AI Revolution in 2020s: Insights from AngelList CEO, Avlok Kohli

Welcome to World of Das, a show for data enthusiasts. I'm your host, Oren Hoffman, CEO of Safegraph and GP of Flex Capital. For more conversations, videos, and transcripts, visit safegraph.com/podcast. Hello, fellow data nerds. My guest today is Avlok Kohli. Avlok is the CEO of AngelList. He's also the founder of Faerie and Fastbyte, and he was a product lead at Square. Avlok, welcome to World of Das.

Thank you. Excited to be here.

I'm excited. I'm a huge AngelList fan. I love AngelList. You've been at AngelList for about 3 years, and a lot has changed during that time, both at AngelList and in tech and in the macro level. What, what are some of the biggest non-obvious changes you've seen?

I would say the biggest non-obvious change is, uh, the degree at which people are underestimating the impact of AI. In the last year. And from where we're sitting, we're seeing it as one of the biggest, if not the biggest tech cycles ever. It is a mega tech cycle. And the reason why we think it's being underweighted is because it has an instant install base everywhere. Everyone's already connected. Everyone has a supercomputer in their hands.. And so the rate at which these, uh, large language models are getting deployed and will be used everywhere, and the rate at which new startups are getting created to take advantage of it, uh, I think is going to make all of the past cycles look small and tiny. So that's what we're seeing from where we're sitting.

Oh, got it. So, um, you're not saying it was just un— under under-rated a year ago, you're saying it's still being under-weighted today because you're saying it's going to be bigger than mobile, bigger than the web browser, bigger than some of these other past cycles that we've seen.

Exactly. And the reason we think that, there are a few reasons for it. One is I think people have a hard time understanding exponentials, right? Like the rate at which something can grow. And people are still applying past examples of what rate of growth looks like, what a great rate of growth looks like. The other is that this cycle, there is no rate limiting, right? If you think about the rate at which the internet was deployed, it was a function of everyone getting internet connection, dial-up, broadband, everyone getting a computer, right? So there was time for all of that to get deployed. And when you look at mobile, it was the same thing. Apple dropped the iPhone in 2007 and it took many years for people to purchase it, adopt it, switch their behavior, spend more time on it, and now just spend all their time on it. And so it—

Yeah, in 2007 I was still using my BlackBerry. Like it took me a while to move over. Same here.

Yeah. Exactly. So it was naturally rate limited, but this cycle there is no rate limiting. You have 7 billion people connected with a supercomputer in their hands. These large language models are all software. And by the way, when you actually have the model, they're so tiny you can actually fit it on a phone. And so the rate at which this sort of intelligence is getting deployed everywhere, we, like, what we're seeing is people aren't, like, people are definitely taking it seriously, investing behind it, but they're not seeing the full scope of it. And the rate of change around it.

And do you think— obviously there's the, the models themselves, you know, where maybe OpenAI is the most famous company developing some of these models, and then there's like all these applications that could use the model. Do you think it works out where because it's like very specific, the applications have to develop that model specific for that use case, or do you think they just end up using some of the bigger models like OpenAI, and it could be the same application company today, just using and benefiting from some of these models?

I think for most of the business cases, I think people will build just on top of a pre-trained large language model. The other option is you could train a neural network from scratch for what you wanted to accomplish, but why would you? When you can actually lean on a large language model that's already built and you just, you, you, uh, fine-tune that for your specific use case. So what we're actually seeing is really good, uh, startups and companies getting built around just fine-tuning an already pre-trained large language model, and that actually works quite well. In fact, we, um, we had a machine learning hackathon a few months ago at AngelList. And I personally, I'm an engineer by background, and I was actually spending nights and weekends learning how to, uh, learning how to code up a neural network. And so I was actually getting right in the guts of it, building a neural network from scratch. And when we did the machine learning hackathon, we actually had two— we had two tracks. One was, let's see if we can actually, uh, train some of the neural networks with our data, because we have so much data right now in terms of venture funds, SPVs, deals, everything, right? And so we had one track around, let's see if we can train our own model. And then we had another track which was just sit on top of OpenAI and let's see what we can fine-tune and see the results. And the results of fine-tuning on OpenAI were so good, we just abandoned anything else and we're like, yep, just go all in on this. Because you now— what you have to consider as a company is, if you're going to do this, what is your moat, right? And for us, our moat is the fact that we already have all of this data, and so it actually serves our needs incredibly well. So I think for most companies, uh, sitting on top of a large language model and fine-tuning it is actually the better path. And in fact, uh, in, in fact, Andrej Karpathy, um, he was originally at OpenAI and then became the, I believe, the director of self-driving at Tesla, and now he's back at OpenAI. Um, he just had a tweet recently, uh, where he stated he thinks that the future looks more like people sitting on top of a large language model versus people training their own neural network, because why would you when you can get so many more benefits out of a large language model that's already pre-trained?

If that's the case, because I mean, if you think of AngelList, which is an amazing company, um, it was started kind of before the AI revolution, but you can now have access to these amazing AI tools Are we just going to see a small number of like AI quote unquote winners, which would be like the OpenAI and then just, you know, the same companies that are doing well today? You know, I do some sort of CRM for sales thing or whatever and a sales tool, and I could just tap into the OpenAI models to allow my salespeople to write better emails or something.

Yep. I think there are a few different places value will accrue. One will be for the companies building these large language models like OpenAI, Google, et cetera. And maybe there'll be a very small number of these that'll actually have the capital and—

And it could be even like the image generation, like the Midjourneys and some of these other types of things, right?

Exactly, exactly. So you'll have these small number of companies that will have an incentive to build these large models that others can use, others can utilize, almost like, uh, almost like AWS, right? And Azure. Uh, you can, you can think of these models that they're going to fit that, whereas high—

because the fixed costs are so high, right? We're talking about potentially hundreds of millions of dollars minimum fixed cost to start building some of these things.

