Andrew Chen (a16z speedrun) on the Future of AI Startups & Why Coding May Be as Easy as Video

I think we're at a point where, yeah, maybe we're going to go blow all of this up.

Today I'm speaking with Andrew Chen, general partner at a16z, about how AI and agentic engineering, formerly known as vibe coding, is changing everything.

I love the idea that maybe like the future will be individual, 1 or 2 people, the same dynamics that shape, you know, Shopify or that have shaped Roblox or, you know, YouTube content creation or Substack. Is coming for the software world. And maybe code becomes like almost like another kind of content. This form factor of, you know, what we think of as software, I almost think about it like, yeah, what if it's just as easy as it is to record a video?

We dive deep to discuss how everything from product to go-to-market, retention, and even company building is evolving rapidly.

The entire aspect of why growth teams ended up being created in the industry was because You needed a combined unit of like marketers and, you know, PMs and engineers to actually ship things that would move, move growth. What happens when every marketer actually can output code themselves? Maybe every job will include software creation as part of the job, and then every software experience you have has software creation as part of it.

He's giving us an insider look into Speedrun and its special sauce attracting the top AI founders.

So it's like inverted fundraising experience where, you You say, hey, I'm going to go raise money, and then you get like 50 investors to inbound you.

Plus, what actually gets founders accepted into SpeedRun?

We get tens of thousands of applications a year into SpeedRun at this point. If you were to really simplify it, you would probably say—

Welcome back to the Agentic Podcast. I'm George Bandarian, your host and a general partner at Untapped Ventures. Let's jump right into today's episode.

Where I want to start is the big news. Andreessen Horowitz. What big news?

What big news?

Nothing much.

I don't know anything.

Just a little press release about how the firm announced the $15 billion across 6 new funds.

Thank you. Yeah, we're all very excited at the firm.

I think a lot of people don't understand what this really means. Why did the firm do that and what is the plan?

Right. Yeah. Well, I think what you're seeing reflected is in a simple idea, which is that, you know, this wave of new technology in AI is, you know, potentially bigger than any, you know, innovation and wave over the last couple years. And so if you can imagine the idea that maybe AI is actually going to have a bigger impact than like mobile, you know, that's like, I think that's where we are now, right? And then when you speak to founders now, you get a sense of that same energy of like, okay, this is the biggest thing maybe that's ever happened in tech. Um, and so I think a lot of, uh, you know, what we do here at a16z really just is a reflection of like the amount of founder excitement in pretty much every industry, every layer of the stack. And so yeah, so we're very excited to be able to support and invest in entrepreneurs.

No, I agree with you. And I mean, I think by some estimates, the impact could be the combination of mobile, internet, and cloud put together. AI could be more than all of those.

So that's right.

That's right.

This is it.

Yeah, and with the firm whose sole thesis is AI, in our case, you know, we're really excited about the wave and continuing to work together. One of our mutual investments is in a company called Nexa. I remember I was in SF last year, I was speaking at a VC event and there was this handsome German gentleman wearing all white. He followed me off the stage, you know, tracked me to the bar, I was grabbing a drink, he wasn't, he doesn't drink. And he told me how excited he was about this agentic AI that he was building for the heavy industries. And we ended up investing. His name is Philip Wen, he's the co-founder and CEO of Nexa. Shortly after our investment, they got accepted into your program, into SpeedRun. You get hundreds, if not thousands of applications per SpeedRun cohort. What jumped out and made you invest in Nexa?

Yeah, and yeah, the bar is definitely high. Yeah, I mean, we get, you know, tens of thousands of applications a year into Speedrun at this point. And I think in the case of Nexa, you know, the interesting thing with investing super early, as you know, is it's so much about people. And there's a funny metric we've talked about at the firm that sort of, you know, words versus numbers ratio, right? And because there's an expectation that as soon as you're a startup that's raising a Series B or Series C, okay, better have a lot of numbers in it at that point. But when it's a Series A, when it's seed, when it's pre-seed, a lot of the assumption that you have to have is that the idea is going to change significantly. From there. And so with Nexa, I think the most interesting thing with it was this is a company that by the time that they applied, they were already, and they're working on AI tooling for sort of industrial applications. And the most interesting thing with it is not only do you have Philip who's just has already had a really significant career, you know, exactly in the industry that he's going after. But he had already actually signed up a bunch of customers. So there was the good news is it was an early investment for us where when you were looking at the words versus numbers ratio, there's actually a lot of numbers already involved, you know, in the business. Like you could tell that it was already working.

So that was one of the reasons why we were so excited. Like you said, you have a founder that has the domain expertise, he has a great co-founder in David as the CTO, so great pairing there. And the, you know, we like the 2 or 3 co-founder combo, so great balance with the visionary sales charismatic Philip and the proven builder David. And from there, they're going after helping with the Genentech workflows in construction and mining and lighting and for the industrial engineers and the workflows and business processes in those areas. So I agree, and they did great. I believe they— you had your last demo day, they— you put them first. It was Philip's birthday. He kicked it off really great. They went on to raise seed round, I think $9 million led by Construct Capital. We followed on, you followed on, and, you know, I think it was through that work that you and I got to hang out more and get closer and do more fun stuff together. So, and also it was through that demo day that I got a much better appreciation of the SpeedRun program and I was blown away. I mean, you know, we're at all the demo days of the top ones and there's only so many top ones And one after another, one after another, one after another, all of the graduates of the Speedrun program were impressive. How did that happen and what is Speedrun and what do we need to know about it?

