677th 1Mby1M Roundtable with David Hornik, Lobby Capital

Yeah, great to see you. Thanks for having me. That's a lot of gatherings. Congratulations.

Well, I guess I'd start with this sense that venture capital is the win, or is the loss, depending on who you are. Some people think it is success to raise venture capital as opposed to build a business, or it is a failure to raise venture capital because you've given away too much of your company. What I'd say to that is venture capital is a tool for businesses that require more capital than the business itself can generate. So if ultimately you can achieve the things you're hoping to achieve without venture capital and without selling a piece of your business, then great, more power to you. While I think great venture capitalists have a lot to add to help you build a business and build teams, I think that venture capital in itself is not a win. Venture capital is a tool. It's the capacity to accelerate your business with capital that you can't otherwise generate from your business.

That's right. That's right. I mean, look, if you could build an amazing business where you paid yourself hundreds of millions of dollars a year in profits, great. Then you have that forever more. There was a great founder here in the Bay Area who started a business and had paid himself a billion dollars before he bothered to sell the business. By the time he sold the business, he sold it for $350 million, and it didn't matter because he'd already made a billion dollars and taken zero dollars in venture money. Having said that, I think there are lots of businesses where the measure of 'is it a good investment' is also very simple: if I am going to give you $10 million for 20% of your business, can you increase the value of your business by more than 20%? Can you accelerate the increase in value in your business by more than 20% with that $10 million? If ultimately the $10 million increases the value by 10%, that was not a good investment. But if you use that $10 million to double the value of your business, that was a fantastic use of equity to acquire the capital you need to increase the value of the remaining ownership. So that's the first premise: raising capital is a tool, in the same way that debt is a tool. If you can build an ultra-light business that requires no capital because you don't need additional infrastructure, you're leveraging other people's infrastructure, you're using an OpenAI model, you're using whatever, and you can make more money than it costs you to deliver that solution, then that's fantastic. You should go build that. Whether or not that is a venture-fundable business should not be your primary concern. The question is, can you build the business you hope to build? If you can, you should be thrilled that that's a possibility. AI is making that dramatically more possible. It's easier to create the experience you want to create, easier to generate the sorts of marketing and sales material you hope to generate. So theoretically, AI should make all of these pieces less expensive and easier to do with fewer humans, and therefore more easily sustainable by you. So that's all great.

For sure. You're going to have to think of new categories. If you thought that content creation was your channel to marketing, as the AI content generation explosion occurred, LinkedIn said they were going to generate a million pieces of content. So okay, great. What are we supposed to do with the fact that LinkedIn is generating a million this year and everybody else is generating a million? There's a lot of noise. And then we have a lot of AI that's optimizing for other channels, but you're all optimizing around the same thing. If you can't appreciate this phenomenon, all you have to do is get on a very crowded highway where everybody is using Google Maps, because it starts pulling people off the highway into neighborhoods. The problem is, if it pulls everybody into the neighborhood, now the neighborhood is terribly crowded and you've just spread the traffic jam. So when everybody is optimizing, ultimately there is no optimal route. If a single person is optimizing, they have a whole bunch of opportunity. So I totally agree with that. The bigger question for me as an investor is sustainability. I do think that in the near term, there are a bunch of opportunities to address things for the first time that allow you to use other people's infrastructure, use the LLM of your choice, and create value. The question is, can you create long-term value? Is it different? Is it long-term differentiated? Is there a moat? Will your business increase value that is defensible over time? My suspicion is that actually that will not be the case. My suspicion is that the low-hanging fruit today is low-hanging fruit, but therefore lots of people are spotting it. There's lots of opportunity. It's going to be competitive, and it will be hard to differentiate in the long term and hard to be defensible in the long run. So that's the thing. It may be fine that people say this is a land grab and let me make money selling jeans to the gold miners while I can. Eventually the gold mine is empty and there are no more jeans to sell. Then you have to figure out: are you Levi's selling jeans to everyone, or are you the folks who are selling pans for gold panning, and the pans have no additional use once you're done with gold panning?