Exactly. And so they'll put in the CapEx, then you'll have a, uh, host of companies that end up building on top of it. And I think the next place value will accrue is where companies have unique datasets. Because if you're building on top of a, uh, pre-trained model and you have no unique dataset, then you have to have another way in which you're gonna build moats, right? And some of the classic stuff comes into play: network effects, brand, switching costs. Like, you need to figure out what your moat is, Um, and, uh, and if you don't have any of those and you really want to depend on a data mode, then you really need to have unique data such that you can actually fine-tune the model to make it work for whatever your needs are.

Um, so it is like, let's say you have unique data. Obviously AngelList has this like incredibly unique data. How do you interact with these big LLMs, um, and not like give them your data? Do you just like Are you doing it in kind of a slightly different way so they can't totally understand your data? Or how do you do it in a way where they don't just suck in everything you have?

Yeah, there are a few different ways to do it. So at its core, what we do is we actually end up chaining workflows. And at AngelList, just for some context, we run 20,000 funds and syndicates, and we run the entire lifecycle of this, for all the way from the fund conception right down to the distributions at the very end, like it actually goes out to LPs, and everything in between around, um, receiving deal docs from companies, uh, reviewing the docs, um, and, um, uh, doing financial reporting, tax reporting, etc. There are 100 workflows that we actually have in a comp— in, in, in— at AngelList to help manage this for all of these funds. And you can think of these workflows as being a mix of human and software. But, and what's actually happening is the humans are needed for, uh, call it human judgment, right? And what has happened now is with large language models, it turns out that you can take a subset of that human judgment and you can actually put it in code now, which is awesome.

Yep.

And the way we do it is we actually take the task of a workflow, and then we chunk it, we break it into specific steps, and then we chain the steps together. And a large subset of those steps actually hit a large language model with a very specific prompt and a very specific set of data that we feed it. So for example, it could be, hey, for this document, is this a— like, what type of document is this? Oh, is this a cap table, right? Is this a share purchase agreement, etc.?

Yeah.

And then we take that and then we chain it to the next chain or the next chain or the next. So that's—

but the document is getting potentially into OpenAI, and then they could use that in the future to have some sort of— where that document is the proprietary data you have. You have all these like share purchase agreements that almost no one else has.

Exactly. And so the way in which it's set up is that, uh, OpenAI can't store this data. So OpenAI actually has these settings now where they don't store it, they don't use it for their own training. And then— got it, right? And then what you can do is—

so all you have to do is basically you, you, you basically click a switch saying you can't store this particular piece of data that I'm sending you or something.

Okay, exactly. But now the, uh, the follow-up question ends up being, well, what if you're— the window, like the number of tokens you need to send in, are very, very long, right? How do you chain it? How do you compress it? So you have to get very creative around that as well. And so a lot of what we do is around taking large data and then we chunk it ourselves and then we feed it in, summarize it, and it feeds into the next step and the next step and next step. And that's actually how we've managed to automate some of the workflows we have internally. And you can think of these as like, you know, workflow, workflow one where you thought you needed, you you could only do through human judgment. Well, guess what? You can now actually do it through code. And the rate at which we're actually identifying these that can be automated, um, even it's even shocking me, right? It's shocking me because it's, it's what you would expect really high-quality human judgment to catch and find. And, uh, it's shocking because it, it, it's, it's incredible to see how accurate, these large language models are, uh, in such a short time frame in terms of, uh, taking all of our data and fine-tuning it and then getting the necessary result back.

Interesting. And, and do you think that is going to lead to really an explosion of new companies that are coming out, just because some of these older companies, maybe they could do it, but they're just a bit more calcified and about how they've been thinking about things, or they already have all these people doing stuff, why rock the boat? Or do you think it's going to, uh, really fall to these like incumbents that are already doing well? They already have— maybe it's a tech company, but it's a 10-year-old tech company that's already has, you know, uh, um, tens of millions of dollars or more of ARR, and then is going to take advantage of that.

Yeah, I, I actually think this one will be a function— like, an incumbent, uh, taking advantage of it will be a function of how dynamic is the incumbent.

Yeah, because if it's still founder-led, right?

Exactly, exactly. Because, because if it isn't, um, it's, it's hard for someone to look at a calcified organization and then try and bring in big real change in the company, right? The way that we're doing it is, uh, we've, you know, we've actually put a whole team around it, and we're like go do this. And this is fundamentally different motion than the way we've done things before, um, and you just need to go do it. It really is, uh, I, I, I really love Sequoia's framing around a crucible moment, right? This is a crucible moment for so many companies right now that a good subset of them are just like walking off a ledge and they don't even realize it. If they're not taking advantage of this, they're just walking off a ledge because the fundamental, uh, the fundamental assumption of how large an operations team needs to be or how long you know, how many operations people you need, um, uh, in a company at a minimum changes because you can actually take what used to be, uh, all done through human judgment and now you can put it in code. And if you don't get ahead of it, then you will see a host of competitors come after you. Now, there will still be companies who have built up other moats, right, that are actually harder to actually compete against, etc. Network effect or stacking, uh, One big concept I actually learned from Square is this idea of stacking innovation, where you have one foundational layer, you know, layer on another, layer on another layer, another—

all these new features essentially, right?

Exactly. And then it takes time for someone to catch up to it. But you re— if you're a leader of any one of these companies, you really have to now take a, take a pause and, and, and ask yourself, um, given this technology that's been released, like, what are ways in which someone can come after you? Right. So if, and there's a, there's a thought exercise I always do, uh, which is if I were to leave AngelList today and go compete with AngelList, how would I do it given everything I know? And that's actually how I think about, uh, our moats is, is that's a thought exercise I go through all the time, which is, well, you know, I would do XYZ and I would do ABC. And if it's easy, well, there's a problem, right? And, and it's actually, it's been getting harder and harder actually every year. Uh, as I keep doing this thought exercise, I'm like, okay, yep, a little bit tougher now, even more tougher now. And so I think every CEO has— she really needs to be doing this, not just tech company CEOs. I think every CEO of every company out there has to do this. That's how big of a moment this is.

And if you're, um, if you're advising startups, like maybe there's many different places to compete in, um, is it okay, go take on this you know, maybe it's a take on this market where all the other competitors are private equity owned and they're going to move slowly. And, you know, how do you advise them, like, where to go, where to compete?