Yeah, well, yeah, so the way that this journey started was in the last 3 and change years, Initially, my fellow general partner Jonathan Lai and myself, we had kicked off this whole effort to look at this intersection of AI, entertainment, gaming, and kind of how that was all gonna intersect, right? So that's where we started originally, and we built a new fund at the firm focused on that. What we quickly realized was, you know, wouldn't it be great if we could take a lot more risk and work with companies even earlier, particularly if they're doing things that involve creative risk? So, you know, one of the big things that naturally leads to is to look at all of the functions. I mean, this is like part of the core DNA of Andreessen Horowitz, which is we have this big operating team And when you start looking at the functions, you look at marketing, you look at talent, you look at, you know, fundraising, you look at, you know, immigration, you know, issues for the founders, you know, you look at partnerships, you look at, you know, growth, you look at, you know, all these things. What are the unique capabilities that you can deliver for the earliest stage companies, maybe at the point when they're, when they're pre-product. And so for example, if you think about launching and marketing a product, um, the, the focus might actually be, okay, well, if you are a prosumer product, how do we get you to, uh, launch on social media, have a waitlist, have the waitlist also as people put in email addresses to be able to enrich those email addresses to figure out what the companies and job titles and, you know, people that you're interacting with so that then you can then start to collect your first couple, you know, design partners, you know, for, for, for your new, uh, product. And that's the kind of thing that a brand new company is going to like. It's insanely valuable for them to get their first couple design partners versus, you know, um, if we were in, in, you know, writing a check for Series A or Series B, they better have customers at that point. They better have at least the beginning of a marketing team that can help them with this. But this is part one of the core things that unique ways that we tried to approach this problem is to say, okay, well, if we're going to have a cohorted program to help founders and we're going to intake, we're going to work with 50 companies at a time over 12 weeks, Well, what are all the programs that we can run that are specifically about helping them inflect between day 1 and day 90 that they're with us to be able to really tangibly and concretely help them launch their products, try to help them make their first couple hires, help them deal with their immigration issues, like all these like nuts and bolts things that,, you know, the earliest companies, especially when they're just 2 or 3 founders that, that they have to deal with.

Um, and so, so, you know, as I mean, and all of that culminating in the demo day and, and, and helping with a, hopefully a successful big seed round at the end, right?

That's right. Yeah. I think, I think the, the, the thought is that, um, you know, if you're, if you're, uh, if you're an entrepreneur, very, very high up on your list of, of, uh, responsibilities is, don't run out of money, you know. And so for, for the folks where they want to fundraise and, you know, most folks take it as a huge opportunity to do so, we want to be able to introduce them to, you know, dozens upon dozens of investors. And so, yeah, so the demo day that, you know, we're very happy that you were able to attend. But, you know, our goal there is how do we make it easy for, you know, a founder who maybe doesn't have might have an early network, they may have a network of like 20 investors that they've gotten to know really well and they have a spreadsheet and they have a LinkedIn and all this. But how do we make it so that rather than this, the typical thing where you sit down, you make a spreadsheet of all the investors that you're gonna talk to, that you have talked to and you're gonna email them and then you're gonna book these meetings. Instead you have, you kind of invert the whole experience, right? So it's like inverted fundraising experience where, you say, hey, I'm gonna go raise money, and then you get like 50 investors to inbound you, right? How do you create that experience, right, for founders?

Right.

And so on the founder side of things, um, if you're able to do that, then, you know, if, if things go well, then hopefully within a week you're done with your fundraise and you can get back to work. Like, how great is that? And then for investors, I think the opportunity is, well, you know, what a time-efficient use of your, you know, Tuesday to be able to walk in, watch 6, you know, 50, 60 presentations. They've all been curated by us. They've all been, you know, been working with us such that you can, you can, you can see the tangible impact. And so investors, it's this thing where you kind of know, okay, cool, if I, if I like one of these guys, I'm going to be able to, you know, go backstage after the presentations and you can talk to them and you can do your work.

And the most recent cohort, it felt to me like it was an AI accelerator. Uh, the majority were AI comp— great AI companies. Like I said that, you know, I was like, okay, I'm interested in that one. Next pitch. I'm interested in that one. Next pitch. I'm interested in that one. Um, and so yeah, maybe you can clarify what's the focus going forward. And as it relates to the overall ecosystem, what founder or startup is this program the best for?

