AI Disruption: Startups vs Incumbents in the Tech Stack

Edward Suh is the founder and managing partner of Alpine Ventures, an early stage venture capital fund. He was previously at Goodwater Capital and Redpoint Ventures. He has bachelors and masters degree from Stanford specializing in AI.

Edward's favorite book: Poor Charlie's Almanack (Author: Charlie Munger)

(00:00) Introduction and Investing Framework
(06:02) Cold Emailing and Honest Feedback
(13:23) Biases and Opportunities in the VC Ecosystem
(20:47) Disruption and Ownership in the AI Tech Stack
(24:15) Robotic Process Automation and AI Agents
(26:08) Consumer AI Opportunities
(30:00) Unlocking Opportunities in EdTech
(33:11) Technological Breakthroughs in AI
(35:04) Rapid Fire Round

--------
Where to find Prateek Joshi:

Newsletter: https://prateekjoshi.substack.com 
Website: https://prateekj.com 
LinkedIn: https://www.linkedin.com/in/prateek-joshi-91047b19 
Twitter: https://twitter.com/prateekvjoshi 

Transcript

Prateek Joshi (00:01.483)
Edward, thank you so much for joining me today.

Ed Suh (Alpine) (00:04.144)
Thanks, Prateek. I'm happy to be on.

Prateek Joshi (00:07.179)
You have obviously published a lot and you invest at the earliest stages, pre -seed to seed. What gets you excited about a startup?

Ed Suh (Alpine) (00:22.928)
Yeah, that's a great question. I'd say every startup, the general framework I use is a three piece approach. And this is not too dissimilar from a lot of other VCs, but every early stage startup is, on some dimension you look at the founder, on some dimension you look at the product, and on some dimension you look at the market. And so those are kind of the three elements. And different VCs will gravitate more towards...

different parts of that. Some are much more founder oriented and care less about what they're building and really focus on just the founder and their background. Some are much more market oriented and try to forecast where the market's going. Of the three, I'd say I'm a little bit more, I see a little bit more on the product. And specifically, it's not so much about product design or a particular UX or features, but I care a lot about product market fit.

So, you know, I spend a lot of my effort trying to look for elements of product market fit that a company has, and particularly around usage. So is there a core product that's starting to be used by a core set of customers and it's used habitually, it's used with regularity, and people really love it? And so that's kind of the main thing I look for. It's not so much financial per se, but more usage -based. And so that's the main thing that I specifically try to look for.

Prateek Joshi (01:48.715)
taking that framework and maybe applying it to the AI wave we are in the middle of right now. So if you look at the AI tech stack, what parts are likely to be disrupted by startups versus what parts are most likely to be owned by incumbents?

Ed Suh (Alpine) (02:08.688)
Yeah, again, great question. If you look just generally at how technology industries evolve over time, and this is not specific to AI, but more broad, incumbents tend to have two very distinct advantages. One is in distribution. So, incumbent companies have really strong brands and a really large roster of existing customers and users that they can distribute new products to.

And then the second is around economies of scale and resources. So, you know, incumbents tend to have a lot of excess cash and, you know, really strong balance sheets, the ability to raise capital at a relatively low cost of capital and invest that into capex and infrastructure, and then amortize that over a large, in a large base. So, incumbents generally tend to have a lot of advantages in...

in areas that are very capital intensive and require a lot of upfront investment. So within AI, the foundational model layer is the most obvious area where incumbents have a natural advantage, right? Because it takes hundreds of millions, if not billions of dollars over time to successfully build, train models, have the infrastructure and compute infrastructure to do that. And then to continually refresh those models.

And then incumbents also have distribution advantages where they can gather data from a very, very large install base very quickly and iterate on those models really well. And so it's interesting that OpenAI, for example, has done a really good job of quickly, despite the fact that they were an outsider and a new entrant, be able to scale up and essentially have the resources of an incumbent, right? Both from a capital perspective, they have raised billions of dollars.

and have the balance sheet of many of the incumbents. But then also distribution -wise, they've done a really good job of promoting the usage of all the products, including ChatGPT, and built up a really large install base. So it's been very interesting to see that. But I think that's a layer where companies, and I consider OpenAI to be pretty much an incumbent today, just given how large they are and how much they've raised in the evaluation.