I'm not sure about that. Here's why I'm not sure about that. AI will allow you to train the engine on domain knowledge. So it is conceivable, it is true, that being ahead of the curve has allowed some of the early AI meets legal tech to use domain knowledge of what the law firms need to create a tool that will make it easier to write a brief or a contract or whatever. But in the long run, I think the capacity to take all of the domain and train an AI will allow the AI to encompass that domain knowledge in a way that may not allow for defensibility in the long run for the folks who got there first because they understood the space.

I guess we'll see. It's interesting to talk about this particular scenario. I taught a class at Stanford Law School this quarter, and this was what we talked about. It was about licensing, and it was AI-generated drug candidates go to a company that uses CRISPR to create drugs. Those drugs get to test efficacy using a big corporation that's good at drug testing. So the question is, which of these companies is defensible in the long run? Is it the company that is generating candidate drugs? I do think that domain will allow for a better-purposed vertical solution. That's what we're talking about here. Drug discovery is a vertical solution of LLMs. I think that's true. I think you will need some domain expertise to understand what the right inputs are and what the right outputs are. The question is, my suspicion is that's not a unique knowledge. I think that there are a small number of people who have that knowledge, and therefore the more of the knowledge you have, the better positioned you are. But my belief at the moment, and I can be disabused of this over time, is that one of two things will differentiate you. Either you will be able to create your own LLM that is somehow uniquely situated to a particular problem solving. There are some folks who are working on LLMs that are really using math, that are self-validating using math and the laws of math and the laws of physics, as opposed to just a standard LLM which uses estimation. Those will be better at some things. If you have a unique set of formulations, you could create a better LLM for a particular domain, and that will be differentiated and protectable. So that's one area. It is certainly possible that one could do that in the drug discovery space. It could be that these drugs, if you have a full understanding of drug candidates, are self-validating. If you fully understand the domain, you can create a differentiated LLM. So that's one version where the tech itself is differentiated and defensible. The other version is that you have unique information. Ultimately, these LLM models are only as good as the training information you give them. If you are somehow uniquely situated to have training data that someone else doesn't have, then you could actually have proprietary outputs. It is conceivable that a major university could keep its research to itself for purposes of training LLMs, which will then be better able to create a defensible set of drug candidates because the underlying information it has is better.

Yeah. From a very simplistic perspective, I kind of look at the universe and what's happening in the technology world. I don't have the knowledge to do drug discovery. I just don't have the depth of knowledge to be able to do this. I'm generally an extremely intelligent person, I have plenty of intellectual horsepower, but it's not sufficient. It's necessary but not sufficient to do drug discovery. That's one of my guiding factors: I cannot do this. I can do a lot of other things. But here's what I would say to you. I totally get that, and yet I often tell this story because I think it speaks to this question. I had four small children, tightly grouped, so I spent a lot of time at pediatrician offices and in the ER with fever kids. By the time I got to kid number four, when I found myself in an emergency room, I described for the doctor all the questions the doctor was going to ask me in advance, using the words he was going to use. Not because I was a doctor, but because I had had four small children worth of being asked these questions and understanding it. When I finished, the doctor in the ER said, 'Are you admitted at Stanford or are you admitted at another hospital?' I said, 'I'm sorry, I'm not a doctor.' He said, 'Really?' It was because I knew his vocabulary, not because I had some distinct piece of information. Over time, I had acquired all the necessary vocabulary to speak his language in advance of him making the medical diagnosis. So that is just to say, a lot of this stuff is not really about having some deep and unique knowledge. It is about understanding the domain. I do think that if what you chose to do was say, 'Okay, I want to understand the drug discovery pipeline, and I'm going to leave the science to the LLM,' you might be able to do that in a relatively short period of time. We have a lot of entrepreneurs probably listening to this who are like, 'Yeah, I don't know how to manufacture things, but I'm going to become a manufacturing expert and do that.' I was literally talking to a guy yesterday who's working on a set of bots for customer service, but he said he has friends who have these manufacturing facilities who have asked him if he could put his mind to creating a set of solutions for them that would take a bunch of labor out of the process of manufacturing. He said, as a non-manufacturing expert, absolutely not.

Okay, I still have to think about that. That's something that I still... I don't feel so comfortable that I can do drug discovery.