Yeah, well, I think, you know, these days where I'm most focused is let's just figure out how to work with the best founders when the companies are brand new. And that matters more than anything else. And so of course, like, it's great if it's adjacent to some of the things we know best. And by the way, over 50% of the companies are B2C companies typically, as an example, that percentage moves up and down a little bit here and there, but we always want to keep that part of our heritage. And so we just want to work with founders ideally as soon as they're ready to start something. And so one of the funny stats that we've had is I think we had a dozen founders in the last cycle where they they had jobs, they applied to Speedrun, we gave them offers and they quit their jobs. Right. So it doesn't get any earlier than that, you know, you know, at that point. And sometimes they have ideas. Sometimes they have ideas that we think are bad. Sometimes they have ideas that we think are good. The ideas always change.

You know, that's part of how it works. That makes a lot of sense. So working with all the founders in the cohorts, what do the best founders consistently do?

Yeah, well, I love when I'm able to work with speedrun founders and they're incredibly, especially in a B2B context, incredibly responsive by, you know, email and text and the whole thing. And I think it's an easy theory to have that, you know, people that treat their, their, you know, their work as part of like with a service mentality, that is one of the big things. I also love, frankly, when folks are just like sponges for the category that they're in. How can you make sure that if you're working in an industry like social networks, that you know every single product that's been built, and all the reasons why something's worked and not worked. And this is, you know, like my, my friend Balaji's, you know, idea maze, like knowing every twist and turn is just incredibly important. Someone that, that is clearly like aggressively trying to assimilate and create that idea maze in their head so that they're not going to make kind of all the, all the, all the obvious mistakes.

Gotcha. Okay. And so then what is that bar to get in? Yeah.

What are you looking for? Yeah, I wish it were formulaic, you know, because in fact, actually, there's, we have an ongoing, you know, project to figure out, you know, when we have, let's say, 10,000 companies apply and, you know, we have our investment team literally goes through every single company, you know, line by line by line, how, like, what are actually the features, you know, in our kind of human review process that causes us to score an application higher or lower. If you were to really simplify it, you would probably say, okay, either you have done something really interesting already with your company, and that means that you have something that you can talk to us about, right? Whether that's metrics, whether that's customer logos, whether that's an open source project that you've put together that has like crazy traction, whatever it is, You have something really concrete. If you don't have that and you have an idea, great. Okay, then we'll probably spend a lot more time then thinking about the founder's background and their motivations. You know, what is their unique capability? And in those cases, you know, we're just very open-minded to sort of elite talent from anywhere and many different markers to show that. They've done some amazing things in the past. And so that doesn't just mean the obvious thing of like, oh, okay, well you went to this top school and you worked at this brand name company. And by the way, sometimes that's a really great indicator. Sometimes that's an indicator that somebody's like a follower, right? And maybe not somebody that's gonna be the CEO and founder, but more importantly is like, well, in the past, have you done something really interesting? If you're a world-class athlete, if you are somebody who's shipped a project on GitHub that has like a bajillion stars, if you're somebody that has like a huge social media following, like each of those things is a demonstration that somebody has very elite talent. And then of course sometimes that's, you kind of get something where somebody has a little bit of each And then we have to debate a little bit what that means. But I think there's also really, we have really nuanced conversations about, for example, whether or not somebody who's a solo founder, is that a disqualifier or not? One argument would be like, well, if you're a solo founder, not only have you not been able to convince a co-founder, how good are you going to be able to convince a customer? That's sort of like the very first job is can you find one person who wants to follow you? On the other hand, we often find somebody who starts a company and, um, you know, maybe they don't have a co-founder, but maybe they've already hired a couple great people. Um, and they're a solo founder with a supporting team already, and they already have some insights and maybe they have some, you know, early customers and maybe that, maybe that's great. You know, so I, I, I think, I think we try not to have like too many rules. In how this works because a lot of the judgment is that. And once we get to a point where it's all rules, then I guess, yeah, then it'll be some decision tree algorithm that will spit out a number and then we'll just decide based on that. But like, I'd like to think that venture capital is, that there's enough fuzziness in it.

Yeah, I agree with you. And like, as an example, the solo founder thing is an interesting point because it's evolved, right? We see some of our best founders using agents and incorporating agents, and so you're, you know, like you were saying, typically start with 2 or 3 founders, make your initial hires, your founding engineer, your salesperson, but now there's agents for all those. And so there, there's the ability not necessarily not to make any hires, and I don't know about the one-person, uh, unicorn and, and, and things of that nature, but I, I definitely think that with smaller teams there's a lot that they could do.

