Vertical AI Agents Could Be 10X Bigger Than SaaS

Every 3 months, things have just kept getting progressively better. And now we're at this point where we're talking about full-on vertical AI agents that are going to replace entire teams and functions and enterprises. That progression is still mind-blowing to me.

A lot of the foundation models are kind of coming head to head. There used to be only one player in town with OpenAI, but we've been seeing in the last batch, this has been changing.

Thank God. It's like competition is, you know, the soil for a very fertile marketplace ecosystem. For which consumers will have choice and founders have a shot. And that's the world I want to live in.

Welcome to another episode of The Light Cone. I'm Gary. This is Jared, Harge, and Diana. And collectively, we funded hundreds of billions of dollars worth of startups right when they were just 1 or 2 people starting out. And today, Jared is a man on fire and he's gonna talk about vertical AI.

Yes, I am. I am fired up about this because I think people, especially startup founders, especially young ones, are not fully appreciating just how big vertical AI agents are going to be. It's not a new idea. Some people are talking about vertical AI agents. We funded a bunch of them, but I think the world has not caught onto just how big it's going to get. And so I'm going to make case for why I think there are going to be $300 billion plus companies started just in this one category.

Nice.

I'm going to do it by analogy with SaaS. And I think in a, in a similar fashion, people don't understand just how big SaaS is because most startup founders, especially young ones, tend to see the startup industry through the lens of the products that they use as a consumer. And as a consumer, you don't tend to use that many SaaS tools because they're mostly built for companies. And so I think a lot of people have missed the basic point that if you just look at what Silicon Valley has been funding for the most, for like for the last 20 years, like we've mostly been producing SaaS companies, guys. Like that's literally been like most of what has been coming out of Silicon Valley. It's over 40% of all venture capital dollars in that time period went to SaaS companies. And we produced over 300 SaaS unicorns in that 20-year time period, which is way more than every other category.

Software is pretty awesome.

Software is pretty awesome. I was thinking back to the history of this because we always like to talk about the sort of how the history of technology informs the future. And the real catalyst for the SaaS boom was— do you guys remember XMLHttpRequest?

Oh my God.

Like, I'd argue that that was quite literally the catalyst for the SaaS boom.

Like Ajax.

Ajax. Yeah. In 2004, browsers added this JavaScript function XMLHttpRequest, which is the missing piece that enabled you to build a rich internet application in a web browser. So for the first time, you could make things in websites that looked like desktop applications. And then that created Google Maps and Gmail and set up this whole like SaaS boom. Essentially, the, the key technology unlock was that software moved from being a thing that you got on a CD-ROM and installed on your desktop to being something that you use through a website and on your phone.

Paul Graham actually shares in that lineage in that he was one of the first people to realize that he could take the HTTP request and then actually hook it up to a Unix prompt. And you didn't actually have to, you know, have a separate computer program that would change a website.

So ViaWeb was an online store, kind of like Shopify, but way back in the day.

Yeah, it was basically like the first SaaS app ever. Like, like PG actually invented SaaS in like 1995. It's just that those first SaaS apps kind of sucked. 'cause they didn't have XMLHttpRequest. And so every time you would like click a button, you would have to reload the whole page. And so it's just a shitty experience. And so it didn't really catch on until 2005 when XMLHttpRequest widespread. Anyway, I see this LLM thing as like actually very similar. It's like it's a new computing paradigm that makes it possible to just like do something fundamentally different. And in 2005, when cloud and mobile finally took off, there is this sort of like big open question of like, Okay, well, this new technology exists. What should you do with it? Where is the value going to accrue? Where are the good opportunities for startups? I was going through the list of like all the billion-dollar companies that were created, and I kind of had this realization that you could kind of bucket the different paths that people took into like 3 buckets. There's a first bucket that people started with, which was like, I would call them obviously good ideas. They could be mass consumer products.. Um, so that's like Docs, Photos, email, calendar, chat, all these things that like we used to do on our desktop, but that obviously could be moved to the browser and to mobile. And the interesting thing is zero startups won in those categories. 100% of the value flowed to incumbents, right? Like Google, Facebook, Amazon, they own all, all those businesses. Folks forget that like Google Docs wasn't the only company that tried to bring Microsoft Office online. There were like 30 companies that tried to bring Microsoft Office Online, but they all lost. Google won. Then there was like a second category, which was like mass consumer ideas that were not obvious that nobody predicted. That's like Uber, Instacart, DoorDash, Coinbase, Airbnb. Those ones came out of left field. Like the, the dot, dot, dot between XML HTTP request and Airbnb is like very not obvious. And so the incumbents didn't even try competing in those spaces until it was like too late. And so startups are able to win there. And then there's a third category, which is all the B2B SaaS companies, and that's like 300 of them. And so like, by number of logos, way more billion-dollar companies were created in that third category than the first two. I think one reason why that happened is like, there is no like Microsoft of SaaS. Like there is no company that somehow does like SaaS for like every vertical and every product. Like for structural reasons, it seems to be the case that like they're all different companies and that's why there's so many of I think Salesforce is probably like the first true SaaS company.