Ed Suh (Alpine) (04:29.552)
But that's an area where I think incumbents will continue to have a big advantage. But I think in the rest of the stack, there are mostly places where I think new entrants have a big advantage because there are a lot of vertical applications, for example, where it doesn't make sense for an incumbent to build a standalone product. It's not really a core area of focus. It might not be a big enough needle mover for them.

So if you look at, you know, verticals like AI, you know, kind of applications for the healthcare industry or financial services or education, you know, those are ones where I don't see a lot of, you know, incumbents, you know, doing a really good job of evolving into those areas. And those are areas where new entrants can be, you know, do a really good job. There are some where incumbents within those verticals, you know, may have an advantage and be able to build, you know,

a new vertical application that can tackle that. So in education, there are existing players like Chegg and Quizlet and call it Generation 1 EdTech companies that are now building LLM -based applications to evolve their products. But I think for the most part, there's still a lot of areas where new entrants with a fresh perspective, fresh product, they don't have to...

kind of maintain a legacy infrastructure or legacy product, we'll have a big advantage in terms of being able to start fresh and evolve.

Prateek Joshi (06:02.411)
I want to take a quick detour. You have a very no -nonsense way of tweeting, which is just fantastic. And I want to explore a couple of topics you've talked about. Let's start with cold emailing. Recently, you had a tweet and a bunch of discussion. Why is that an important skill for founders and even VCs to develop?

Ed Suh (Alpine) (06:27.984)
Yeah, I think cold emailing is a really good skill for everyone. Just philosophically, I think being able to sell and persuade others, particularly strangers, is really valuable in anything that you do. A lot of life and any kind of career is about selling. And cold emailing is part of that. So I think for both founders and then also for VCs too, kind of in an inverted way, being able to...

get in front of people who you want to get in front of, but you don't have a connection to is a pretty important skill to have. And I think the nature of it is a little bit, it's a little bit nuanced, a little bit different for each constituent, but I say generally, there are a few things that I've seen work well and resonate. One is just to do your homework on the person that you're reaching out to.

you know, kind of learn everything you can about them, about, you know, them personally, about their, their, uh, their background, um, you know, what drives them, what are their incentives and try to craft your note in a way that resonates with them and would be, you know, they would find interesting. So for a founder reaching out to a VC, you know, that would mean, you know, thinking from the perspective of VC, what would pique their interest in what you're doing? Maybe it's, you know, something, um, some kind of metric about you that's really superlative.

Maybe it's something that you've built that aligns with their investment area of interest. Maybe it's something more personal, but there are things that you can do there. And then related for VCs reaching out to founders, which I do a lot. I do a lot of cold, cold emailing to founders that I don't know particularly well. And I'll make a point to try to use their products and provide feedback and just come across as really genuinely interested in what they're doing. And then try to develop personal rapport too.

If I have a sense of what their hobbies are or interests, or if there are people that we know in common or we have common interests, I'll try to highlight those. I think I've seen a lot of people take an approach where they try to impress the other person. For a VC, cold emailing a founder, they might brag about their portfolio companies and how big their fund is and things like that.

Ed Suh (Alpine) (08:47.376)
But I found those generally don't work too well. It's much more about expressing genuine interest in what that person has done and coming across as a much more interested person has worked really well for me.

Prateek Joshi (09:01.515)
Right. And also when you work with founders, extending this theme a little bit, what are the hard truths that you end up surfacing? Usually things that are not cool to say are not maybe not politically correct, but it is the hard truth. So when you advise founders, what else comes up?

Ed Suh (Alpine) (09:24.752)
Yeah. Are you talking about founders I've invested in and I'm providing feedback to them or?

Prateek Joshi (09:30.443)
Actually, yeah. Founders have invested in and also just in general, maybe you'll react to invest in, but like maybe they're doing it for the first time and what other hard truths do they need to know?

Ed Suh (Alpine) (09:43.856)
Yeah, yeah, that's a good question. Um, you know, I think it depends a little bit on the founder and, and where their blind spots are. And, you know, I think everyone has blind spots, right? I have blind spots. I'm sure you do, you know, everyone, uh, no matter how good you are, you have certain blind spots. And, um, you know, so part of what I try to do with founders is, um, it's, it's about striking a delicate balance because, you know, as VCs, it's not our job to.

manage founders or manage companies. My perspective is it's up to the founder to really own their particular company and their team and the roadmap. And you're trusting that they have the capability to do it. So I never wanna be in a position where I'm dictating what they need to do. But I do sometimes try to give genuine feedback that can be hard to hear. If I...