Well, I think, I think, may, may, you know, to your point, I think it's unclear what the form factor for the future of a company might even look like, you know, like we— it— people forget. But, you know, there was a period where, you know, the big automotive companies, you know, they were the first to really create, you know, get big enough that you had to have business units and like functional departments and, you know, the whole thing. And I think it was GM and, you know, Alfred Sloan that like created all this. And then, you know, McKinsey and many of the management consulting firms then went and taught that. To a ton of the corporations, you know, over the, that was one of their core, but before digital transformation, there was like, you know, this is sort of like corporate transformation of, you know, this is how you set up corporate structure and all that. Exactly. Yeah. That was a major part of their business. And so, you know, somebody had to like invent the structure of these companies and there's all these heuristics of like all the things that we've come up with. And I think we're at a point where, yeah, maybe we're going to go blow all this up. And I think the extreme is maybe to think about this idea that we've had content creators for a long time and content creators are often like very young people. They're often individuals and maybe code becomes almost like another kind of content. And so in a world where social media is primarily dominated by youth culture and young people and people creating content, Doesn't that imply that because they're so time rich, doesn't that imply that most of the code on the internet and most of the software that's being written in the world will mostly be kids writing software potentially for other kids? And then maybe they are solo. And the current instantiation of that is looking at Roblox. I'm on the board of a company called Voldex and they've been going and buying and building kind of a long series of Roblox games. And one of the recent ones that they acquired is called Brookhaven, which is like, it's the largest, it's like over 100 million active players kind of thing. And this whole thing was written by, I think like 2 guys and has more active users than 99% of what's in the games world, for example. And so, And similarly, we've had this whole wave of people that built dropshipping businesses and people that, so I love the idea that maybe the future will be individual, one or two people, the same dynamics that shape Shopify or that have shaped Roblox or YouTube content creation or Substack is coming for the software world. Which will be wild, but I think that's where it's all going.

So it's an exciting time and I do want to see and touch on some more interesting things about Speedrun. So community, community is a big part of Speedrun, right? Maybe you could talk about that from the cohort to alumni to the investor network around it.

What role does community play? I had the, you know, very lucky experience of moving to San Francisco in 2007. And that was kind of the beginning of Web 2.0. That was, you know, when the iPhone was released. That was, you know, when we also went from this period where you had very early stage VCs, you know, literally first round capital and, you know, Josh there and, you know, Mike over at Floodgate originally, you know, Maples Investments, you know, Ron Conway, you know, all those guys. Like that whole era was one where the community was very small in San Francisco, but very clearly a super power of that dynamic was that somebody knew, you know, I was living in Seattle at the time when I was 25 years old, could move to, you know, the Bay Area and all of a sudden you just meet a ton of other people that are all kind of interested in the same things. And you had these investors that could support you and you were quickly integrated in the system, which I think ultimately like this is kind of like the secret sauce of What it is that makes Silicon Valley really San Francisco these days, like special is its ability to do that. And so I think from a speedrun standpoint, I think what we really seek is the ability to sort of like replicate some of those dynamics and just do it in a structured way. And so if you have somebody that is, we have founders that are coming from, we have half dozen companies that are coming from Israel, and from, you know, we have a whole bunch that are coming from London, coming from, you know, Asia, coming from many other places. We want to put them into a structured environment where, you know, they're in a kickoff, they, you know, are in smaller groups that are aligned. You know, if you are a prosumer company, you probably want to talk to other prosumer companies. And how do you just sort of like create that shared experience? And so we house them all at the a16z offices in San Francisco. So they're coworking working every day together. And then, you know, they're doing office hours every week. And so you just kind of, you know, create all of these environments where whether it's, you know, sort of productive, you know, like these things where they're trying to learn things or it's like purely social that you just have like your, like your crew, you know, you just, you just want to get them connected in that. 'Cause we often find, you know, as the speedrun team, we want to be able to help and support founders on everything. But sometimes they, before they want to ask us something, they want to ask each other, like for Rocket Money. And frankly, like sometimes we don't know the answer and sometimes it's better to just get a survey of everybody. So we try to create that environment and obviously it's such an intense experience that at the end of it, hopefully people feel like they've made friends for life. And one of the roles that actually I'm, you know, we're hiring for right now is an alumni experience head. And so this person will actually design and create like, okay, you go through the speedrun program, you're done a demo day, you graduate, congratulations. Okay, well what is the speedrun experience after that? Right. Right. And there should be hopefully an experience that lasts for years and years and years, even if you've left your, you know, the company that you originally started or you exited the company you originally started.

There are different ways to work with us as, as a16z and as a16z Speedrun. Then that gets into growth, and that's an area that, you know, I know both of us are really passionate about. I know you've had a very hands-on vantage point both as an operator and an investor, and obviously, you know, you are the expert when it comes to this cold start problem, having written the bestselling book. So let's talk about that. I mean, obviously the environment for those ideas was very different back then. What aspect of it do you think still applies and has stood the test of time? And what aspect of it should we modify for the current AI era?