Um, and I, I remember Marc Benioff coming to speak at YC and he tells the story as just very early on, people just didn't believe you could build sophisticated enterprise applications like over the cloud or via SaaS. It was just so, um, there's just like a perception issue, right? It was like, no, like you don't, you buy like your box software and that's like the real software that you run.

The way we always do it.

It was, it was quite contrarian because the early web apps sucked. They were like via web where you had to be a visionary. Like PG and understand that the browser was going to keep getting better and that eventually it'd be good.

Which feels like quite reminiscent of today, right?

Yeah.

Where it's like the, yeah, the same thing. Like, oh no, like you won't be able to build like sophisticated enterprise applications that use these LLM or AI tools because they hallucinate or they're not perfect or they, um, they're kind of like just toys. But yeah, that's like the early SaaS story, exactly the same.

And so when I think about the parallels with LLMs, I could easily imagine the same thing happening, which is that there's a bunch of categories that are like mass consumer applications that are obviously huge opportunities, but probably the incumbents will win all of those. So that's something like a, like a general purpose AI voice assistant that you, you know, you can ask it to do anything and it'll like go do that thing. That's an obvious thing that should exist, but like all the big players are going to be competing to be that thing.

Right. Wow. Apple's a little slow on that one. Why is Siri so stupid still? What year is it? It makes no sense.

I mean, it's interesting. Like, a counter to that is like the very obvious thing is search, and maybe Google will still win, um, on search, but Perplexity is definitely give them a run for their money, right?

Yeah, this is the classic innovator's dilemma at the end of the day. I mean, you could argue, going back to what you said about Uber or Airbnb, these were actually really risky things from a regulatory standpoint. So if you're Google and you have basically a guaranteed giant pot of gold that, you know, sort of comes to you every single month. Like, why would you endanger that pot of gold to sort of pursue these things that might be scary or might ruin the pot of gold?

I think that's— I think that's like probably the primary reason why the incumbents didn't end up building those products and didn't even clone them even after they got big. And it was obvious that they were going to work. Google never launched an Uber clone. They never launched an Airbnb clone. I was listening to this talk by Travis, and one of the things that he said that really stuck with me is that in the, in the first years of Uber, he was very scared that he was going to personally go to prison for like a long time. Like he was actually personally risking going to prison in order to build that company. And so, yeah, no highly paid Google exec was going to do that.

What do you think about why the incumbents didn't go into B2B SaaS? Is it part of the reason is that a lot of the use cases are very— there's a very wide distribution?

I— it's a great question. I love to hear what you guys think. My take is that it's just too hard to do that many things as a company. Like each B2B SaaS company really requires like the people who are running the product and the business to be extremely deep in one domain and care very deeply about a lot of really obscure issues. You know, like take like Gusto, for example, like why didn't Google build a Gusto competitor? Well, there's no one at Google who really understands payroll and has the patience to like deal with all the nuances of all these like stupid payroll regulations. And like, it's just like, like, it's just not worth it for them. It's easier for them to just focus on like a few really huge categories.