See a blind spot or you know or have a different perspective than then they might So, you know sometimes it depends a little bit about on the nature of what that is Sometimes it's about you know, I've seen sometimes founders can get a little bit complacent if you know things are Good, but not great. So, you know if they're growing at a certain rate, but it's not a really high rate of growth They have a lot of runway, you know, there's a tendency to

to not be as driven or not push as hard. And sometimes I'll see that where a founder will have a little bit of comfort in thinking that they have two years of runway and they're growing at a certain modest base rate and they think kind of everything's fine, everything's good. But the reality of a lot of startups is you actually have to hyperscale and push really hard. You have to grow really, really fast. There's a lot of competition coming at you.

And so, you know, sometimes I've given tough feedback to founders where, you know, I feel like they need a greater sense of urgency in driving faster and driving harder. So, you know, that happens sometimes. Sometimes it's more tactical. Like, you know, with first time founders, I've sometimes seen, you know, when they're hiring for executives in particular, you know, they might not have a very

Ed Suh (Alpine) (12:10.)
accurate sense of what excellent is. And they might meet a candidate that checks all the boxes, that sounds really great. But there are, just given I've met so many different executives of different flavors and different styles, I might pick up on things that are more of a red flag or think that there are better candidates. But just because they've met a smaller number, they might think that that candidate is really top notch. And so sometimes I'll kind of...

not necessarily override, but give them really blunt feedback on candidates that they might be excited about and saying, hey, they're actually not as good as you think, right? There's a world of other people out there that are potentially better. So, there's all kinds of different things, but I do try to strike a balance with founders where I try to be really direct and honest. As a VC, as an advisor, we're also doing a disservice to founders if we're not giving accurate feedback.

and we're sort of trying to hide the truth. But also give them the leeway to make their own decision and make their call and give them space to do that. So it's a delicate balance trying to get feedback to founders for sure.

Prateek Joshi (13:23.627)
100%. Moving on to the VC investment process. So let's start with in the early stages, there isn't that much financial data or market data to look at. It's a different, it's not like analyzing a NASDAQ company. So in the early stages in a VC investment process, there are many biases that exist. Some are maybe fair, most maybe not fair. So what

biases bother you the most if you look at the VC ecosystem.

Ed Suh (Alpine) (13:59.888)
Yeah. I think a lot of the biases in VC come from a very specific characteristic about it, which is that it's insanely network driven. So, you know, I think the vast majority of VCs, you know, will say something to the effect of, um, we'll only invest in founders who come in through a warm referral, right? And, you know, we, we want every founder to come in.

through someone we know really well, whether that's another founder or another VC that we know. And so, you know, just that characteristic alone drives a lot of biases, right? Because, you know, as humans, we tend to know and be more friendly with people who are similar to us, right? Similar backgrounds, probably live in the same city, similar age, demographics, you know, upbringing, maybe went to the same schools, worked in the same fields.

And so that just perpetuates a lot of intrinsic bias, I think, just given the nature of that and what that means in terms of investment funnel. But then there's also, I think, certain stereotypes that VCs have about founders and whether they're accurate or not. I think they're maybe not as highly accurate as people think. So certain characteristics about founders like...

VCs tend to bias towards founders who are younger, founders who are more technical, founders who went to an elite school versus not, who worked at certain companies. So I think there's a lot of different reasons why that exists. Ultimately, part of the reason for this is a lot of VC investing is about you have to narrow a funnel very, very quickly. You have a lot of potential investments you can make, a lot of potential companies.

And you have to narrow it down very quickly to a very small number that you can evaluate and dive deep on. So VCs naturally have heuristics to do that. And, you know, I think all those things tend to perpetuate certain biases about founders that they're more interested in versus.

Prateek Joshi (16:08.491)
When you look at the process of evaluation in the early stages, like pre -seed and seed, obviously there isn't that much company data available. So what data can be useful in the evaluation process of an early stage startup?