Yeah, one of the first things I'll point to is that when we got the internet and the web browser and all that, it came with SEO, right? It came with people being able to share links and also then further email invites and all this great stuff. When mobile came out, we got the App Store. And then as part of that, we had text messaging. We had, yeah, exactly. We had mobile ads. We had a bunch of that. One of the really interesting, difficult questions right now, I think for anybody that's building kind of consumer-facing or prosumer applications with AI is, you have this incredible engine and technology underneath it, but what is the new distribution that you're getting above and beyond what existed in the past? And I think what you're seeing is it's not like the browser and mobile where new kind of consumer surface area is getting created that then that surface area then requires you to number one, rewrite your whole experience. If you built a a travel website, you know, then when mobile comes out, you're gonna have to redo the whole thing, right? That doesn't quite exist, right? So I think the AI era has been interesting in that it feels like it's easier for incumbents to add AI by adding like the Chrome, you know, Gemini button or, you know, adding, oh, Baba to WhatsApp, you know, into all their text fields and that kind of thing. That's one thing that's happening. And they're able to use all of their preexisting, and I love clicking that little Groq icon on Hacks, they're able to use all their existing surface area. And then I think for all these new companies, the question is, okay, well, if you make your product AI native, if you're gonna build a brand new experience that's mostly chat-oriented, let's say, rather than, that's powered by an LLM rather than having it be like a little sidebar, is that enough of a consumer upgrade and like an end user upgrade that they're gonna be, that they're gonna, you're gonna have vastly different retention curves or monetization curves or otherwise. So I think that is like really one of the core conflicts right now. So let me call that out and put that aside. Second one that I think is interesting is this like B2B versus B2C thing. In B2B you're seeing so much top-down interest where executives and leaders and investors and everybody are like, have you thought about putting AI, you know, what's your AI strategy, you know, et cetera, such that if you're selling a product to other businesses, there's just this magical pull from the market that's happening. And if you put AI, if your company was you know, product.com and then you call it product.ai, all of a sudden people want to try it more. That's in B2B. The interesting thing in consumer, now that I've dug through so many companies, has been consumers still actually have the view, I think, that we're at this point in the cycle where if it's something is like.ai, that means it's worse, right? It means like you've taken my travel company where I could interact with a human and you've now made it so that I can, you know, I have to interact with this AI thing.

Or hey, instead of—

Got it. So you feel that there's early adopters that might consider it better, but the mass market considers it worse 'cause you've dehumanized the experience from it.

Yeah, well, and you know, not only dehumanized, but also just the idea that the actual level of quality of the output is worse. Got it. Right? And so on the consumer side, I think what's happened is, and you can look at this across many verticals, certainly within social media, that's why people kind of talk about AI slop, right? Certainly in gaming, if literally if a company says that they are doing AI art in their game, consumers are up in arms about it. And then, yeah, and then if you have a product where it's meant to be a concierge experience and then all of a sudden you're actually dealing with an LLM, that's very negatively perceived. So I think there is unfortunately like a broad base like negativity on the consumer side for this right now, right now. And of course I'm an optimist, of course I'm like all these things, but I think because of that, this is again, I'm just adding kind of like, this is like one of the things that are in the conflict And then maybe the last thing, you know, I think I'll, you know, add on all this is that, you know, is that AI, just the inference costs are actually expensive right now, right? We're gonna get to a point, and I think it's, you know, within sight where you can have an AI consumer experience. It will, every screen that you tap on, every thing that, every feature you use, will trigger, you know, an LLM in the background. And so there's, there's gonna be a shitload of inference costs and also you're gonna be able to just put ads on it and then the economics will all work out. We're not there yet. We're not there yet. And so what, so naturally as a result, what we all see is people build these mostly prosumer experiences. They're, they're productivity oriented. They're asking people to put in credit cards at the very, very beginning., and then they're charging you for per token or per seat or, you know, per whatever. And it's much more of a productivity kind of businessy use case, right? Because we're not at the point where you can do the Spotify ad-supported freemium subscription thing and the math all pencils out.

Right, right. That makes sense. And when you wrote the cold start problem, um, have obviously evolved. You've written that marketing channels suck right now. A lot, a lot of them. Yeah. Uh, some of the new strategies, tactics, playbooks, right, that we have, We've moved to, you know, first we had product-led growth and then now we have founder-led sales. We have FTE for deployed engineers, we have AEO, GEO, right? AI search engine optimization. You still have other channels like TikTok that people talk about as relatively green pastures. So from your vantage point, what do you see? What's working? What's not? How would you advise early-stage founders trying to get initial traction.

Yeah. Well, you know, I'm going to, I'm going to take a little segue here and, you know, just, just talk about like, just one of the presentations I put together when I first joined a16z was a presentation that where the title was, you know, technology changes, but people stay the same.

Right.

