Well, currently I think there's three types of AI companies and startups right now. One is in the model AGI research layer, so foundation model companies like magic.dev, openAI and then you have LLM driven algorithms. So typically like the Devin Sakana AIs, they've invested heavily on algorithms on top of LLMs and then sort of like where we sit, nullify. And a third type of company is integrating AI with existing environments. We have our own name internally for it called Agent os and really it's making the LLMs highly operational within today's current environments. And that's sort of where the landscape of AI seems to be shifting up. We've been very deliberate with where we placed our investments and where we place our time. And I think a lot of people very quickly, and this is my hot take is a lot of people quickly think oh let's just use OpenAI tool calling and it'll work. It doesn't. And most importantly, where you invest your millions of dollars and where you develop a plan for development is probably like the most critical step towards building something of value towards customers.

So you.com had a little bit of a different origin story. It actually started off as a search engine and our original vision was to create a search and do, what we call the search and do engine. Turns out that, you know, once we built that search engine aspect there was a lot of interest from the business community on that do engine side, what we now call the agentic AI side. And we started exploring that area and we found a lot of traction with media publishers, with pharma, both on relatively simple agents but also multi step chain of thought type of agents and we found a good degree of success there that we're continuing to build into as time goes on.

So this is Bounty is my second company prior to this built another company called Blameless and what we were trying to do was automate a very, very specific use case and workflow around managing outages and incidents. Right? So that was kind of the first foray into hey, how do you break apart a really well defined or just somewhat complex workflow and then how do you figure out how to automate it? And then when we started Bounty we started as a very different company which was like, hey, let's ingest signals from all of your internal data and then use that to arm the go to market teams with some information, some intelligence. And to Saurabh's point where we saw a lot of excitement and interest was just like the do part. Okay, like now that I have this information, the signal that's great, that's commodity, thank you for giving it to me. But like what do I do that's still like you've just raised my anxiety level because you've given me something to do that's painful. And so that's kind of when we shifted into the agentic, you know, sort of domain and said like hey, we've got this really kind of core infrastructure for the other part of the product. I think the again, maybe a little bit of a contrarian ish view is like, I think it's just what my learning, our learning so far on the agentix side is that it's so early and it's evolving so quickly that it's just really hard to figure out where the puck is really going to go. We know it's in some direction of maybe 90% AI automation and 10% human involvement. I think that's going to be the big first milestone. And then maybe, or not we'll have AGI. Who knows, maybe it's like a big black hole after that everybody dies. I don't know. But yeah, the whole sort of agentic sort of space, what we realized is that hey, what's really important is the ability to build tooling that allows us to rapidly iterate and learn from workflows and humans very, very quickly. So it's almost like if folks have used LaunchDarkly and a bunch of CICD platforms. So there's these ideas and concepts from the world of DevOps that was like really, really popular, you know, that you can like. We've actually taken a lot of those to build our Agentic platform layer. And then what we're doing is we're building user facing apps sort of on top of that, but the ability to morph very quickly. And again, I know you we're going to talk about vertical versus horizontal, which is why I'm like, you know, like once you have that capability you can't really tell the difference between vertical and horizontal anymore because you're just going to go to wherever the, the pull is.

Yeah. So one of the things that I think was talked about in the last session and I think we could all agree on, is that a big part of the success in Agentic is going to be based on building agents that actually work. Right. So like in your case, you know, it's got to be able to prevent the, the app security issue. Right. In Aria's case, it's got to be able to build the machine learning model and be able to have the right insights. Right. And so on. And in our prep calls, you guys, a number of you guys mentioned that that's kind of been like your biggest priority and biggest lessons learned. So we'd love to hear from you guys. Maybe we could start with Aria on this one. So you know what, how did you approach building agents that actually work with the high degree of efficacy and what were some challenges and lessons learned along the way?