Ed Suh (Alpine) (16:26.8)
Yeah, and this depends on stage and it depends on, you know, I'd say VC. So, you know, I wouldn't say there's a, there's one, you know, perfect set of metrics or one perfect, you know, process to evaluate a company. I think the beauty of VC is that it's as much art as it is science, right? And, you know, there's different frameworks that you can use to evaluate companies. And so this is not, you know, certainly a ground truth.

I tend to focus a little bit more on, like I mentioned earlier in the conversation, on product metrics and on usage in particular. So, you know, I'm less, for example, financially driven. You know, obviously if a company has strong revenue growth and has great margins and, you know, you know, economics, that's better than, than not. But I'm not wedded to that in the sense of, you know, I'm comfortable investing companies that are pre -revenue that are really early in monetization.

But I try to look for companies that have strong signs of product market fit. And that to me is a combination typically of user traction engagement and retention. So, are there examples of a core user base that's using a product very habitually with a high degree of intensity, high retention?

is that core usage really strong? And so that's the main thing I try to look for is elements of that. And so it's not so much about a specific number, right? Like I'm not looking for, okay, it has to be, you know, this many times per week is my cutoff or this percentage retention. But it's, you know, kind of in the context of what the product is, but generally looking at, you know, are there signs that this is something that's really strong in terms of usage and, you know, product?

So that's the main thing I personally look for.

Prateek Joshi (18:29.547)
Where can AI be effectively infused into the VC tech stack or the VC process?

Ed Suh (Alpine) (18:37.616)
Yeah, so AI is a pretty broad term. So, you know, within it, there are a lot of different sub products and, you know, and subcategories. So different, I think parts can augment different parts of the investment process. So, you know, recently there's been a lot of attention, obviously on LLMs and on what, you know, LLM based foundational models can do. And that's generally more in the vein of

evaluating large corpuses of text and generating and synthesizing large corpuses of text. So I think there are areas where that can help with the investment process. If you think about a lot of areas where you need to ingest large quantities of text and media and then summarize that. So that could be things like evaluating companies' pitch decks and company data room and materials that are in written form.

you know, what used to be a manual process can be largely automated and synthesized through LLMs. I also apply a pretty different type of approach in my investment process, which is to use quantitative signals to source companies. So this is a little bit different than LLMs, but, you know, there are, you know, ways that you can apply more traditional machine learning techniques to

both uncover and qualify companies that have some kind of superlative metrics about them. So whether they're growth metrics or usage metrics and say, okay, quantitatively, this company is outperforming a bunch of other companies that are at the same stage, pre -seed stage, series A, et cetera. So there are a bunch of things that I do around that that are more quantitative, more time series based, more for sourcing and evaluation. So...

There's a lot of different ways that technology and AI can leverage in the investment process. And depending on the specific type of AI it is, there's different parts where it'll be more applicable.

Prateek Joshi (20:47.211)
I want to touch upon a couple of themes that you've either invested in or you've talked about. So let's start with robotic process automation in the AI era. Obviously, UiPath is a shining example of RPA, big success when public, it's fantastic. But in the AI native era, what does a great company do? Like, what does it look like at the seed stage?

Ed Suh (Alpine) (21:15.888)
Yeah, I think generally speaking, you know, great AI companies do a few things right. One is to build some kind of product differentiation that is very unique to that particular company and their data set and their processes. So no, not just an API call to an LM, but you know, doing a lot more on top.

That's not just UX. It could be leveraging different data sets that the underlying LLMs themselves are not doing. It could be something around the UX that is very different from a chat -based interface. So I think there's a bunch of things there that are pretty compelling. But I also think there's a set of things that are not product -oriented that great AI companies do. And this is generally true of

you know, of all technology companies, which is there's a lot more to a successful startup than just the product, right? And just the feature set. There's a lot of execution that's required on messaging that's required on marketing, sales, customer success, pricing, et cetera. Um, things that are a little bit separate from the product, but that can differentiate and, and really help drive success. And I think that's going to become more and more important.

in a world where a lot of software, and by extension, a lot of AI products, will, I think, increasingly become more commoditized at the feature level in the sense of it's going to become easier and easier for products to be replicated. The base set of features, the base functionality that a lot of these apps do, they're not that hard to replicate and rebuild. And you can do that with greater efficiency.