And, and I think it's just the recognition that especially when you're thinking about, you know, marketing and growth, that the same ideas actually appear over and over and over and over again. But they appear, they manifest in different underlying technologies. And so, maybe I'll start with the headline on this. The headline is like every single thing that you can think of that's like coupons, referrals, advertising, whatever. I imagine every single one of those, just given how big and important the AI wave is, will remanifest itself in a new form, right? And my argument for this is actually, if I go back to like coupons as a great example, people often don't know this, but the reason why Coca-Cola is what Coca-Cola is today is because it was the first company to actually use coupons. And so I wrote about this in in my book many years ago, but Asa Candler was the first to actually create a little thing that said, hey, if you take this piece of paper, you can get a free Coca-Cola if you take it to your local soda fountain. Now, the interesting thing with this soda fountain, think of it as like, it's like a bar where they were serving many different kinds of fizzy drinks. And so naturally all these customers show up at the local soda fountain and if they keep asking for Coca-Cola and you don't have it, you as a business owner, well, you're gonna be like, well, I better get this thing, otherwise they're gonna go across the street and you know, and so on. And that was the actual problem that the coupon was meant to solve, which is actually how do you get B2B distribution for Coke products in the first place, which is super interesting. But if you kind of zoom out and you're like, well, why is that super interesting? Well, the reason that's interesting is because coupons could only exist because newspapers existed, right? And newspapers could only exist because like the steam-powered printing press existed. So if you have the steam-powered printing press, this is all kind of like early to mid-1800s. If you have that and then you can actually print newspapers, then naturally what will happen is you have a new surface area for consumer engagement. And when you have new surface area for consumer engagement, that's when the marketers can come in and figure out new innovations to kind of attach to all of that new attention. And so no wonder over 150 years later or whatever, if we have search engine results, if we have websites, if we have mobile apps or any new consumer surface area, then people are gonna reinvent, continue to reinvent the concept of a coupon just in different forms. So I certainly think that, you know, there will be like, given the amount of time, at least, you know, for myself personally, that I'm interacting with LLMs on everything now, that that is such an incredible surface area for consumer engagement that, you know, that the natural push will be that all of that will eventually point to other things. And sometimes that will be commercial and monetizable and sometimes it will just be, helpful to you. And maybe it'll be Mark-sponsored, maybe it won't be Mark-sponsored, maybe it'll tell you about stuff that you might want to buy, not just answer. That would be the hope, right?

It's all coming, I think. Because this is the perfect time to actually give me an ad of something that's 100% relevant to what I was searching for and likely to buy, right? It would be advertising at its best. And if it's done that way versus the highest bidder,, right? So if it's relevant, if I'm searching for Hawaii deals and Hawaii flights and Hawaii hotels and it's from their Hawaiian advertisers, right? And it's the best flight deal relevant to what I searched on, then that makes a lot of sense and it's going to be high conversion for the advertiser. It's going to make the consumer happier. So I think there could be some interesting models where it provides a delightful experience if done right.

And actually, you know, I encourage everyone listening to, you know, just plug into their LLM, like, you know, tell me everything that you know about me, you know, and what you quickly realize is that the kind of advertising and kind of media profile that is being built about you is far, far above and beyond, you know, anything that's available on just like, you know, online ads kind of more generally, you know, because what, you know, what's happening for the rest of the online ads business is you are being cookied, you're going to different things, you're being retargeted, and sometimes you end up, there's these various services that let you do this where you can go and inspect all of the user segments that you've been placed into. And so it'll say, oh, well, you're like a car interested, high-intent interested, car automotive shopper, right, for instance. But it doesn't have, the point of view of like, tell me about an SUV that is going to be the best one for the new kids that I'm having next year. Right, right. You know, like that is like a whole new kind of level of like marketing information. And one day, you know, we started this conversation talking about Coca-Cola. There will be some kind of coupons format, I assure you, you know, that sits inside of, inside of every LLM.

Right. So, moving from growth to retention, you've also written that retention is hard and, you know, you were talking a little bit on the B2B side, you know, on the plus side, there's a lot of interest, there's, you know, board level all the way down to executive leadership mandates to experiment with it. That's great, but I think what a lot of us acknowledge is that there's also going to be a lot of churn, right? Because it's an experimentation budget. And not everything's going to stick. You know, many cases they say, okay, we've identified a use case, the use case is this process, okay, go try 5 different applications. Not all 5 of those use, you know, apps that they tested are going to win. It might be 1 or it might be 0 in the end. And so there's concern about what the churn numbers will see after that 1 year.