Yeah, so I think this is a very important question because it's very helpful to take a little bit of a step back and actually think about like how do LLMs work and what are they doing? Right. So during pre training they're basically learning some Distribution of some data that we fed it for pre training and a bunch of data for post training. Right? And as a vertical agent company, one of the most important notions is that the big value that you'll be able to give to customers is over. General agents is going to be in the outer tail of distribution of tasks that your agents are doing, right? So if you think about like your agent's performance as, think of it like a decay curve, right? Where you have all the way to the left of the graph the most reliable tasks that any general agent can do, right? And as you go further into the tail of distribution of tasks, the more complex the tasks get, right, and the harder it is for general agents to do that. So what you have to solve for is that further outer tail of distribution of tasks. So this makes it. So there's two notions of two different ways to do this, right? We can either take enterprise existing workflows and existing tooling, like he mentioned, and get a bunch of like let's say synthetic data of sequences of tasks that these elements are doing and fine tune some model to use the existing software infrastructure that exists. This is typically a lot harder to do because you need a lot of diverse data to be able to fine tune an ll. Or you can go the complete opposite direction and in a way think of it like representing existing software infrastructure in a way that's in distribution with an LLM's training and what they're good at doing, right? This is that exact agent OS that he's talking about, some sort of middle layer that makes it very easy for these agents to interface into existing software. That's exactly what we've been very, very focused on. And that so far has been very successful, especially for, let's say, like building data pipelines, right? There's a lot of tooling out there. They're very complex to use. But by building a middle layer that makes it very easy for agents to use set infrastructure, these agents end up becoming very, very reliable very quickly. So it makes it more of an engineering problem rather than a machine learning problem of trying to capture a wide enough distribution of set tasks.

That's great. Anybody else want to tackle the accuracy?

I think the thing that we found to be super successful is just a constant feedback loop with your customer, with us, with the data set, with eval, we need that constant evaluation, we need that constant feedback from customers. And we found that to be just super critical in building great solutions. Now some of the problem with that is that sometimes you can overfit towards a particular Customer set. And so again, this just goes back to basic product development. We don't want to overfit. We want to have a set of design partners and be in constant back and forth communication with them around making sure that we're building a system that, that meets their exacting needs.

Quick, I'll do. Yeah, quick thought. I think again maybe contrarian throw humans at the problem is kind of how I see it. It's like if you got a workflow, you can either spend your $2 to $5 million seed round trying to build all of the technology infrastructure and all of this stuff around it to really kind of get the tuning right or take what's already off the shelf, figure out some automation, some workflow and just like inject humans into it. Because that can do that fine tuning, that feedback loop, that prompt engineering like and just make that invisible to the customer. Customer doesn't care. Customer, you know, users only care about, you know, to the point we heard earlier. It's just, I just want the outcome. I'm not looking for AI, I'm looking for somebody to solve my problem. So in many ways it just doesn't really matter what the, you know, if you, you can fall into a trap of just making the agents like just focusing on efficacy when you can solve the same out, get the same outcome by finding the right balance of agent efficacy and human in the loop.

So, so what you guys have done is, is packaged in the human in the loop in your go to market offering, right? So that kind of goes back to what the panel, the VC panel was talking about in terms of focusing on services and outcomes. And so do you want to expand on that a little bit? And I'm curious if anybody, I don't think any of you guys are packaging humans in the loop in your solution, but I'd be curious if you are.

So maybe I can just answer from our side. So on the one side we work a lot with our clients and we need also the business owner ultimately to own the process and to give us feedback. Because one, there is information about how to solve certain problems.

Problems.

But there's a lot of procedural knowledge as well in these companies that you need to understand in order to really build agents that can do the full task. When we work with clients, we on the one side sell our platform, we sell the agents, but we also do fairly high touch sales. So essentially solution sales where we sell them consulting as well to a certain extent. And we see that it really helps us in the beginning to make sure that we get to These reliability and accuracy levels if we don't. And we have always, like people wanting to try out the platform, and I want to build a certain case that just barely works. So we really need to focus on very specific problems and then work with our clients. And to be honest, in the beginning it took us months to get certain agents to work, and now we got this down to a couple of hours or days, which is really great. So the effort on our side really is minimized.

Yeah.

Just to add on to that, that's exactly what we've seen. While the solution itself is not human in the loop, the whole package, the whole bundle is solution is human in the loop. What we found is that a lot of our clients would be one or two super users. Right. And that they would over time expand that knowledge to the rest of their organization. That'd be great for us. But when we added in a learning component, when we added in almost like a consulting component to help that organization understand how to use this usage took off much faster. So that's critical part, especially in these early days where people don't. Again, the point that we've all been making, people don't understand or care about the AI. They care about the solutions. And then showing them what's possible is a big part of it.