But there's a lot of execution that goes into building a brand around marketing, around positioning, around pricing, around all the execution behind the scenes. But I think that's actually gonna be what sets a lot of companies apart. And it's already setting a lot of companies apart in certain verticals that are more commoditized. So it's interesting, because that's an area where traditionally, VCs have not focused as much in terms of trying to differentiate products. We've been...

Ed Suh (Alpine) (23:41.104)
historically much more technically oriented, much more product oriented, where we think about what features someone has that someone else doesn't have, or what kind of technical advantage some company has. But I think it's gonna become much more of an execution advantage and what that means in terms of driving differentiation. And that means just, we have to evaluate companies a little more holistically than we used to, where a product is one.

part of a lot of other constituents that are really gonna help private company success.

Prateek Joshi (24:15.403)
Right. And do you think that AI agents will replace all the RPA products or will they exist side by side or is that a completely different category? So what do you think of the AI agent wave we are seeing?

Ed Suh (Alpine) (24:37.488)
Yeah, I think we're still really early in the AI agent wave. I think a lot of the innovation to date has been more around automating one particular process or one particular workflow, and specifically generally around either synthesizing large corpuses of information or generating that new information from an existing training set.

So on the RP side, I think we're not quite at the point where more complex, multifaceted of interactions and multifaceted workflows are automated to 100 % accuracy and trusted yet. I think we can get there. And this is, I think, generally a broad trend around AI agents that I'm really excited about.

You know, on the consumer side, everyone has this notion of, you know, an AI travel agent or an AI tutor that's, you know, more general purpose. They haven't really been built yet. You know, there's a little bit too much complexity in present day to build a really successful one, but, you know, theoretically it's totally possible. And I think that'll happen more and more. And so on the RPA side, I think we'll see a lot more. I don't know if we'll ever get to the point where, you know, it's fully, fully automated and there are no humans in the loop.

That's what you're asking. I'm not sure yet, but I think there's way more than what we can do today.

Prateek Joshi (26:08.523)
You mentioned consumer and just looking back, there was a time when consumer was everything. Everyone loved it. We invested heavily in it. And obviously, over the last decade or so, enterprise has become more prominent. But if you look at consumer AI opportunities, what are you seeing in the market? Like what's exciting to you and what do you think can become big?

Ed Suh (Alpine) (26:38.064)
Yeah, yeah. And I think it's going to be an evolution. I think today the most obvious opportunities are ones where there's a natural chat or conversational interface that used to require a trained human operator, but can be automated through multimodal LLMs. So, you know, I'll point to a few really obvious and specific examples.

One is in language learning. So there are now a series of companies that are building language learning bots, essentially, that are vocal. So they're conversational and voice -based. So, you know, traditionally the issue with language learning was it was really, really hard to practice conversation, right? Which is actually the way that you need to get fluent in a language and conversational is you have to practice speaking it to a native speaker.

So, you know, back in the day, there used to be a series of companies that were trying to build marketplaces with humans, where they would try to hire, you know, a bunch of native speakers of, let's say, Mandarin Chinese. They tried to hire, you know, a bunch of people in China or Hong Kong to go into this marketplace and chat live with people in the West. And, you know, they could practice Chinese or practice English, et cetera. It was a really hard problem to solve, right? From an economics perspective, you know, it's costly to do that.

And you have to synchronize everything and manage a large population of supply and demand. But now there are a series of tools that are leveraging AI so that on the supply side, they don't have to have any humans. It's just a bot that is trained on speech and is perfectly fluent in whatever language that you want it to be trained on.

and can use a combination of speech recognition and synthesis to have a real -time conversation with a person that sounds like a native human with the right inflection, emotion, et cetera. So that's one specific example. There are things around AI tutoring and education that are similar to that. So those are the more obvious examples today, where there's a conversational interface that used to be manual that's now automatable.

Ed Suh (Alpine) (28:59.248)
But I think over time, you know, I'm personally excited about a world where there's going to be much more agency that an AI application has to take actions on behalf of a user, oftentimes in the background. So, you know, the whole concept of agents and agents for different tasks, it hasn't been developed as much today for a variety of different reasons that are more technical. But I think over time there will be much more

You know, one examples of agents that have more potential to take a wider breadth of actions. And we as people have a high degree of trust in those. And so that's a world where I'm personally really excited about, you know, having an agent that could be, you know, an assistant for travel, for, you know, in the workplace, note taking and scheduling, you know, all kinds of different tasks where, you know, they're not just synthesizing information for you, but they can take actions on your behalf.