Yeah, it's interesting on the private equity point. I actually wonder if private equity will be the first to deploy it because what is the cost savings? Yeah, it's, it's so clear, right? It's, um, it's, hey, this will actually give you even more cost savings. You can move quite fast. So But that's more, you know, that's more of an untested hypothesis there. In terms of startups that are being created now, it's still the same advice for any startup, which is you need to understand what— first, you just need to solve a pain point that a customer is having, right? You have to hit product-market fit. There's only two modes in a company: pre-product-market fit and post, and that's it. Then you have to understand how these moats could develop over time, at least have a view on it. Because if what you're building is, uh, really easy for someone to create and compete with—

it's just like a skin on top of OpenAI, someone else is going to do it.

Exactly, exactly. And just because it's a skin on top of OpenAI doesn't mean it's not a good idea. Yeah, it's just a skin on top of OpenAI quickly afterwards, right? Exactly. And so if If it's skin on top of OpenAI, that then allows you to get very fast distribution and you build up network effects, awesome, go do it. There's some other areas of opportunity that I'm seeing specifically in and around developer tools because a lot of the way in which you need to interact with these models is around chaining and these agents. And there's a lot of different companies like LangChain, VoCode that have come out.. And so I think that's another area where building out all the necessary developer tools for this way of building software, I think will be another, you know, huge, huge opportunity. And going back to Andrej Karpathy, he had a really good blog post. It was actually in 2017 or 2018 where he basically made the case that neural networks are basically software. I think he said 3.0. Right? This idea that the next evolution of software and writing software just looks very different than the way in which we've written software before, meaning all of the architecture is the same now. Everything's a neural network, right, underneath the hood. And the, the, the people that are going to win in the space of this new way of writing code are actually the data labelers, right? The data labelers and people who can actually work in around data to fine-tune these models are actually the ones who have real value. So if you translate that into the tooling, a lot of the tooling that exists today isn't that good because it's not really purpose-built for, uh, for building applications in and around large language models. I think that's the other big, uh, big opportunity.

And just, do you think just the type of developer who's going to succeed is going to be different? It's the one who's going to be more just like trying a bunch of different things out, or maybe a bit more creative? Or how do you, how do you think of that? Like, my, as a former developer, my, my biggest frustrations with development I feel like are, are going away. Just like all the stupid misplaced semicolons and all the other types of things that just so frustrated me in the past. Um, and it seems like the, the joy of coding is going to, you know, whereas before it'd be like 70% frustration and maybe 20% joy. It seems like it's starting to flip with all these new tools.

Yeah, it's funny you say that because I, I recently got back into coding because of that exact thing. It was, you know, using Copilot and ChatGPT. It just— the frustration's gone, right? Because you can actually now work with it. It's like your pair programmer. And it'll just tell you, hey, here are the errors, here's what to watch out for. Um, I, I used to hate reading Stack Overflow, and you're like scanning, right, right, scanning for gold, and you're like, you could spend like 3 days just reading Stack Overflow for some weird corner case, right? Exactly, exactly. Um, I, I think in this, with this paradigm shift, I think everyone everyone who was a 1x engineer just becomes a 10x engineer, right? I think you actually just give everyone that superpower, and mainly because you've given everyone that Copilot, right? A 10x engineer basically had the Copilot in their head, right? They were just very good at solving these problems and, you know, and finding issues. And, um, you've now given everyone that capability, uh, with, uh, with something like a Copilot. I do think that the value that is going to be— or is going to be differentiated is going to be the creativity now. A lot of it will be around the creativity of the ideas that you pursue. And I think the second and third order effects of this will end up being you need a lot smaller of a team to have a lot larger of an impact. Right. One of the, one of the things that I, I think the tech industry lost over the last few years, um, was this idea of you need to keep growing headcount as a function of your growth when the entire concept of technology is leverage. Yeah, right. So I, it never quite, like, it never quite, um, clicked for me where you would see a company's tech company's headcount increase at the same rate as their growth rate. It's like, wait, where's the leverage?

In fact, sometimes you saw that increase faster than their growth rate.

Faster. Yeah. And I get it, you know, some— I can see the narrative, right? The narrative is we need to hire engineers, they're going to go build this thing. Well, but it discounts all of the coordination costs required. The coordination costs are— gets very taxing. Um, and of course the diffuse accountability. So I think the second and third order effects of this will be you'll just have more smaller teams building like really impactful software, right? And so I actually think as a whole, you're going to see smaller teams building, uh, really great companies, um, and the leverage per engineer just ends up being incredibly, incredibly high. Um, and so maybe the dollar—

like the, the, the ARR per employee you would expect to go up. Right now it's very— you look at even these super successful SaaS companies and it's really low. You'll see, okay, it's like a 90-person team has a $10 million ARR or something, and then you'll see that same company when it hits $100 million ARR has like 1,000 people. Like, it got worse over, over time and not better. You think real leverage, you're always getting better. You're getting the dollar per employee should be going up.

Exactly. Now, now, an interesting thought exercise here could be what's happening in those companies, right? Is it that the cost to reach the customer has gone up? And so because the cost to reach everyone— yep, exactly. So I haven't thought through, would there be leverage on sales or not, right? In this, in this world, there's definitely leverage on software, right? The software that you're building, but you still then have to get that software to the people that that can use the software. And so will this do anything to the sales and marketing expense of a company? Maybe not. So maybe you would still need to linearly grow your sales and marketing.

Well, I think it almost certainly does. I mean, there's certain things that you probably still, you still need like a certain number of AEs to talk to customers because, and then certain number of customer success people to manage those customers at enterprise software. But, you know, maybe your SDR to AE ratio changes because you can automate some of that. Maybe some of the marketing spend can be— you can use more tools. So there does seem to be just on the sales and marketing side of the house ways to reduce it. And then of course on the G&A, you should be able to reduce it. And of course, and then, and then the product engineering should be able to reduce as well.