I think I have some disagreements with the rest of the panel. Just like typically, I agree that there's a general idea that you want to invest in testing and evals, but it's very difficult to know which evals to invest in, especially with like, vertical AI agents. And so we've very deliberately made the decision after testing not to invest in evals yet. And we're just seeing how they run in production, learning off those mistakes, and only investing in evals when there's a really, really clear, like really, really clear path forward that this is something that's not going to change. And a lot of these decisions are largely dictated by how you orchestrate your compound AI system, which is really how you orchestrate agents in general. And where a lot of people have made mistakes on building agents is assuming similar. Like Sierra AI's Tao bench paper is like, you discover that when you're doing really complex scenario modeling, especially for enterprise situations, the Markov decision trees are crazy. Like, the paths are like, they're not consistent, they're not reliable. But what agents are really, really good at is information synthesis and decision making, like getting to a right start, but they are not reliable. And like, you have to reprogram agents in such a way that they are Reliable and that's like an art in itself. And so that's our take and I think that we invested heavily in the software 2.0 layer where we're not treating agents as if they have to do good instruction following, we're treating agents as information synthesizers developing programs off that. And so like, and I think one thing that we've made a mistake in is that we've, we've let these agents as a result go haywire and we blew up 10k in bills over a day and it was nuts. But like, you know, it's one of the mistakes that we learned and like we're actively iterating towards anything else on.

The technical front before we move to the kind of customer and traction side of things.

No, one of the hot takes I have here is that I think traditional cloud managed infrastructure will bleed very heavily because of this agentic paradigm shift that's coming. Because if you think about it like this, right, a company like MongoDB, they have some open source offering for their database that you can deploy, but the value of the managed service is that you as a developer don't have to deal with managing that database and deploying it to the cloud. But once you have these semi general agents that can come in and build you production level infrastructure using open source tooling, a lot of these infrastructure platforms will bleed very, very hard. And this is, I don't think I've heard much of it so far.

Awesome. So let's go ahead and switch to the customer side of things, right? So in the last panel there was a lot of discussion. It was interesting for me to hear about this idea of enterprise burnout with AI. When I talk to CEO groups and executive boards, what I find is that a lot of them still haven't implemented it. They're still learning. Maybe not the Fortune 500, but everybody that's in the mid market. A lot of people still haven't gotten chatgpt or maybe they have Microsoft Copilot. But that from my voice of customer as I'm a keynote speaker and I travel around the country in the world and I speak to the customer audience about AI, I find that generally people have not adopted it and are not familiar with it. Not what like Andrew was saying as far as like too many tools, failed projects, burnout. I don't want another AI tool but I could see that that could be happening in different areas as well. So how are you guys approaching go to market and customer acquisition? You don't have to share specific numbers if you're not Comfortable. But how far along are you? You know, Char, you mentioned the pull. So like, you know, and I know you guys had a launch, but would love to hear about like, how, how is that going? What's working, what's not going, what's not working on the, on the customer acquisition side.

Yeah. Can I ask an audience question real quick? How many folks here are like early stage founders or thinking about founding a company or work for an early, like super early stage company? Decent amount. Okay. So the only reason I ask is because the lens that we look at, again, like if you think about the, the, the most precious asset for any startup is like your Runway. And especially when you're thinking about the cost of building versus the value that you can provide and the value that you can consume. So that's kind of like a very general high, like high framework. Right. If you've only got 2 to 5 million bucks, you got to make that 2 to 5 million bucks last a long time or you've got to start generating revenue very, very quickly. So where I'm trying to go with all of this is for us, you know, my co founder, Matt Cooley is very, very, he is the domain expert around go to market. So that's one, one great thing. But then also we practice what we preach. And so what we've done is we said, look, time to value. Like in this market that is not focused, like it's really not the technology that's going to differentiate because every, like there's a new agenting company launching every three hours. Right. So it's really not the technology that I think is going to. Maybe technology creates incremental benefits, but like the true benefit is going to be time to value. So for us it was like, how do we get, how do we deliver this outcome on a go to market site for you as a customer within minutes versus hours or days. So really what I'm talking about is how do you take an agentic experience and make Excel service? So you know, we launched out of stealth about three, four weeks ago. Three weeks ago. And you know, you can go on the website and you can basically get a taste of like, hey, through a free trial or product LED experience. Oh, holy crap. Like this thing can go do research for me, generate really personalized messaging for me and all that kind of stuff. Right. So I think that is really critical because the alternative, which we've tried by the way, is that, which is also, by the way a completely fair approach, which is the sales light approach. But the problem is that this, there's a lot of. To George's point, there is still a tremendous amount of education that you've got to do in the market, and that's fine. So you're waiting for the market to catch up. Maybe you're doing some education, which means you're raising money very different, differently. Right. Which means you're raising larger rounds of funding because you got to go and build the sales team, the solutions team and all that kind of stuff. And you can be very successful doing that as well. So those are some of like our learnings. We just took a very different path and we said very, very niche. You know, our ICP is early stage founders and company. You know, like the folks on this panel, maybe a lot of you, which is like, hey, get that agentic experience, but anybody on the planet can go and use this thing.