In the B2B SaaS world, it's, it's sort of about the unbundling bundling of software argument that comes up a lot as well, I think. And why didn't, why did all these vertical B2B SaaS products evolve versus just like Oracle or SAP or NetSuite. Yeah, NetSuite, just owning everything. And I think it might be also is another thing that's attributable to the shift to SaaS and the internet is in the old ways of selling software, again, you had this box software that was really expensive to install and you had a whole ecosystem around it. And anytime you wanted something custom, the integrators would just say, oh no, we can just build you a custom payroll feature or something like that. And then Salesforce comes along with a SaaS solution and it just seems like it could never be as powerful or sophisticated as the expensive enterprise installation you just paid for. But they proved that it totally was the case. And I think that just opened the gates for all of these vertical SaaS solutions to emerge doing exactly what you're saying.

I think the other problem is that with a lot of this enterprise software, if you're a user of Oracle and a NetSuite, because they have to cover so much ground, the user experience is actually pretty bad. They're trying to be jack of trades, but master of none. So it ends up being a bit of a kitchen sink type of experience. And this is where if you go and build a B2B SaaS vertical company, you could do literally a 10x better experience and more delightful because there's this stark difference between consumer products and enterprise user experience.

Well, there's only, what, 3 price points in software. It's $5 per seat, $500 per seat, or $5,000 per seat. And that maps directly to consumer customer, SMB, or enterprise sales. And then I think time immemorial has taught us that in the past, and this is less and less true with new software, thankfully, but enterprise is terrible software because it's not the user buying it. You know, some high-up muckety-muck inside Fortune 1000 is the person who's getting wined and dined for this, you know, mega 7-figure contract. And, you know, they're going to choose something that maybe isn't that good actually for the end user, the person who has to actually use the software day to day. And I'm sort of curious to see how this changes with LLMs, actually. I mean, to date, one of the more salient things that we've seen for both SMB and enterprise software companies is that— or all software companies, all startups, period— is like, you know, there's this sense that as revenue scales, the number of people you have to hire scales with it. And so when you look at unicorns, even in today's YC portfolio, it's quite routine to see a company that reached $100 or $200 million a year in revenue, but they have like 500, 1,000, 2,000 employees already. And I'm just going to be very curious, like even the advice that I'm starting to give companies that are, you know, a month or two out of the batch it's, it's feeling a little bit different than the kind of advice I would give last year or 2 years ago. In the past, you might say, you know, let me find the absolute smartest person in all of these other parts of the org, like customer success or sales or different things like that. And I want to find someone who I've worked with, who is— I know is great. And then I'm going to go sit on their, you know, on their doorstep until they quit their jobs and come work for me. And I want them to be someone who can you know, build a team for me, hire a lot of people. That might still be true, but I'm starting to sense that, uh, the meta is shifting a little bit. Like, you actually might want to hire more really good software engineers who understand large language models, uh, who can actually automate the specific things that you need that are the bottlenecks to your growth. And so it might result in, you know, a very subtle but significant change in the way startups grow their businesses, sort of post-product market fit. It means that I'm going to build LLM systems that bring down my costs, that cause me not to have to hire 1,000 people. I think we're right at the beginning of that revolution right now.

I mean, we talked about this in a previous episode. We talked about there will be a future unicorn company that's only run, if we take it to the limit, with only 10 employees, that's completely plausible.

And they're writing the evals and the prompts.

That's it.

I think what you're saying is a trend that was already underway pre-LLMs. I remember when I was running Triplebyte, for example, we needed to build marketing or customer user acquisition, basically. And especially after we raised our Series B, the traditional way you were supposed to do that is to hire a marketing executive and build out a marketing team and and just like basically spin up this machine to do like sales and marketing. But I'd actually met like a YC founder, um, Mike, who was— his company was basically building like a smart frying pan. Sounds like bizarre, but like he was an MIT engineer. Yeah, you remember this? Um, he's an MIT engineer, and to sell the smart frying pan, he had to get really, really good at understanding like paid advertising and, um, uh, Google Ads and just a whole bunch of stuff. And so he'd taken this engineer's mindset approach to it, and I remember just talking to him about it and realizing this would be so much better to have an MIT engineer working on our marketing efforts than any of the marketing candidates I've spoken to. And he was able to scale us up to— I mean, we were spending at one point like a million dollars a month on just marketing and various campaigns.