And that's a pretty exciting field.

Prateek Joshi (30:00.363)
You mentioned language learning. So let's extend that to all learning and actually all of education. Historically EdTech has been a tough sector because if you go to sell to schools, the budgets are tight, it's regulated. If you try to sell to students, they don't have money, the churn is high. So is AI unlocking any...

net new opportunities in EdTech and what do those opportunities look like if they exist.

Ed Suh (Alpine) (30:31.28)
Yeah, totally. So I think there's a lot of net new wallet that's being unlocked. One is on the teacher side or the district side. So post COVID, there is still a very strong and continued push towards digitizing classrooms, both K to 12 and higher education. And so nowadays pretty much every student in a public or private elementary school and middle school and high school in the US,

has access to some kind of technology. They get a Chromebook, they get a laptop, they might get an iPad, and that's just given to them for the year. And there are more and more examples of apps and services that are being used and incorporated in the classroom. So I think that'll continue and AI will be a core part of that. But then also for at home learning and enrichment, there's a lot of propensity for parents to pay for educational tools to help their kids. And that's a pretty...

high propensity to pay. I would say for the most part, parents are very vested in their children's education. If your kid is behind and needs help and improve their grades, you're willing to pay for that. Or if they're already doing well, but they want to get more advanced, there's also pretty high parent willingness to pay and encourage and support that.

So I think there's a lot there that, you know, at tech companies traditionally have done a good job of unlocking parent demand and AI, you know, will be an extension of that. And so, you know, I think AI will help to unlock a lot of opportunities. I think, you know, the biggest opportunity honestly is unlocking services that used to be cost prohibitive and access constraints. So tutoring is a great example of this.

And AI enables that to be offered at free or no cost to students around the world. So this is a much more global opportunity than just in the US, where I think the US people are blessed with generally having high quality education and access to good resources. And of course, the wealthy nation too. But around the world, there are a lot of emerging markets where the school systems are really poor or there's very limited access to even basic education for students.

Ed Suh (Alpine) (32:49.936)
But if they can have a free service through AI that's available on their mobile device and it's 24 -7 and it's very low cost or free and high quality, I think that can do a lot to improve educational standards and provide access to a lot of knowledge and services for kids around the world.

Prateek Joshi (33:10.987)
I have one last question before we go to the rapid fire round. What technological breakthroughs in AI are you most excited about right now, specifically from the lens of, hey, that's going to unlock net new seed investments for us?

Ed Suh (Alpine) (33:15.248)
Mm -hmm.

Ed Suh (Alpine) (33:29.936)
Yeah. Yeah. I think there's a little bit of rehash from earlier in our conversation, but I'm still, you know, continue to be really excited about the potential for agents in particular. You know, um, like I mentioned before, there had been, there's been so much attention brought on automated services and, um, you know, and, and automated call it information synthesis and generation, right?

Um, you know, most of the AI tools today are very specific in terms of, you know, taking a large corpus of data and then analyzing it and summarizing it or generating something similar, right? Based on a prompt. Well, I think there's a whole new world that's still not yet really unlocked around, um, AI applications taking action on behalf of a user. And that's still in its infancy, but that has, I think the strongest potential.

Right. Think about just all the tasks that can potentially be automated. You know, just think about all the times that you're clicking around different applications on your phone or on your laptop. That should all be automatable, right? Or the vast majority of it should be automatable because they're pretty, you know, rote repetitive actions that a decently trained agent should be able to do on your behalf. And so I think there's a world of productivity unlock that can come from that. That's really exciting. It's a, it's a pretty broad.

you know, thing, but I'm pretty excited about that. I'd say that's the main thing that I'm still keeping an eye out and really excited about in terms of development.

Prateek Joshi (35:03.787)
With that, we're at the rapid fire round. I'll ask a series of questions and would love to hear your answers in 15 seconds or less. You ready? All right. Question number one, what's your favorite book?

Ed Suh (Alpine) (35:13.84)
Sure, yeah, go for it.