Yep, definitely. And the other impact would just be you accelerate your roadmap. Right? You, you can just push things a lot faster. Features. Yeah, exactly. And, you know, there's this, um, there's this interesting chart that I've always just had in my mind of the rate of acceleration right now of innovation. And it really feels like we're looking up this exponential curve right now, and we're like at that bottom, or just looking up, because the rate at which there's like— there's so much happening in terms of— from, you know, across the board. In terms of, call it space innovation, intelligence innovation, which I'm kind of thinking about as the large language models and all of that. And the rate at which all of this is happening right now, it's all coming together, all compounding, and it still feels like we're at the very, very beginning of what that exponential curve looks like. And so I actually think anything we even project out today or trend out today of what a company is or what a company could look like in the future, I still think we're underestimating the rate of change and the rate of the impact on these companies because the variance at this point of outcomes is just so high. And I was actually— AngelList does an annual conference called AngelList Confidential. And as part of that, I interviewed Sam Altman in one of the fireside chats. And he had this frame of reference that just stuck with me and has me thinking about this concept of this rate of change and, uh, and how we're just looking at this exponential curve, which is imagine if all of scientific progress happened in one day versus the last 100 years, right, the last few hundred years. And if you, if you play out the curve of improvement on these large language models it actually could eventually get there, right? It actually could eventually start producing new scientific knowledge. It's crazy, right? It's absolutely wild and nuts. And so, uh, that's a lot of also what sits in my head of, wait, if the— as Sam says, if the cost of intelligence is going to zero, or it's like, you know, well, maybe it doesn't go all the way to zero because you have human creativity, right? These large language models are like pattern matching. They can't— they're not— they're not simulating human creativity today. Maybe in the future that changes. Um, I, I mean, you'd have a different model. I don't think it would be a large language model. You need something different. But it's still pretty wild, even if you get the cost of indulgence in year zero for all the things that we do, and you start producing net new, or even assisting humans to produce net new scientific knowledge very, very quickly. And so maybe it's not all in one year, but even, you know, or all in one day, but maybe in a year or two years. That's still fucking wild. Right, so amazing. Um, so yeah, it just actually has me like very excited about the future. I think, you know, I've been in tech since 2009, and, uh, this has definitely been the most interesting, interesting year ever. Um, and, uh, yeah, pretty excited about what's to come.

Now I'd love to also chat with you about these like dual threat CEOs. There are these CEOs who are also investors And, you know, in the past we had, let's say, SPVs, then we went to angel funds, then we went to rolling funds. Like, how's this dual threat CEO evolving?

The broader trend, um, that's enabling this, and let's define dual threat CEOs as CEOs who are operating a company and they're also investing, uh, from, from a fund.

Yep.

The broader trend that's enabling this is that the cost of running a fund has gone down over the last few years, and AngelList has been a key driver of this. Context is, 10 years ago, if you wanted to start a venture fund, you would have needed to bring together many service providers, a lawyer, an accountant, tax professional, they each charge several hundred thousand dollars over the lifetime, which means definitionally the minimum fund size you would need to go raise, call it, it has to be like $10 million, $20 million minimum just to go start a fund 10 years ago. And the only way to raise a fund of that size is to get an anchor LP. And the anchor LP is typically an institutional LP. Think of them as like endowments, foundations, universities or university endowments. And so the only way to do that would've been to go work at a large VC, build a track record, build the relationships, and then spin out and start your own fund. And so the entire issue there was actually just around the cost to start a fund was actually bottlenecked at the service provider level. Yep.

In many cases, the cost to start a fund was the first couple years of your fees, even at that, even at that size.

Exactly. And what AngelList did is basically in 20, um, in 2016, uh, or no, sorry, in 2015, launched SPVs, uh, which are single-purpose vehicles. It's to invest in a single company. And then over the next few years launched venture funds, and then over the next few years launched rolling funds. Now, in hindsight, I've actually reflected on this. If we knew then what we know now, would we have done it all? Probably not, because it was a very hard undertaking. But what we, you know, what we decided to do back then was we said, okay, We're going to build the best experience for the GP where all they need to do is just raise capital and invest, and we handle everything. And so what this meant was we absorbed everything, and this included all the potential service providers, and we actually approached the problem from, uh, from, uh, from ground zero. And we said, what would this actually look like in an age of software? And we had no choice. We had to go build all of the software to support this. Yep. And what ended up happening was we compressed the cost of what it, uh, you know, cost of starting a venture fund, which meant you could now raise a million-dollar fund, $2 million funds. And of course, over the years, uh, you know, we've continued to move up market and supporting $50, $100, $150, $200 million funds. And that change now made it simple for, uh, founder CEOs to also raise a fund. Because raising $1 million from friends for your first fund and then scaling that up as you continue to make great investments actually ended up being pretty easy endeavor, especially for founder CEOs who are operators and have built a, a credible company. And so that is the underpinning and, and the key change, the key why now of what happened. And so what we've actually seen over the last few years on AngelList is a lot more founder CEOs are actually running funds, and they're typically founder CEOs of companies that are post-product market fit and they're scaling. And what's— what happens is founders know other founders, right? So, and founders where their company is working, right, post-product market fit and scaling, it's a positive selection bias. Because then they're typically around, in and around friends that are just like them, right?

Yep.

And so, uh, we're seeing that these dual threat CEOs, um, who are operating their company and running a fund are actually, uh, able to successfully invest from the fund, successfully raise multiple funds. And in fact, Naval, um, uh, who, who of course is a, the founder of AngelList, started Spearhead, which is this idea to back technical founders and to give them a fund, their first fund to invest from, because of how, like, how strong the data is. Like, yes, they are the, you know, they're actually great people to give funds to because actually, uh, the, these folks typically have a great network because it's all positive selection bias.

And there, and the data shows, at least the data I've seen, is that these dual threat CEOs are consistently outperforming the traditional VCs, but the traditional LPs probably have a real hard time investing in these dual threat because they're, they're not full-time. Um, so it must be a, you know, so you have to get like maybe non-traditional LPs, so your friends, your family, um, just random strangers, the random dentist or whatever, to, to invest in those things.