Right.

Regardless of where they are. And again, time to value execution. That's been our focus.

Yeah. So it's kind of interesting. U.com started off as a consumer focused company and with that we ended up building capability just like, you know, to build these agents directly on the site.

Right.

And that's. It's led to a lot of things, but one thing that it led to was it essentially acted as a showcase for our, for our sales led growth as well.

Right.

And so people would say, hey, great, I built this thing on your site. I've got it like, I've got it pretty far along the way, but I need. That last mile is always tough. The last mile of accuracy, the last mile of production level capability is all this tough. So that it essentially served as an inbound lead generation for our slg.

Right.

For our sales led growth. And that's been great because in a way that helped. There are still two things that we see. Number one is the people that come in through that consumer site, they're already aware, they're already prosumers, they already came there looking to build these agents. And so they were primed for our sales motion. But our sales motion is also finding, you know, as a point that was just made, there's a lot when people come and a lot of times people will come and say, hey, look, I'm trying to figure out how to navigate, you know, AI for my company. Right. And then that education piece comes in and it takes time, but what we found is that those deals can be very, you know, really great win wins for both parties as well.

Is Everybody familiar with you.com? have you guys tried it? A few people. Are you guys familiar with perplexity? So you.com is like perplexity but better if you ask me. Obviously they're one of our portfolio companies, so I have a little bit of bias. But yeah, you should give it a shot for your general purpose kind of perplexity replacement tool. Thanks. Were you going to say something on the traction side, customer acquisition? Just now going back to the question we were on around customer acquisition.

Well, we have a very different go to market than probably everyone here because we kind of try to enable others to use AI. We provide a quality assurance or a quality testing platform.

Right.

So maybe tying like just a few points, like some of you said, like getting the human in the loop for quality assurance or quality testing is really important. You actually said like just try it out and get it and figure out the problems of production, but that it really matters at what level of company stage you are. So when we talk to lots of seed companies, we often hear that as a seed company you have lots of problems. You don't want to get into proper quality assurance automation systems once you're at a large company enterprise level. One of the reasons why they have not adopted AI is well, there's many reasons but number one, they really lack the talent, the technical talent to build these things out and then have quality tests and quality assurance processes in place because they have that for all their old like deterministic software and they know what to do there. But they don't have that with this probabilistic software called AI. And the problem is it's not even about is it human oversight or is it some other system oversight. But they need to have some control, they need to have these things in place, reports in place, etc. To know, okay, we hedging against this risk and we know what to do. And I think a lot of your earlier point, a lot of education needs to be done there to say hey, like you can test this. We coming from the space a little bit from like self driving technology where we say hey look, self driving has come a long way through simulation and like testing things like humans are probabilistic agents we mock up on a bad day and because our company's a lot of money, you need to think about it similarly. But like also adopt your testing your quality assurance process similarly. So for us it's like a lot of education but going that hard way, I guess.

Yeah. So it's a very enterprise sales led growth kind of motion. Got it. Okay.

Yeah.

Maybe to add on that and what we see in the market, you have a lot of people right now that Have a mandate to try out AI and they come to you and be like, hey, we want to test out all of these AI solutions. And those are not really great customers most of the time because they just look around and they're like, oh yeah, we built this agent, we dump all of the data and then we do beautiful things. Right. There's this never happens and sales. What you have to do is you have to listen a lot. And when you actually start talking about their problems, then you see they have a lot of jobs to be done in their organization that you can actually cater to. And then the beautiful thing about agents is that you can say, hey look, you have a very specific job to be done. You should hire somebody for that and maybe you should hire an AI agent and then you can build very specific solutions and then it becomes really easy to do the sale. Yeah, but if you keep this fairly broad and it, it just doesn't work and you really have to narrow it down also with clients, specifically with enterprise clients that want to go in all sorts of different ways.