Ed Suh (Alpine) (35:18.704)
If your book is called Poor Charlie's Almanac, it's a book that is a collection of writings and speeches given by Charlie Munger. So Charlie Munger, who was at Brookes or Hathaway running it alongside Warren Buffett, late Charlie Munger, he died recently. And I just love his outlook on life and investing, you know, for someone who was so successful, had really high stature. He had a very down to earth personality, very witty, dry sense of humor.

and very simple view on life, which I really appreciate. So I just, you know, I always come back to it from time to time. And, you know, it reminds me to simplify a lot of things in life, which I really like.

Prateek Joshi (35:55.115)
I'm a big fan of Charlie Munger, especially there are a bunch of YouTube videos with his best moments, like his quips and like one -liners. I really love, I really enjoy watching that. Also, the bookshelf behind you, I love that. I love the collection, the look here. So fantastic. All right, next question. What has been an important but overlooked AI trend in the last 12 months?

Ed Suh (Alpine) (36:04.176)
Yeah.

Ed Suh (Alpine) (36:20.88)
Yeah, I think one that's been a little bit understated, but I think is really important is how much AI is being leveraged or can be leveraged internally within companies to make them much more efficient from an operational perspective. Right. So all the attention I think has been around AI, new AI products and services that are being launched externally, right. And being sold to other companies, but there's been a ton that's also happened behind the scenes around AI.

enabling organizations to become much more efficient. So, you know, one of the trends that's really both potentially scary, but also exciting is the concept of AI developers. So, you know, there are a couple of companies that are trying to build this now, but there's a world where in theory you could imagine most, if not all the software development that's being done for most companies can be automated and done through AI instead of a human developer.

I don't know if that's actually gonna happen. There's a lot of nuance behind software development that I think is hard to automate. But I think there's a lot more that can be done than is being done now, which is around cogeneration of very simple blocks. And that's a really interesting area because you can imagine a world where you no longer have to hire 30 engineers when you're Series A to build the net new feature of your product, right, or Series B.

you could potentially get all that done with maybe one or two or zero engineers. And so, you know, it's a very interesting hypothesis around, you know, can we get to a world where companies are much more efficient from a headcount perspective, you know, just given how much AI can be leveraged to automate what used to be manual processes. And that's something that hasn't, you know, it's been stated a little bit, but I think it's been more understated versus a lot of the external developments that have happened in the industry.

Prateek Joshi (38:16.139)
What's the one thing about early stage investing that most people don't get?

Ed Suh (Alpine) (38:22.544)
Yeah, so this is something I think a lot of people do get, but I'd say isn't stated enough, which is how critical sourcing is. So, you know, investing is some combination of sourcing, meaning finding new investments, picking, winning, and helping to build and support investments, so those four things. And there's a lot of emphasis put around picking and winning, but I think there's a lot less that's traditionally been emphasizing around sourcing.

But sourcing is what drives everything else. Meaning even if you're a great picker, you win and charm founders really well and founders love working with you. If you're not seeing the best companies and the best deals, everything else is kind of irrelevant because you're not going to build a great portfolio. So, you know, and this has been true for me where I put a lot of my emphasis in my firm around making sure that I'm sourcing great opportunities, seeing great things at the top of the funnel. And I think a lot of, you know, VCs.

you should be reminded that that's really critical and the foundation on which everything else is built.

Prateek Joshi (39:27.499)
What separates great AI products from the merely good ones?

Ed Suh (Alpine) (39:33.584)
Yeah, I think it's actually less about AI. So I have a little bit of a differentiated or maybe more controversial opinion on this. I actually think that incremental feature development is less relevant in AI just because, one, features can be copied so quickly. And then two, there comes a point at which incremental improvements in features or speed.

or cost are fairly incremental to the average customer and a lot of software products become commoditized. So I think actually what matters more and more is an execution on all these other dimensions that are not product related, meaning marketing, sales, customer success, all these other things that are more business oriented. And I think that's actually what separates great from good, where it's not just about building

the best feature set or slightly better product than everyone else, but it's more around execution, around distribution, go to market, retention, all those things. And so that's kind of what makes a great company versus just a great product.

Prateek Joshi (40:45.803)
What have you changed your mind on recently?