Yeah, and the way to think about the, the LP world is you can split them into individual LPs and then larger LPs, uh, typically like a firm, uh, or a larger institution. And the individual LPs will be the fastest to make a change, right? Because they're managing their own money and they're the fastest to look at the data and go, yep, this makes sense. Like, I'm okay with someone not doing this full-time. Right? And then the firms just take a little bit longer, but we are starting to see this change now, uh, especially as, uh, the portfolio of these dual threat CEOs are maturing and it's turning into DPI, right? We're starting to see this, uh, change. It's just the case when it comes to venture and larger LPs, it just takes time to change, right? Because the, the, the durations are so long, right? Venture like you don't even know if you're good. I mean, you get an inkling of you being a good venture investor maybe 5 years in, right? But really the true test is like your 6th year or 7th year or 8th year or when you start getting into DPI. And so we are starting to see this shift, but yes, it is, it's such a, it's a new thing. Whereas the non-founder CEOs and non-dual threat CEOs, like the traditional VC, that's been around for several decades now. Um, but we are starting to see that shift.

And a lot of the traditional VCs, they're like former founders. Um, maybe they sold their company 10 years ago, and so they still have like some founder mentality. They have the empathy for the founder, but like everything's changed in the last 10 years. How you go to market has changed, how you hire changed, how you fundraise— everything has changed. So maybe the, the really big things they can still help the, um, their CEOs with, but like the, the medium things, all those have changed, and, and probably a current CEO could help them a lot more.

Yep, that's exactly right. I, I think a, I think a, a founder CEO needs support in many different ways, and I do think that the CEOs who are operating a company today and scaling a company today, they're their judgment on some of the items related to operating are probably a little bit more finer tuned and is more relevant today. Whereas judgment around what's typically timeless around how do you think about business models, how do you think about moats, strategy, right? That I think, you know, having someone who's been around over a longer cycle can help. And so I, my personal view is I actually think both are actually important to have some of the table. Exactly. And we're actually seeing this happen more and more now where, you know, AngelList has a, um, has a product called Rollup Vehicles where founders can actually take on a lot of checks from operators, um, as well. And we're seeing traditional VC come in and, uh, lead around and then asking the founder, hey, go start a roll-up vehicle and bring on some of the operators into the, into the company. Or even putting aside the roll-up vehicle, we're just saying, they're just saying, hey, go bring on, like, just go bring on other operators.

Yeah, bring on 10 more people. Why not?

Exactly. Bring on 10 more people because they'll actually help increase the odds. Okay. You know, what are founders really looking for when they go to raise a round? They're looking for the people around the table to increase their odds of success, increase the probability of success, because the, the, the, uh, the, the natural outcome of starting a company is failure, right? So anything you can do to increase the odds of success, um, a founder should and will do. At least great ones will definitely do that. Um, and so, you know, the dual threat CEOs are absolutely, in my view, going to increase your odds of success, because again, when we say dual threat CEOs, these are folks who are already running post-product market fit companies that are scaling. And they've been through a lot of the different loops around, like, even, you know, the one thing I wish I could impart on founders is what product market fit feels like, right? You know what it feels like when you've been punched. It really does feel like a punch in the face. Yep. And you just like, the demand is coming in from all places. And you're like, oh my god, I can't, I can't keep track of this. This is just too much. And it's the one thing I wished I viscerally knew in my first company. But unless you have a lot of people around the table that really understand it, and especially the operators, dual threat CEOs that are going through it, it's really hard to internalize. But yeah, I'll pause there.

Now, it seems logical that a CEO would make a better investor, but it's not clear to me that being an investor would help someone be a better CEO. How do you think about those things?

Mm, that's a good one.

Um, you don't want to reduce— like, for the, for the, for the dual threat CEO, it's not good for them if it— like, their main source of wealth is going to come from their current company, not from like this investing thing on the side. It might be fun for them and stuff, but ideally it also helps them make their current company better.

Yeah, I, I do think it does help them make their current company better. The key thing it does is it raises the level of their ambition when they invest behind another like-minded founder, CEO, uh, who also has high ambitions and is also crushing it, right? I, I think that is one of the biggest benefits you get out of it because you're staying connected to what great looks like. I think a lot of what founders sometimes miss is what is the definition of great, right? I think sometimes that gets lost, right?

Yeah.

People will have like, oh, I want to go raise venture funding. And then you ask them, well, you know, what's the scale of your ambition? And they say, well, $10 million in ARR. And it's like, well, you know, that's, that's, that's not that high, right? Like you need to aim higher. And so I think what investing does for dual threat CEOs in and around being in and around other great founders is it just raises the level of their ambition. And I think that's a huge positive for their own company as well, right? You just say, okay, the marker of success was here. Nope, we're going to 10x it now. Like, how do we do that? Um, I think the other benefit for the dual threat CEO is, um, they have a good sense of where the future is going, where technology is going, and they can think about how it applies to their company so they don't get disrupted, so they don't, you know, get behind the 8-ball. That's actually another really, really key feedback loop. Now the flip side of it, which is can you be a great CEO by being a great investor? I don't know if that translates as well. And I think the main reason is when you're operating as a CEO, it's just, and when you're just in operating, it's just very different motion versus you just being a pure investor. I remember I, after the acquisition of my last company, I had stepped back and I actually decided I was not going to start anything new, not going to do anything. I was just going to retire and start investing. And I, that lasted 6 months, but 6 months of me walking around texting my friends, hey, you want to go for coffee? I was like really just craving, I was craving the like adrenaline, right? And, but what was interesting was the, when I, when I stepped into AngelList, I actually remember the first day back felt like a punch in the face. I was like, oh my God, this is in, like, this is like, this is what operating looks like again, right? You're kind of going from not operating to operating where you have so much coming at you all the time that it, did feel like two different worlds. And so I just don't— I don't know if anything around pure investing would immediately translate to operating, because operating, running a company, yes, it's one part strategy, one part building product, but a lot of it's just dealing with people, right? Dealing with human problems in a company and motivation, hiring execs, right? Um, and so I think that you don't get as much of when you're investing and dealing with the human you know, nuances and problems.