But give us your update on state of retention. Well, I think the Very two very interlinked topics. Both, you know, I wrote very grouchy, you know, complaining, you know, blog posts about, you know, first that all the distribution channels suck right now. And then also that retention sucks, right? And then I'll talk about, you know, and yet of course we're seeing companies that are obviously adding insane amounts of ARR very quickly. So like, let's talk about that too. Like why are they succeeding kind of in spite of that? So The first thought on marketing distribution sucking is we're sort of in this world where as we were talking about, like we haven't had all these new distribution channels created and so we're still existing on kind of like the old stuff, right? This is SEO, this is paid ads, this is email virality, these are all the things. And what you realize is, well, SEO is like way too slow, so that sucks. Paid marketing, okay, well, do you want to compete with, As a startup, do you want to pay money and compete with all these other companies that are going after the same customer and they are like hyper-optimized as to what they can do? Probably not. Super expensive. It's annoying. Social media landscape, crazy, right? Like very, very busy, you know, et cetera. And so I think one of the things that you are seeing over and over again and that relates to retention is that a lot of the, most effective early, early stage startup tactics right now have to do with awesome launches on social media. And so if you're on AI Twitter or, you know, I don't even know what to call it now, you know, it's like X, the X community for the AI enthusiasts all hanging out together and you're able to do a really cool launch video and then everybody tries your stuff. That's awesome. But you're also getting this really kind of potentially like flighty kind of early adopter group of users that come in. And no wonder, I see a lot of data rooms of kind of many AI companies, many AI products. You see that the retention curves are absolutely awful. You see when you get a big spike of users, those users extremely experimental, 30 days out, you'll have lost 95% of the users, right? I mean, just like huge drop-off, huge drop-off. And so you're kind of like, well, what do you do with that? And this is repeating over and over again. And so then you're like, okay, well, these spikes are happening all the time whenever there's maybe like a new product release or whenever there's a new underlying kind of upgrade in all the models that then like increases all the capabilities, then you kind of get another spike. But I think this is, you know, this also comes back to why is it that AI, I think is, you know, products are working even better on B2B and prosumer over consumer is that if you have a retention curve that sucks, there's only a couple ways for you to turn that around, right? One version of it is, okay, well, you kind of remove all the enthusiasts. It turns out you're really creating value for a smaller percentage of people. Maybe that, maybe it is only 5% of your users, but with those 5% of users, you can then go figure out, oh, they work at these companies and then you go sell to those companies, right? And then all of a sudden it's like, yeah, even if your per user retention curve isn't great at an account or logo level, all of a sudden you're starting to grow, right? So, so that, that's like one, you know, amazing way that, that, that can happen. The other is of course individual kind of like just enthusiast programmers that are just using like hundreds of millions of tokens like per day, you know, just like working on random projects, right? You know, and like, how awesome is that? Well, from a retention curve standpoint, what that means is, yeah, maybe you lose 95% of your users, but maybe the 5% that stick are actually, you know, so engaged that from a purely revenue standpoint, they actually cancel out all the churn. And so what ends up happening is, is that is how you end up, um, uh, seeing this paradox of both really, I think, flighty user bases. And then in combination with that, that like when it does stick, and especially when it's in a B2B environment, they seem to grow incredibly fast as well.

All right, so now let's shift into how AI is changing how we build and ship product, right? So a year ago you wrote another great seminal blog post on this idea, this new nascent idea of vibe coding and how LLMs are getting good enough that people could sit there and just kind of vibe code a product and wow, it feels like that was another one of these AI light years ago because that field has progressed so much. What has surprised you the most about how massively the coding agents and vibe coding has taken off?

Yeah, I mean, if you had told me just how good LLMs would be at coding, I mean, it's just unbelievable. What it's able to do. I, like many of my coworkers and many folks out there over the last couple weeks, have been simultaneously trying Cursor and all their new models and Antigravity and Claude Code and building Ralph loops and doing all sorts of stuff. The project I was doing over break was actually just taking, how do you take a single, line prompt for a product and then turning it into a really detailed PRD and then feeding it into Cloud Code to go code for, you know, an hour plus on your super in-detail PRD. And specifically what I wanted to do was to be able to, um, you know, give it something like Windows Notepad, but with a twist, right? Like a product that's a well-known product with a twist. And then that way you'd have a really detailed PRD of like everything that all the features and all the things that you'd expect from the known, but then be able to give it like a little bit of a like kicker and then just see like what, what that experience would be. And so I think I probably built like, you know, 50% of a dozen different projects like over, over, over the break, you know, between, between, you know, evenings and weekends on, on, on, you know, while we were on, you know, vacation, you know, do that. And so I just think, I just think it's been like incredible to see the progression. And then today I was just reading about how,, you know, the Cursor team, um, had a bunch of agents, uh, coding nonstop for a week to build, uh, like a browser. Like it was a one-shot on a browser. Um, and, and, and just like, like this amazing thought that, that, uh, you know, you can have this like week-long, you know, sort of lack of intervention and it, and it can just like code something, you know, obviously it works better when it's a well-understood thing, like like a browser. So I think all this is to say is, you know, I just kind of think about this idea and it's hard to extrapolate of like what happens when writing software is like so low cost that it's almost disposable. You know, it's a disposable piece of content, right? The way that we think about like a photo being, you know, it used to be that we, you know, we had film cameras and you had to like ration like how many photos you would take. And then now we're like in the digital world. So it's like how many photos you wanna take? I don't know, you take as many as you want. Like there's no sense of like cost, you know, I think building software products kind of like has the potential to sort of be that way. And so I think the thing that I imagine is, yeah, what happens in a world of that? Well, so let me speculate on a couple things. So first, what you might think is typically people undertake software projects and they're big things. And so of course you either want to build products that are for millions of people, apps that are for millions of people, or if you're gonna build it for a small number, then, um, you know, they're gonna be businesses, you know, but you, but you don't really think about building products for, you know, like your 5 family members. Mm-hmm. You know, you don't think about building, you know, like disposable products like for yourself, but like, yeah, we'll, we'll certainly get there. You, you also think about building products almost like, like we don't have the idea of like, like a meme on the internet is a photo or an image like that has, that somebody will create in reaction to some world news. Oh, there's a thing happening in Iran, or oh, there's a thing happening in, you know, in the stock market. Okay, cool. I'm gonna make a little image about it. But you never think of somebody being like, I'm gonna build a software, a meme app, a meme app, you know, for that. But it could be a cool idea. But yeah, but, but we could get there, which is to say like, oh, you know, well Maduro gets, you know, like taken out and so, you know, he's, he's been, he's been captured. And so that day an app gets created that is like a simulation of that experience and you can play through it and, you know, the whole thing. And then the next day, zero people use it, right? You know, but for the one moment where it's super relevant, it is like a super interesting product and it may maybe make some money because people throw some ads on it or, you know, whatever.