And Triplebyte had great marketing. I remember the Caltrain station takeover that you did, all the out-of-home stuff that you did. It was really high-quality stuff. Stuck with it. You could tell that it was not being done by some VP marketing person.

And that was all Mike. And the comment I would often get when people would ask me around that time, like, how big is Triplebyte? And we were like 50 people. And it would be so much better.

Yeah, yeah.

Everyone would be like, I thought there was like hundreds of people. I was like, no, it's all because if you put a really smart engineer on some of these tasks, they just find ways to make— they find leverage. And now LLMs can go even way beyond the leverage you had with just pure software.

Okay, so here's my pitch for 300 vertical AI agent unicorns. Literally every company that is a SaaS unicorn, you could imagine there's a vertical AI unicorn equivalent in like some new universe. Because like most of these SaaS unicorns beforehand, there was some like box software company that was making the same thing that got disrupted by a SaaS company. And you could easily imagine the same thing happening again where now basically every, every SaaS company build some software that some group of people use. The vertical AI equivalent is just going to be the software plus the people in one product.

One thing might be just enterprises in general right now are a little unsure about what exactly they, like, what agents they need. And one approach I've seen from especially more experienced founders, like Brett Taylor, the CTO of Facebook, started his company Sierra. I don't know all the details, but as far as I can tell, it's essentially more like Broadly about letting enterprises deploy these AI agents and spinning them up custom for the enterprise versus like, oh, hey, we have this specific agent to do this. It's something I've seen from one of my companies called VectorShift that we funded about a year ago. They're two really smart Harvard computer scientists. And what they found is that they're trying to build a platform to make it easy for enterprises to build their own, use no-code or SDKs to build their own internal LLM-powered agents, but enterprises often don't know exactly what they want to use these things for. And so bringing it back, I wonder if in the box software world, you started off with just a few vendors who just basically were trying to convince people to use software at all. And it was just like, it does everything. And then it gets more sophisticated and higher resolution, and you get lots of vertical SaaS players. We go through that same period with LLMs where the early winners might just be these general purpose, hey, we make it easy for you to do LLM stuff. And then it the vertical agents will come in over time, or do you think there's reasons it's different now and the vertical agents will take off on day one?

Yeah, that's interesting because if you think about the history of SaaS, the consumer things worked first. Like 2005 to 2010 was mostly consumer applications like email and chat and maps, and people got, people as individuals got used to using these tools themselves. And I think that made it easier to sell SaaS tools to companies because, you know, the same people are both employees and consumers.

Yeah, I think the answer might just be like, this is all just a continuation of software, and there's no reason it has to reset back. Like, LLMs don't have to reset back to a few general-purpose, like, enterprise LLM platforms doing everything, because enterprises have already been trained on, like, the value of point solutions and vertical solutions. And like, the user experience is not going to be that different. These things will just be a lot more powerful. And so if enterprises have already built the muscle of believing that startups or vertical solutions can be better than legacy broad platforms, they are probably going to be willing to take a bet on a startup promising a very good vertical AI agent solution today. And I feel like we're all seeing that in the batch now, where some of our companies are getting faster traction in enterprises for these vertical AI agents than we've ever seen before.

I think we're just early in the game, right? Like all software sort of starts quite vertical and then as the industries actually get much more developed, then I mean, I just answered my earlier question. It's like, you know, why does a company end up having 1,000 employees? It's actually that, you know, early, early in the game everyone's making these specific point solutions and then at some point you've got to go horizontal, like you're already doing this crazy spend on sales and marketing. And then the only way you can actually continue to grow once you sort of get 100% or, you know, some large majority of the market is you actually have to do like not just a point solution, but things that sort of work together.

Maybe the other point of why the bull case for vertical AI agents could be even bigger than SaaS is that SaaS, you still need it. Operations team or set of people to operate the software in order to get all the workflows to be done. I don't know, approval workflows, or you have to input the data. The argument here is that you will get not only replacing all that set of SaaS software, so that will be like one-to-one mapping, but it's also going to eat all of the, a lot of the payroll. Because if you look a lot of the spend for companies, big chunk is still payroll and software is tiny.