Now, you've been really open and saying that more founders should be starting funds. Is there— where are we on that journey? Are we still like in inning 1 and inning 2? Are we kind of now further along on that?

I think we're actually further along on that. Um, and I, I would probably, um, edit that view slightly and say I think more founders who are post-product market fit, post-product market fit So post product market fit, $10 million or $5 to $10 million ARR type of thing. Yeah, exactly. I think the nature of the role does change when you're post product market fit. When you're pre product market fit, the only thing that matters is getting product market fit. Once you post product market fit, you have some amount of momentum that the company has and there's a different set of problems to solve. And I do think we're farther along on that journey. But I, I still think that, um, for founders that are post-product market fit, there's still this mystical thing about venture, right? And, and it's fascinating, right? Because these founders will go on and raise $5 million, $10 million, $50, $100 for their company. And then when it comes to venture and understanding how to raise a venture fund, it's like, oh no, it feels too hard, or it feels too foreign. Um, and so I think there is this like education gap around normalizing and just understanding like how you actually raise a fund, how you get started, how simple it is. And some of these founders, by the way, are investing their own personal capital. That's how they start. They'll invest $5K, $10K, $20K, right? And, um, the, the ability to amplify through a fund ends up being a natural evolution of how they're already investing. So I think AngelList can do more to help bridge that gap, and we are actually doing a lot, um, we We're gonna be launching a set of products pretty soon, and one of them, we're actually launching a product to help founders and non-founder VCs alike. I mean, we do support a very broad swath of venture at this point, and we think there are a lot of extremely, you know, extremely good non-founder VCs, of course. And what we're doing is we're putting together a product to help these folks with understanding how to do portfolio construction, portfolio modeling. Um, really just helping them understand venture a bit more, right? Demystify venture, because I think a lot of it's been, you know, a lot of the work happens behind closed doors. It's not in public. You don't, you don't raise a fund in public, right? Does— so people don't really know all of the inner workings of it.

I remember just trying to figure out my IRR as I was angel investor and angel invested over 10 years and just had all these different portfolio companies. It's just like actually figure out, like, what— I had no idea what my IRR was, and just like building this like very complicated spreadsheet. And if I just had like AngelList to do that for me, it would have been so much easier.

Yep, exactly. Yeah, you know, we spend, we spend a lot of time on building the, like, the full-blown infrastructure, the financial reporting, tax reporting, everything, handling all the complexities of a fund, larger and larger funds. And we're now With all of that foundation, we're now going, great, let's now move up that stack and let's get into some of the front office, what we call front office tools. Also, like, how do we help GPs raise more capital from LPs? Because we, at this point, you know, across the 20,000 funds and syndicates managing a portfolio of more than 13,000 companies, actually portfolio companies, we, we see everything. And we think that there's a lot we can do to help educate, guide, share best practices. Um, and, uh, you know, almost every time we do some sort of data poll— and in fact, you know, as we were looking at the, uh, I, I never actually looked at the— personally looked at the performance of, uh, you know, founder CEO running VC funds and non-founder CEO running VC funds. And I was like, oh, this is interesting data, this is awesome. Yeah. But almost every time someone asks us a question, we're like, oh, that's actually an interesting question, let's go run the data on it. And we're like, you just learned something new, right? Um, Uh, and so we're, we're very excited about starting to share some of the best practices.

Now there's this thing that people talk about where there's like too many VCs, it's too competitive, there's too much stuff out there. Um, and you, you historically kind of push back on that idea. What's your take on that balance on supply and demand in the ecosystem?

Yeah, I feel like this has been the, the statement for, for 10 years. It's just every year there's too many VCs. I remember when I first started my, my first company in 2011, um, there were 20 people you could go raise from. 20. Naval was one of them. Scott Bannister, who's an amazing angel.

You mean like angels?

Angels, yeah. But even for like the early, like, investors, like, there were just not that many, right? And you fast forward to today and it's several hundred, right?

Yeah.

And the reason I push back, at least Exactly. And the reason I push back on that view is when you think about early stage VC and you think about early stage startups, it is definitionally non-consensus, right? You're building something that the world doesn't think should exist because the world thinks that if it, you know, if it could have existed, it would have existed. And so it's definitionally non-consensus. And what you really need at the earliest stages is you need someone to take that bet with you. To believe that, you know, that this could be something big. And so my view is you want more people at the earliest stages investing that check, writing that check, helping, you know, helping put this startup into existence so they can go experiment with it. Who knows, maybe it works, maybe it doesn't. We don't know. I, you know, I won't go into naming specific companies, but I know several companies that are massive companies and that are generating like really large cash flows that went through seed extensions, for example, that no one else would have done, right? No one else would have done. I know of many other companies where they really did try to raise that initial round and it was so hard. It was just incredibly hard to go raise it because the non-consensus view was it would not work. Um, and so my view is you want more VCs at the earliest stages because you want to reduce that friction for people to go take that bet, to run that experiment. Now, taking the other side of the argument, you know, what folks would say, well, you're rate limited by the number of great founders, right? And I think Marc Andreessen, I was listening to an episode of his recently, and I think this is right. He's like, look, if there were 10 more Elons, he'll find the capital, go invest in 10 more Elons.

Yeah.

And I think the question to ask there is what's causal, right? Are the great founders just born and they just show up? Or are they forged through running their first company, second company, and then eventually getting into the third? You know, and that, that could be a big hit. And I take the view that they're forged, right? So you actually want more people to become founders. You want more people to take that bet. You want more of those VCs that can back them, and you'll get more founders out of that. In fact, YC is a founder university, right? And They'll have folks that come in for the first company, they'll go back in again for their second company, and then off they go. So anything in my view that does any— anything that can help put more founders into business for their first company, I think will actually increase the total number of great founders that will go change society, right? And, and so with that view, I actually think more VCs should exist, not less.

Now, um, on the VC side of things, you know, VCs are often giving startups advice to create a monopoly build a differentiated product, take dilution, grow fast, have one CEO, eventually go public. And yeah, none of that advice do they ever take themselves. Um, why is that? Why is there cognitive dissonance there?