Yeah. One just. I wanted to highlight Sourabh's point around showcase. I think that's one thing that has been a big learning for us is that using our agents for ourselves and I think that applies for most folks here. There's nothing better than that being your sales pitch. Like this is how we for, for us, hey, this is how we prospect it to you. What do you think? Oh, that's great. Well, this is what you can do for your customers or hey, this is what we do for our security appsec workflow. What do you think? Oh, this is what. So that's actually, I think one of the, the really big things that's coming out of this agentic, you know, agentic workflows is how it sort of just unlocks you on so many different dimensions of the business, particularly on the go to market.

Yeah, let's stay with you. So let's go ahead and tackle the horizontal versus vertical. We didn't plan it this way, but if I'm correct, three of you guys are horizontal or at least started horizontal. Three of you guys are vertical. Is that right? I mean you're very focused on one niche, right? One, one, one problem to solve versus more of a platform.

So.

And again, I don't have a horse in the race. I'm just kind of curious. So what are your thoughts? And you know, we heard from the VC panel earlier, how are you thinking about where you've started and where you plan to go? And I think the General takeaway was, you know, if you're vertical, make sure there's a bigger picture and you can connect to the big picture. You know, I could see all three of you guys overlapping at some point. Right? Similar. Ish. So, and even if not with each other, there's plenty of competition in the market. So what are your thoughts? Why don't we make it interesting? So why don't we start with you, Char? What are your thoughts on this?

So I have a bunch of thoughts that I think are even like I'm conflicted by internally. So I'll say something now and then 30 seconds later I'll say the complete opposite and all of you will think I'm crazy. So first of all, horizontal versus vertical. I think it all depends on which lens you're looking at. I think we're starting vertical, which are vertical, being go to market and then like early stage founders and companies. So maybe it's a question of like, what is your vertical mark? Are you going after a vertical market or a vertical horizontal? Again, open question. I think the, the, the key thing in all of this stuff that was said in the, the VC panel was like, where are you lining up value and time to value and cost, like fastest? Like all of how are, where, where's that happening? And that's what determines, in my opinion, horizontal versus vertical. So it's like you approach it from a place of curiosity. As a founder, you're like, okay, you know, I've got this really awesome platform and I just really want to find the place where like things move fast. I can close deals quickly and I can start like, just, that's how I start winning. Because that's the reality of it. When you have money flowing in the bank, that's when you're winning. Nothing else. Like everything else is just kind of talk. On the flip side, I think with agentic, with agents, here's the good scary thing about it is that you can essentially tackle multiple verticals at the same time. If you wanted to, you have the, you have the capability to do it with like general purpose workers. Right? So you can go. So it's really attractive to say, like, oh my God, like I can build five companies at the same time, or, you know, I can build five products at the same time. It used to be really crazy to think that way, like five, six, seven years ago. I think it's less crazy because maybe it's still a little bit on the crazier side, but I think it's really starting to shift. And if I think as Founders, if we're able to show, like, hey, I can. If it took me 12 months to capture one vertical, then it took me nine months to capture another one. All those. All it means is that with agents, that timeline has shrunk rapidly right down to. And it all comes down to how much money can you spend on sales and marketing to go and capture marketing. Again, pretty controversial. Maybe a little bit. A mix of everything. Maybe didn't answer anybody's question.

So for me, this is like, personally one of my least favorite questions that I get from investors. Because the notion of vertical and horizontal is so subjective. Right. Like, what does that even mean?

What.

What is vertical? Right. Like, we see ourselves as vertical to a certain degree. Right. We're building autonomous data scientists. They can help you build a data pipeline and then train a machine learning model off of it. Is that horizontal? I guess to a certain degree it is within data science, but it's pretty vertical in terms of engineering tasks. Or even if we were to drill down even more. Right. We just focus on data engineering. What is vertical data engineering? Right. There's so many niches within data engineering that you could even further verticalize on that. I think it's like a very subjective lens to look at this through. So I think it really depends on the person that you're explaining it to. But for me, it's a tough question to answer properly to a general set of people that I talk to.

I just wanted to double down on this because I feel similar about it. Like the entire vertical versus horizontal. Yeah. We do quality assurance in a way. This is like one specific thing, but it actually applies throughout verticals. I think the vertical and horizontal definition is by. It's just. It's so arbitrary. And it helps as a thinking pattern. And I know VCs love it and there is a lot of looking back and finding patterns. But I think what actually is important, it's coming to earlier is like, it's not about selling AI, but solving a solution. Again, whether it's a vertical or horizontal, find something that you're extremely good at and other people are willing to pay money for it and then execute faster than anyone else. And then it doesn't matter where you are, as long as you have the passion for that and you're really good at doing that.