Ed Suh (Alpine) (40:50.576)
I used to be much more adverse to low gross margin AI companies. As you know, a lot of AI companies today, at the application level in particular, are pretty low gross margin because they have very high computing and inference costs, a lot of which is passed on to OpenAI. But my view on that has changed a little bit. I recently come across a few where the...

the gross margins have improved significantly in just a period of a few months, and there's a lot more nuance to it. So, you know, what happens actually a lot of times is there are AI applications that are freemium. So, you know, they have a massive base of free users that are being subsidized by a very small percentage of paying users. And the actual gross margin on those paying users is very, very high, but they just have a big long tail of free users that are weighing, dragging that down. So, you know, and...

As time goes on and they improve their pay conversion, that can dramatically improve their overall blended gross margin. So this is one of those metrics that I think a lot of people have a very simplistic view on where they say, oh, I will never invest in a product that has negative gross margins or that has really low gross margins. When in reality, there's a lot more nuance to it. And so I've become much more accepting of companies that on the surface might look like they have a certain unit economic profile that's weak.

but I do a lot more work to uncover why that is and whether there's a positive trend behind the scenes.

Prateek Joshi (42:24.971)
That's actually a good one. I think I've heard that reason a number of times where gross margin is like used as a reason to not do some more homework. But I think there's a lot more to it. And I think that's a great point. All right. Next question. What's your wildest AI prediction for the next 12 months?

Ed Suh (Alpine) (42:45.008)
Yeah, my wildest one, and this might take more than 12 months to come true, or maybe I'm completely wrong here, but I have a perspective that I think there's gonna be a lot of consolidation at the foundational model layer. We're starting to see some of that, right? There are one or two players that seem to be struggling, and there are a few that are starting to get acquisition interests and get acquired.

I think there'll be a lot more consolidation because I think we're in a world where there are a fairly reasonably large number of foundational models, all of which need to raise a lot of funding and have really large user bases and customer bases to sustain. But I don't know if the world can handle that many. And what I think will actually happen is there'll be a massive consolidation into two or three or some very, very small number. I think OpenAI will certainly be one of them.

and they're arguably the leader of the pack. But outside of that, I don't think there's room for that anymore. And so over the next 12, 24 months, as these companies continue to need massive amounts of funding to build and refresh their infrastructure and their models, I think we'll see more consolidation in the industry.

Prateek Joshi (44:02.123)
All right, final question. What's your number one advice to founders who are starting out today?

Ed Suh (Alpine) (44:09.584)
Yeah. So first of all, I'll say, you know, as a VC, it's not necessarily my place to try to dictate to founders what to do, um, you know, or impart X, you know, general purpose advice. Cause I think every founder is very different and very unique and have different styles. Um, that being said, I think the founder journey is one where it's important, but yet very difficult to balance to diametrically opposing viewpoints.

I think founders have to both be impatient and patient at the same time. And what I mean by that is on one hand, the founder journey is really long, right? The typical startup, you know, can take anywhere from eight to 10 or more years to exit, right? It's a very, very long time. Um, and, you know, all along the way, you know, founders have to drive really hard and continue to push their team and themselves.

It's important to be patient and understand that there are going to be a lot of bumps in the road, but it's a long journey. But at the same time, it's important not to be complacent and take things slow because you have to move quickly and execute quickly because competition is there. It's important to continually make sure the company is funded, to keep recruiting, keep hiring, keep elevating the bar. And so it kind of feels like a marathon and a sprint at the same time, which is really difficult, right?

Prateek Joshi (45:35.787)
Yeah.

Ed Suh (Alpine) (45:37.04)
But I think great founders find a way somehow to have stamina throughout the whole thing, take rest when they can, take care of themselves, but balance the sense of, okay, I'm going to be really patient and even keeled in the long run, but every day I want to continue to push myself in the team and go fast. So I think that's a really important skill to have.

Prateek Joshi (45:58.987)
That's actually a really good point. And then a good way to end the episode is to learn to be patient and impatient at the same time. And it's that push and pull that you've got to balance to keep moving forward. Edward, this has been a fantastic episode. Loved your views on all these topics. So thank you so much for coming onto the show and sharing your insights.

Ed Suh (Alpine) (46:23.152)
Yeah, thank you, Prateek. This was a lot of fun. Really enjoyed it. And thanks for having me on.