Exactly. They spend way more on employees than they do on software.

So it'll be these smaller companies that are way more efficient that need way less humans to do random data entry or approvals or click the software.

I agree. I think it's very possible that the vertical equivalence will be 10 times as large as the SaaS company that they are disrupting.

I mean, there's two cases. It could be that the vertical point solution could be just big enough and you don't need to do that broad breadth thing, right? That could be a nice scenario.

Should we give some examples? I feel like we've all been working with so many verticals AI agent companies, we've got like news from the front of how it's actually going.

Well, your former head of product, Aaron Cannon, is working on a YC company called Outset that I worked with. And basically they're taking LLMs to the surveys and Qualtrics space. So Qualtrics is almost certainly not really going to build the best of breed large language model with reasoning. And then the funny thing about surveys is, you know, who's it actually for? It's for people who run products, for marketing teams, is for people who are trying to make sense of like, what do our customers actually want and what are surveys like? Guess what? That's language. So and then I feel like these types of businesses actually have to thread this needle because enterprise and SMB software often is sold based on a particular person who is the key decision maker.. And you have to go high enough in the organization so that the people you're selling to are not afraid that their whole, their job and/or their whole team's job is going to go away.

Totally.

That's kind of the move that I seen that a lot of companies that sell need to do, because if you're going to go and sell to the team that's going to get replaced by AI, they're going to sabotage it, man. It just does not work. So I think this is an interesting way that a lot of these are top down and you have to go through at some point, even get the CEO to sign off on it.

A company I'm working with, Momentic, that's essentially an AI agent, but for at least where they're starting, it's QA testing. They're actually getting really great traction right now. And it's interesting because do you remember a decade ago, Y Combinator, we worked with Rainforest QA. Rainforest was a QA as a service company, and they had this exact tension of where they couldn't actually replace your QA team. And so they needed to build software that made the QA team more efficient, but that obviously meant trying to replace as many of them as possible, but they couldn't replace the whole team. And so they were always on this sort of tightrope between trying to sell the software to the head of engineering and say, this will mean you'll need less QA people. And great, but then you also have to go sell that to the QA team who don't want to be replaced. And so I think that was always a friction for that business, for how it could scale and grow. But now, like, Momentic with AI can actually just replace the QA people. So their pitch is not, oh, this makes your QA people faster. It's like, this just means you don't need a QA team at all. So they can just focus the sell onto like engineering and engineering doesn't need buy-in from QA at this point. And you can also go in, I mean, to start with, you can go and sell to companies that don't even have big QA teams at the moment. They just use something like Momentic and then it will just like keep scaling with them.

Scaling and they'll just never build a QA team ever.

Yes.

That is a real life case study of what Diana was saying about why these vertical AI agent companies can be 10 times as big as the SaaS companies.

I'm seeing this interesting now, um, like in recruiting too. I had this exact same issue with Triplebyte where to build the software, um, to build software that makes it easy to like screen and hire software engineers, you need buy-in from both the engineering team that they're joining but also the recruiting team. And effectively, the software we were building was trying to replace the recruiters, but we couldn't completely replace the recruiters. But now within YC—

and so the recruiters were always like opposing, yeah, opposing it because it was a threat to them.

Yeah, so it was just always like friction on like how, um, on like how far you can get when the customer you're trying to sell to is worried about being replaced. Um, but yeah, I think now it's still early days, but now with AI you can build things that do the whole stack like of recruiting. We have a company we worked with last batch, like Nico worked with them, aPriori, which is actually just doing like the full like technical screen, the full initial recruiter screen. And getting great traction. So I think as those things keep going, like they won't have the same thing. You won't have the friction of, oh, I need to convince recruiters to use this. You're probably just like not build a recruiting team in the same way that you used to.

I mean, other example is even for dev tool companies, they have to do a lot of developer support. And I work with this company called Kappa.ai that basically built one of the best chatbots that responds to a lot of the, a lot of the technical details that are hard to answer. And I think a lot of the companies that started using them, they actually ended up having DevRel teams that are a lot smaller because it ingests a lot of, uh, the developer documentations, even the YouTube videos that DevTools put up and even a lot of the chat history. So it just keeps getting better and better. And it gives really good answers, actually. It's one of the best I've seen.