Yeah.

Kind of seconding that. Overall, I think what is important if, like, in our case, we built a horizontal platform so we can build different agents on our platform. Platform. Yeah. Nonetheless, now if we go and build one agent here, we build one agent there and we talk to that company and that company. That doesn't make any sense. So what we need nonetheless is focus. So we need to be very focused on our approach and we need to be like, if we build that particular agent, we need to be able to tell it, sell it to 10 other companies as well. Yeah, then it does make sense. Then it does make sense that I build agent one, I go to the second company and be like, hey, we built this for a, do you want to have it as well? Yeah. Then we have synergies there. Then it does make sense for us. So this focus is really important. The second, like kind of my history. So I built companies with rocket Internet before and there was like literally every month another country. So it was like bam, bam, bam, bam, bam. And kind of naturally, now obviously I build a platform where we can do something similar where we can say, okay, every month we're going to launch something, we're going to launch something and we're going to do it again at rapid speed. And so I think what you ultimately need to find the right balance with is focus because, because that really helps you to do stuff repeatedly at the same time, just speed, speed, speed, because then you can actually do more.

So I'd love to, instead of calling it horizontal and vertical, the way I tend to think about it is general purpose versus specific.

Right.

And in our world, what, what basically happened again, we started off as a consumer site, we started off as a general purpose consumer site, but we found lots of value for building for enterprises, specific solutions in particular vertical. And so what we realized for us, the thing that works for us is that we solve, you know, as you were saying, a very specific problem in a very specific vertical. Turns out you can take that technology and apply it downstream to more general purpose problems as well. It's not going to fix, it's not going to upgrade every general purpose problem, but some particular areas it will upgrade. And that's been a motion that's worked for us. So we're always looking when we go attack a new vertical or a new specific problem, is there an accruing advantage that makes everything around it better? Right. And so that's been what's worked for us as we've transitioned more and more towards enterprise offerings.

So I think I disagree with most people. So I think like I actually like horizontal vertical as an analogy towards like jobs to be done theory. So like the theory just states that you analyze every startup based on, instead of being like, oh, this is an idea, you analyze it based on what's the job to be done by that professional and that really targets use cases like Google Perplexity where tons of people have to do research, but it's a general purpose tool. I think that like with horizontal, the hard bit is having a unique enough insight such that your product offering will not be consumed by OpenAI. It's like really obvious that that's a good definition. If I was building in this space, I'll be shivering. Like there's a lot of what keeps me up at night would be like, oh, what's OpenAI strategy? What's their plan? Can I execute fast enough in such a way that their business model, you know, like is not going to move faster than me? It's a tough question to ask. And I think even with Perplexity, they're facing tough questions like Google's Gemini. Once it's sufficiently good, will it just outstrip growth in that place? And so I know these investors have to place a bet on the right startup, but properly thinking through what that's like, how you think through that paradigm is it's a complex question.

So just one line closes. Why don't we go in this order? Do you have any ask for the audience? Is there? Are you looking for customer? Are you looking for, are you recruiting? Are you fundraising? Let's start.

Yeah, I'll start. So I'm very technical. I'm building an AI agent space. If someone here is also building technically, I'd love to connect. Yeah, probably not looking to buy any tools. Just to be clear, no vendor pitches. Thanks.

I like that. Yeah.

If, if you have trouble building data pipelines, come talk to us. That's basically it.

If you build anything with AI, come use our product to make it better.

We're also hiring.

He stole my pitch. But no, I think, you know, we're, we're, we'd like to think we're on the cutting edge of many of the folks on the stage. Go to u.com check us out. You know, really, you know, really try those agents out, see what breaks them, see what makes them work really well. And when you break them, please let us know. We'd love to learn more.

Yeah, thanks also to you for these great conversations. So would love to continue this both on the investor side, but also then potentially on the talent side. So we're growing really fast. So happy to have these conversations. Looking forward.

Yeah, all of the above. And was I gonna say the. I saw there was a bunch of hands up early stage founders. I mean that's the hardest part of anything. The emotional journey, the physical journey that you have to go through, and I don't think that gets talked about enough. So if folks are looking for just somebody to listen to them and just talk about that and how painful it is, I can be your therapist for $0. Now, we have an AI for that. Or you can use that.

All right, you guys don't go anywhere. You guys don't go anywhere. Round of applause for our founder panel. It.