Agents, I think in terms of actually automating a huge amount of workflow, agents are perfect. The APIs are starting to be available to be able to integrate. I mean, you can do an enormous amount with agents. This week, NVIDIA's earnings call, Jensen talked about these tens of millions of agents that are going to be running around all over the world, eating up the technical horsepower, the CPU horsepower. All that seems completely reasonable.

Same problem, right? I mean, ultimately, I just don't think anything the AI thing is different from every other trend I've seen in software for the 25 years I've been investing. It may be more powerful in some ways, but it's not different. All we have to do is look back to the internet. The emergence of the internet and connectivity, everybody said, 'Oh my God, everything's going to be reinvented.' Then everything started being reinvented. There was a bunch of economic value, and then there was a crash because it was like, 'Wait a second, everything's not going to be reinvented.' In fact, they were wrong. In fact, everything was reinvented. It just was a question of time frame. Now if you look at it, the impact of connectivity has been, in my mind, the greatest impact in our lifetime of any technology, including AI. All these people who say AI is the most important technology of our lifetime, it is not, because if you disconnected all these devices, AI would be pretty meaningless. But because there is an internet, and because we have a universe of connected devices, then AI can be astonishing. When the internet occurred, everybody said, 'What are the problems that we've been trying to solve forever that now we can solve in a differentiated and better way?' The answer was basically everything. Over time, everything was solved. Everybody said, 'Oh, we should have a digital calendar, an online calendar.' Then there were a bunch of online calendar companies that attacked it. Then you say, 'Oh, the problem solved. No more calendaring.' I represented When.com. Bill Trenchard, now managing partner at First Round Capital, created Snap.com, another first calendar company. Everybody's like, 'Oh, the calendar thing's been solved.' Then Calendly came along and said, 'Wouldn't it be easier if you just had this sort of more agentic way of figuring out when to meet with people?' Boom, suddenly there's a new billion-dollar business. I would anticipate that the AI wave will solve this problem in an even more compelling way, as it has the capacity to not only understand your calendar and your behavior and your opportunities, but those of the people you're trying to meet with. It will keep track of things like, 'Oh, but you know, right now I will occasionally say to someone, 'Next time I'm in Boston, I look forward to connecting,' and I copy my assistant as if she's going to set aside a list of these people are in Boston, these people are in Washington DC. She does have that list, but most people don't have a Jenna. Even if you do have Jenna, is Jenna going to remember it? Am I going to remember it? But with AI, suddenly when I put a flight to Boston on my calendar, my calendar should be able to unearth all the people that I claimed I wanted to see. More importantly, AI can actually figure out which ones I actually wanted to see. Suddenly the calendaring problem that we thought was solved with Outlook, and then we thought was solved with When.com, and then we thought was solved with Calendly, is going to be solved with calendar AI, which suddenly organizes my life in a way we thought. So I think that every problem that we previously had and we thought we had solved with software can now be solved with agentic AI, and will be done in an interesting way that better understands the broader universe and allows for better solutions. Some of which will eliminate humans, but not all of which.

Well, I think you're right. There's a very smart young student who ended up at Stanford Business School who I knew when he was an undergrad. At the time, he was helping put himself through business school because it turned out that there were multiple e-commerce channels where he could buy something on eBay, or he could buy it on Amazon, or he could buy it on Shop.com or whatever, and then he could sell it on those same platforms. He had written a set of tools, bots, not AI bots but in early days bots, that would constantly track the selling price and the buying price of these particular things. He would list when there was a gap, he would list them on the respective target platform, sell it, and then buy it from the other one and drop ship it to the person who bought it, and keep the margin between the two platforms. He did nothing. His agent would sell it, buy it, ship it. The two platforms did all the work of taking the money, taking the goods, pushing the goods. That was fantastic. Of course, it was only a question of time until that disunity came together, but by the way, it still exists. If you go, you can find better prices on one thing than another thing. Agents are going to be able to do these things. They're going to be able to do a bunch of these things. The existence of infrastructure companies that enable the component parts is what's differentiated. Now, it's great to start a shop, but for a bunch of handmade stuff that we're selling in a shop, that was Etsy. Now you could do that, but most of that money is really someone else making a thing, you're marketing it on your shop, you're not even holding inventory, you're just selling it. So it creates real value in the manufacturing, it creates real value in the shipping, it creates real value in the marketing. There are a bunch of businesses that are solving these problems so that small business owners can pull together their solutions to make a little bit of money. The solutions are going to make a lot of money. So what are the solutions you can make? What are the infrastructure pieces that will allow others to leverage the technology advancements to build their own businesses? Those will be the better businesses, and they're still bootstrap for sure. But that distinction of 'I can build a bootstrap business that makes me some money sitting on top of other people's platforms' is dramatically less interesting than 'I can build a platform that allows others to take advantage of the technology that I've built, and I get to scale and grow with their businesses.'