Yeah, I also worked with a customer support, like an AI customer support agent company called PowerHelp. Well, actually, we both did last batch, and I learned a couple interesting things from PowerHelp. The first is customer support, like AI agents for customer support was like the category that's like famously crowded where there's like supposedly like, you know, 100 of them. And if you go and you Google like AI customer support agent, you'll get like like 100 results on Google. But what I learned through working with PowerHelp is like, it's actually kind of bullshit. Like, almost all of those companies are doing very simple, like zero-shot LLM prompting that can't actually replace a real customer support team that does a lot of really complicated workflows. It just kind of makes for like a nice demo. To actually replace a customer support team for like an at-scale company that has like 100 customer support reps that do lots of complicated things every day, you need like really complicated software that does all the stuff that like Jake Heller was talking about. And there were only like 3 or 4 companies that were even attempting to do that. And cumulatively, they had like less than 1% market penetration. And so the market was just completely open.

I could also see that being another case of hyper-specialization or hyper-verticalization.

Yeah, like there's not going to be— I mean, maybe eventually there could be a single general-purpose customer support agent software company, but we're like in inning, you know, that'll be like an 8th or 9th inning kind of thing. And we're literally in the first inning. So Instead, you're going to have companies like GigaML that, you know, it's doing it for Zepto, doing 30,000 tickets every single day and replacing a team of 1,000 people. But it's very specific and it has, you know, it's not a general purpose demo where kind of thing, like it's 10,000 test cases and a very detailed eval set that, you know, is basically just for Zepto and things like Zepto.

Yeah.

But if you are, you know, any of the other marketplace companies, you're probably going to use it because like that's a very well-defined kind of marketplace that's, you know, instant delivery marketplace.

I think this is the kind of dynamic that led there to be like $300 billion SaaS companies rather than like one like $10 trillion like meta SaaS thing that provides all the software for the world. It's just like the customers just require really heavily like tailored solutions and it's hard to build one that like works for everyone.

Exactly.

I mean, we already gave 3 examples of customer support, but they're very different verticals. It's like dev tool companies, very different kind of support that you need and the training set to marketplaces, very different, right?

Yeah. I guess whether you have agents or real human beings working for you, you end up with the same problem, which is every company bumps up against Coase's theory of the firm, which says that any given firm will grow only so much. To the point where it becomes inefficient to be larger than that. And then that's why there are sort of networks and ecosystems and, you know, a full-blown economy, you know, like every firm will sort of specialize to do what it is particularly good at. And then the limits, the outer limits of what those firms can be, it's actually based on your ability as a manager. So yeah, that, that part a little bit breaks my brain because when we spend time with Parker Conrad at Rippling, one of his favorite points is actually, well, everyone's very obsessed with the fact that the rocks can talk and maybe they can draw. But the more interesting thing for him, running HR IT software that he spends a lot of time thinking about HR, like actually the coolest thing about the LLMs is that the rocks can read. And from his perspective, He's— I think he has 3,000 employees. He still runs payroll for all 3,000 employees through Rippling. So I think he spends a lot of time thinking about like, how can one person extend their ability as a manager? And I think we're going to see a lot more there. That would be a reverse argument that if we're at this moment where tools for managers and CEOs are going to get much more powerful Oh, it could increase the scale of the firm that you can run.

Right. And that's certainly what Rippling is trying to do.

Yeah, that's sort of the war, right?

He's attempting to build this like suite of HR tools where if he wins, he's going to eat a whole bunch of billion-dollar SaaS companies in like one giant company.

It's a very interesting point, Gary. I think what made me think about this is that with having all these AI SaaS tools, it's going to give the ability to all these leaders and all these orgs to basically open the aperture of the context window of how much information they can parse, because there's a limit of how much as humans we can have meaningful relationship. There's like the whole thing with the Dunbar number. It's about 300 people that you can—

150.

150 that you can have a meaningful relationship with. But with AI, because all of these rocks now can read, I think we will be able to extend that Dunbar limit that we have.