Right.

Um, uh, you know, and, and so, so, so I think although this form factor of, you know, what we think of as software, I almost think about it like Yeah, what if it's just as easy as it is to record a video and like what happens? And you know, as I mentioned earlier, I think it also means that a lot more kids actually end up making software. They may end up making software for each other. It means that you get a lot more randomness in the kinds of software that people create. But also, you know, just because MrBeast can make a MrBeast-style video and you can also like look at that and you can make the exactly same video, that doesn't mean that you're going to be as successful as MrBeast. Well, why is that? It's because of brand. It's because of IP. It's because of like these moats that are actually not typically things that like a software company would think of as like that being a moat. But that might be just actually be the case, you know, over time. And then, you know, rather than having a company think of itself as like, you know, because today what we do is we build like, you know, Microsoft Word and then we operate it for decades and people just like buy into the franchise. You know, there's another model that's quite popular right now and you see it with like various like Turkish game studios and you know, that kind of thing where they will go and make copies of like many, many genres of things and their entire portfolio may be like hundreds of different games. Well, it's like maybe what, and in the same way it's like, well, you know, as a blogger, rather than being an author that writes like 3 books in my lifetime, instead I've written like 1,000 essays. Maybe that's what happens. Maybe if when you're, the future of being an app developer is you actually write thousands and thousands and thousands of pieces of software and many of them are ephemeral and you build a following of a certain type of software that people like. And, and that's the future. Like, who knows, right? I think that's kind of the part of the hypothetical that I'm I'm super interested in. And then we talked about this a little bit earlier, but part of the reason why we've had these kinds of delineations between jobs, a marketer versus a product manager versus a software engineer, a lot of that has to do with just the fact that the roles are— coding is hard, right? Programming is actually super, super hard. You have to go and like study it and learn it and it takes years and then one day you can finally do it. And so the entire aspect of why growth teams ended up being created in the industry was because you needed a combined unit of like marketers and PMs and engineers to actually ship things that would move growth. So, but I mentioned that because like what happens when every marketer actually can output code themselves? Exactly. Like how does that actually change the dynamic? Well, I think we end up changing all the job titles because none of them are relevant anymore. And it may mean that like, you know, if you're a marketer and rather than having, you know, the mobile phone was like the supercomputer in your pocket, this is having like superintelligence and like an infinite, you know, like access to labor and you have like Nike's digital agency like in your pocket. Like what can you do with that? Right. And I think it's going to just create a very different structure to like the idea that maybe every job will include software creation as part of the job. And then every software experience you have has software creation as part of it, right? So you have extensibility in all the software products and then all the jobs, like, you use that extensibility as part of doing the job, right? So I think there's like, we could sit here and speculate on many, many different avenues for where this is going.

But yeah, it's certainly very exciting. Yeah, and I mean, I think a lot of what's exciting about it is that a good part of what you said in many ways is here, right? Those, the non-technical people can use the vibe coding tools to build stuff that's workable right away. It may not be scalable to thousands of simultaneous users, but the disposable apps, whether you're using the lovelables and bolts or whatnot, and then for the engineers, right, more of the coding agents, to make them 10 or 100x better, faster, and increase the quantity and sometimes the quality of their output. Those are here, they're being used, they're being adopted in astronomical numbers, and I think it's really amazing. Andrew, thanks again. This has been a ton of fun. For everybody listening, tune in for the next episode, and signing out. Wow.

That was an amazing conversation. Huge thanks again to Andrew for joining us and sharing his perspectives on speedrun, product and company building in the GenTech age, and what really matters at the earliest stages. So if you want to follow Andrew's work or connect with him, you can find his writing on Substack at andrewchen.substack.com. You could also find him on LinkedIn. And if you're a founder interested in the A16z Speedrun program, definitely definitely check it out at speedrun.a16z.com. Now, during the episode, we discussed Speedrun along with Nexa, a company we co-invested with Andrew. And in an upcoming episode, I'll be sitting down with Philip Wen, the co-founder and CEO of Nexa, to hear directly from a founder who's been through that journey and very successfully. So stay tuned for that. So if you enjoy this episode, make sure you subscribe to the Gentic Podcast so you don't miss future conversations with AI founders and investors building at the forefront. And if you're building an AI company and you're in the pre-seed or seed stage and you're interested in an investment from Untapped Ventures, these are exactly the kinds of conversations and companies we love to chat with. We write checks of up to $1 million and work closely with founders early on. So you could learn more and pitch us at untapped.ventures. Thanks again for listening, and we'll see you in the next episode.