I, I think the— I think the characteristics and the structure is a little bit different. Um, when it comes to VCs, uh, capital is a commodity, right? Just definitionally, it's a commodity.. And so why would a founder ultimately choose you and, and, and, and, uh, accept you into their round? And it ultimately comes down to the individual, right? What is the founder, you know, purchasing, if you will, with their equity? They're purchasing a great investor who can be their thought partner in building this company to increase their odds of success. But it all comes down to the individual, right?

So it's about the individual service business, which is not service business scalable.

Yeah, exactly. And so there, you know, there are some elements of differentiation that do build up like brand, right? And the brand comes from the individual's brand that's then amplified to the institution's brand. But it's all about the individual and it's all about the ability to attract that individual who can then amplify the overall brand. And so I think it's just a, it's a different problem to be solved. In venture around how you build differentiation when you are fundamentally in a services business, um, unless you are a platform play, um, and you're building products, right? So I think it's a difference between you're a service or your product, right?

I think to me, the founders approaching the service, there are, there are some venture firms like Sequoia which actually have like all these different types of products that they do. But to me, the, the most interesting companies in that realm are AngelList, Y Combinator, where, I mean, most people can't even name out, you know, they might be able to name Naval, but they can't really name anybody at AngelList, you know, um, and they can't name anyone at Y Combinator. Like, they actually don't know. It's because it's like a product. It's just like, you, you don't, you, you buy, um, you know, salesforce.com or something like that, and you use that product. That product is beneficial to you. They're building like a real thing that you use. And it, to me, if like that, if, if the, if one person gets hit by a bus and your entire firm is going to go under, because then you don't really have anything of super value.

Yep. Yeah. And I think that's a good way to frame it, which is what's the bus factor, right? Like, yeah, if, if, if you take your top 2 best people and they get run over by a bus and your firm collapses, yep, it's a good, it's a good litmus test for is it a product company or not. AngelList is definitionally a product company. Um, I agree Y Combinator is definitionally a product company, right? There's some— you can— I mean, it's had now, uh, you know, I think on their third CEO and still really, really strong, right? Maybe fourth. Yeah, yeah, I can't keep track. Um, yeah, I think maybe the fourth CEO. Um, and so, uh, it— yeah, those are the product, you know, product companies. But when it comes to a VC firm, um, the founder is purchasing with their equity, they're purchasing a partner, they want to work with a partner because they believe that that partner is going to increase the odds of their success. And then that partner's brand gets, you know, gets almost like transfers the brand value to the institution over time. And so that's the way I think about VC and like why it's harder for them to adopt, you know, to take that same view in building the VC fund. I think I think some have tried, by the way, you know, Andreessen famously has tried to build out all these products from— yep, recruiting, uh, to, you know, go-to-market. But ultimately, all the founder wants is a great partner, a great thought partner.

They really— that makes sense, especially pre-product market fit, to have a great partner. Even a little bit post-market and product market fit, Series B or something, to have that really good partner. At some point, we have all this like growth capital firms and stuff like that A lot of those, sometimes they don't even join your board. Like, you just want the money. Um, and yeah, it's nice to call them up because they're smart people and they can give you some advice when you go public or other types of things, but you're— you, you most care about just the dollars and the dollars coming in. So at that point, does it make sense to just upend what, what is doing at the very high end?

Yeah, I, I think as you get later and later stage becomes more of a science than an art, right? Like, early stages is more art. Pre-product market fit and around product market fit, it's very— it's art because you have to understand, you know, non-consensus bets. Exactly. As you get later stage, it is all spreadsheets. All spreadsheets, right? It's like, great, you know, sure, I believe, you know, I believe the narrative, but let's see the numbers, right? The numbers and the narrative have to match at some point. And, uh, that's, that's really what growth stage investing is about. And I think that's where It is just all numbers and could actually be very quant-driven. And I think in this market it'll be a little bit tougher for someone to go do that. But as we start spinning back up, getting back into the next bull market, I suspect we'll see a lot more activity at the later stages again.

Oh, this is fun. Couple personal questions. You had a really unique childhood. You lived in the Middle East, in India, you went to high school in Toronto. What was that experience like? And how does that, you know, and it seems like, and that's probably not uncommon, like when you meet people in tech, like they've lived all over the world, they've done all these different things. Like, how do you think those experiences help you be a better CEO?

Coming, having lived in different cultures, immersed in different cultures, not visiting but actually fully immersed in it, um, and having been on the other side of being discriminated against, um, you know, I, I the position and my view on how to think about change and political change and cultural change is just, it's a little bit more grounded in that I do think you, I think those are important in society and there's a way to do it and a successful avenue to do it. It is not a company, right? I don't think it's a company's position to make these statements or to make these claims. Mainly because starting a company is so hard, right? Achieving a mission is so hard. It's a singularly, like, taxing thing to do, and you need everyone focused on that. And so I'd say, you know, if I was to identify one thing over the last few years that kept me grounded, um, given, uh, you know, obviously a pretty, um, you know, politically charged climate, that was, that was the key thing that I pulled from having lived in so many different countries.

That's really interesting. Like, how did you— I imagine, especially in like the peak COVID 2020 when there was all this different like political things that were going on and stuff like that, um, I imagine you had a lot of CEOs coming to you for advice and asking you for things. And, you know, I remember people asking me my political opinions on just sort of the most random things that, like, you know, important issues like gun control, but like have nothing to do with my company, and probably things I haven't really thought through very well. Um, what was your advice to these folks?

It's what you just said. I haven't thought through these very well, right? If you ask me anything about AngelList and AngelList's mission and our purpose, our purpose is to accelerate the rate of innovation in the world. I will give you all of my opinions on it.

And there's political things about AngelList, like maybe, you know, opening up to new investors. Like right now, maybe investors aren't allowed to invest in private companies. And so you'll have, like they're attached to AngelList, I'm sure you have very strong opinions on.