Yeah, no, great. I think there is a distinction between venture-viable and a good business. I think there are lots of these tools that are going to enable people to build great businesses. I had not thought that agents were going to be one of those businesses that one could build a purpose-driven agent that does a particular thing for others. But you're right, if there are marketplaces of agents and one can create it and make it available, then it will be like a physical good that does things. It solves for scheduling my kid's soccer team. Great. Someone creates small point products, basically. That's great. I think that if one can create a solution that generates more capital than it costs to create, and therefore is a great leverage on one's human capacity and time, that's a win. The more people we can do that for, the more of a win it is. I do hope that it creates some genuine innovation that we start making the planet a better place, that we don't just sort of optimize humans out of making a living in the name of extracting a few pennies. But the good news about these shifts is that humans who are doing jobs that involve other humans will always be engaged. I have a nephew who is currently applying to nursing school. Nurses will be needed. Plumbers will be needed. Roofers will be needed. AI is not going to, in the near term, fix my broken toilet. That's a good thing. What's missing on that front is the financial infrastructure to offer good quality nursing at scale to 10 billion people. We should solve that problem, not by creating efficiency that makes the human nurse's lives worse, but by creating efficiency that makes people healthy and makes the nurses have a more rational life. So let's hope we find that.

Yeah, I mean, I think there are a couple. We have invested in a company called Stack AI. A little bit to your point about how these agents are going to allow for real innovation and solutions for folks who have a certain domain expertise but don't want to or have the AI expertise. What Stack has done has created, you know, sort of the Zapier of AI. It has a set of AI solutions that allow you to abstract out the problem from the AI. So if you are getting a bunch of...

Inbound interest in your product, but you don't know how to take AI to transform it into a sales pitch that is specific to the person who it's reaching out to, etc. That's an easy problem for AI. You can use Stack to connect up your inbound engine with your outbound engine and create a personalized message. You can use Stack to solve all sorts of interesting things that AI agents can do without having to understand the underlying AI. So I think there is going to be a profound opportunity for medium-sized businesses, not just small businesses. We have universities using it, real businesses using it, because they have product managers who want to incorporate AI but don't have the AI expertise. So that's one, Stack AI. I have a really fun one that's launching today. Go to vid dot fun. Yeah, my team has built an agentic director of AI-generated video. You can tell it what you want the video to do. Make an ad for my thing that emphasizes these features. It will use AI to say, here's a storyboard for what that video should look like, and you can tweak it. Then you say generate it, and it generates an astonishing video that is completely generative. It's using the latest AI video tools and LLMs to create the storyboard and messaging. It's astonishing and takes seven minutes. That's fun. The last one I'll share is a company called Feros AI. This is big. The team that built Einstein at Salesforce, the initial Salesforce AI platform, left to form Feros because they felt there was no good solution helping engineering teams truly understand the quality, velocity, and capacity of engineering teams. They built a platform that integrates with more than a hundred of your systems, if you have big engineering teams of 100, 1,000, or 10,000 engineers, and allows you to understand what the engineers are actually doing and which engineers are creating the most and best code. But the thing it is best used for now is that if you're using AI to generate code, which most folks are, there is no good way to understand the impact on your codebase. Is it good code? Is it working? Well, if you have integrated Feros with your engineering pipeline, you can determine: is this copilot increasing our velocity? Is it increasing quality? Where are the bottlenecks? As agents and LLMs become more important to software development, Feros will become necessary for you to appreciate what it is doing to your codebase.

100 percent. It will be able to track because it tracks not just code check-ins but quality and QA, and allows you to say, it turns out these three copilots, this one is dramatically better in that regard.

Thanks for having me.