Exactly, 100%, right? I mean, we, in 2012, Naval and the team were involved in helping get the JOBS Act passed, right? And so there are obviously things where we have a unique viewpoint and we think it's a well-reasoned, and this is the key part, a well-reasoned viewpoint of how the world should work and how we can help it get there. And in those, we can be influential and we think we can make a difference. But having an opinion on every single topic, right? It's a diffusion of focus and nothing actually gets done ever, right? It's just, it makes you feel good, but it doesn't actually lead to any real change. And that's basically the feedback I shared with others who asked me, which is, look, are you actually going to impact real change with this opinion? And candidly, do you even know anything about it, right? Because most people will see what's on Twitter, see what's in the news, and then that becomes their view of the world, their opinion of the world. But it's not actually resting on any key logical framework and any depth of thinking. It's just sort of what is the next kind of thing that's in their head that they're now, you know, taking on and sharing their opinion on. And so that was, that was my feedback to anyone that was trying to sort it out and figure it out. And I think the other key thing is, look, do you authentically believe this? Because people create brave authenticity. And the second you're not authentic, you will, you know, sure, you can sort of keep up the facade and the mirage, but at some point the mirage and facade will be gone. And at that point you've lost trust with your team. And so you need to stay authentic to who you are and just stick to it because it's all you got, right?

The second you lose it, Well, it's a good point because we definitely saw some folks where, you know, maybe they felt pressure to say things they didn't necessarily believe. It's not like they didn't like— they weren't like completely— it's not like they lied or something, but they, they, they, you know, they hadn't given it the thought that maybe was required to go public or something. And, uh, and, and then you could— and, or, and they may be privately saying something different than they're saying publicly. And I can imagine I could lose trust from people as they, as they maybe have a better learning about what these people truly believe.

Exactly, exactly. And look, you could still even believe, you know, something, something about a particular political topic and in your personal capacity be involved. That's great. But again, if you're representing a company, you have a duty to the company to represent like what is scope for the company and what's, what, what is impactful and useful for the company's mission. And that's always been my position on is like, look, I'm a representative of AngelList and of course in personal capacity I can have and I can go get involved in whichever way I want. But let's be clear, we're here to also build this company together, right? And so you want to have these separate lanes, if you will, of like how to think about things and how to approach things.

Yeah, that's really interesting. Okay, this is great. Last question we ask all of our guests: what conventional wisdom or advice do you think is generally bad advice?

Scale your company with headcount.

Okay.

Bad advice. It inherently is increasing the coordination costs and it doesn't apply the necessary constraints for creativity and to build real leverage in the business. And that's the number one thing that I've noticed, you know, folks do, and I get it. It's like, hey, go after the opportunity faster, hire all these people. But what happens? You always get slower, always get slower. And it's just more, more people problems to deal with, more coordination problems to deal with. And it's not always, of course there are exceptions where it makes sense, but for vast majority of companies, just don't do it. Like it, you know, adding more people is not gonna, doesn't naturally lead to more product innovation. And actually, in fact, so leads to less product innovation in my view.

One of the things is like people kind of want, they wanna be able to like brag about their company and, there might be certain things about their company that's private, like their revenues or something. So they can't say, hey, we're at $25 million ARR, but they could say, hey, we're at 200 people. I know.

Yeah.

And, and so, and that, that's like a way of showing, okay, well, I'm at 200 people, therefore I trump you because you're at 100.

Yeah.

But that company at like 20 people may be way better.

Exactly. It's a cultural thing, right? It's actually ingrained in the culture. Like, uh, you know, my parents, uh, I'm Indian, my background, my, you know, my parents always like, yeah, so how many people at AngelList now? Like, that is their marker of like, right, of like, oh, oh, only this? Okay, okay, okay, yeah. And I'm like, all right, fine, I gotta try harder now.

But well, it's interesting, in the, in the, in the, let's say in the 1900s, like the most important business skill was being able to hire manage and grow people. Mm-hmm. That's the number one skill you could have. There's thousands of books and courses about how to do that better. Yeah. And that's how people really often continue to grow in their career. In the world we live in today, you know, it's often about like managing vendors. Okay, I want to manage this API. I want to manage this type of thing. Um, to the best of my knowledge, there's like no books about it. There's no courses in business school about it. It's a very different type of personality that's going to— you see these like random entrepreneurs in like Bucharest or something, and they've got like 3 people, but they have like 1,000 vendors. Yeah, like doing all these different things. Like, those are the types of people that are going to be much more successful in this new world.

100% agree. Um, The, the, and you know, those, those people, like what they're actually doing is rather than getting leverage through people, they're getting leverage through software in the form of vendors because the vendors are typically software enabled or true software.

Yep.

And this is actually, this is a concept I picked up from Naval. He's talked about it publicly, which is there are really 3 forms of leverage, people, capital, and software. And people is the worst, right? And so you really, when possible and where possible, you want to lean in on software. And I remember I saw some of your tweetstorms around the vendor management, vendor optimization. Wholeheartedly agree, right? It is, it is like, it fits into the framework of software leverage, um, and you want to, uh, you want to optimize for as much software leverage as possible before you need to go get people leverage because that scales the worst.

Even big companies, if you think of these like gigantic companies that are out there, the Home Depots and all these other companies that are out there, you'll see these things where like the number of people has been roughly flat for the last 10 years, um, and the company might be doing like 3 times as many amount revenues as they were doing 10 years ago, but like the number of vendors has gone up at least 10x., during that time. So even like, like, like older school companies have embraced this quite a bit.

Yep. Yeah, I mean, you know, it's easy to fire a vendor, right? And, uh, it's a lot harder to like people manage a large team. So I didn't actually know that, Seth. That's actually fascinating, uh, that some of the legacy companies have kept headcount fixed, uh, while revenue has grown. They've done better than tech companies in that respect.

Yeah, that's a good point. That's a really good point. All right, well, this has been great. Thank you, Avlok Kohli, for joining us at World of DAS. I follow you @avlok on Twitter. I definitely encourage our listeners to engage with you there. This has been a lot of fun.

Likewise, this was awesome. Thank you.