Yeah, I think Flo Crivello had this interesting post on Twitter that went viral around. I think someone had made a voice chat, like just weekend project as a CEO, but it would call all 1,500 of their employees. Yeah. And, you know, it was, you know, very short call, like kind of sounded like it was from the CEO just asking kind of personally. I mean, it sort of reminds me of that scene in Her where it zooms out and actually, you know, you're following the experience of one person using the Her OS. But actually that her OS is actually speaking to 15,000 or tens of thousands of people all at one time.

How many others?

8,316.

Yeah, I mean, large language models can talk and can have conversations. And then to what extent can this power actually extend the capability of one or a few people to understand what's going on?

I heard about that use case. It definitely got me thinking because as I understood it, the product is something like it just It will call up all your employees and then your employees can just ramble about what they've been doing and it will just extract the meaning out of it and give the CEO a bullet point summary of here's the most important stuff.

And there were a bunch of SaaS companies that attempted to do these weekly pulses from employees using traditional SaaS software. But that version is literally 100 times better than the pre-LLM version of this idea.

But I wonder with that particular tool, Just like, it's going beyond just reading and summarizing. This is the argument of if writing is thinking, then there's actually just a huge amount of work that's involved in the effort of figuring out who's an effective communicator and what are the most important things to be— what are the key things to be focused on as a company? And I just wonder if that at some point do the LLMs do— they go beyond just summarizing and reading and doing actual thinking, at which point who's actually running the the organization. That's an interesting thought.

I guess the other thing that's kind of interesting about how Parker Conrad's thinking about it is, um, I found out about this recently off an interview with Matt McGinnis, his COO, that there are more than 100 founders who work at Rippling now as sort of specific people who run like an entire SaaS vertical inside Rippling.

It's super cool the way he's built the team. Hard probably knows heard about it because you've done a bunch of interviews.

Yeah, I mean, it's definitely very focused on recruiting founders. And I mean, Parker, like, Rippling is essentially the case against vertical, like, or verticalization.

He's actively trying to horizontalize and take over all of HR and IT software.

Yeah.

Like, the whole thesis is basically there's this underlying platform that has, like, lots of value, and he wants to recruit founders and teams that build on top of the platform. Like it's almost a little bit more sort of like Amazon-esque, whereas like shared infrastructure.

Um, yeah, I think every product that they've released, I mean, things like time tracking and whatnot, I mean, basically they launch a thing and it hits like multi-millions of dollars in ARR on day one of launching. And that's exactly what we were talking about earlier. Like once you, once you have a vertical, once you have a toehold, what you're saying is, well, I have to spend this money on sales and marketing anyway, can I, you know, basically get higher LTV and hold my CAC constant? And that's sort of what you— if you look at all the top software companies today, it's like that's what Oracle is, that's what Microsoft is, that's what Salesforce is. Rippling, knock on wood, going to be the next. But it's, it's an interesting alternative to going from 0 to 1 totally on your own.

Do you guys want to talk about some of the voice companies that we have? I think that's like an interesting like subcategory of this, of this stuff that's like really blowing up now.

I have a company that I work with called Salient that basically does AI voice calling to automate a lot of debt collection in the auto lending space, which traditionally—

So they like call up people and they're like, hey, you owe $1,000 on your car.

Yeah.

Which actually—

What's up with that? Actually, this kind of job is one of those butter passing jobs. It kind of sucks because a lot of, these low-wage workers work in all these call centers and it's like a terrible, boring job. So very high churn and giant headcount to run these because there's just so many accounts with these banks that have to do that. And this is a perfect task that AI could automate. And what Salient has done is has been able to actually get very, very accurate and it has been going live with a lot of big banks, which is super exciting. And this was a company from last year. And demonstrating that, that part of it that they were able to get in because they sold through top down.

I guess the space feels like it's moving very quickly and that we have incredible companies that are voice infra companies like VAPI, and then people can sort of get started right away. And retail also. I mean, these are companies that have reached pretty fast scale just because it's one of the more exciting, like mind-blowing things that you can get up and running within I mean, literally the course of hours. And then some of the question that remains unanswered, and we hope they figure it out, is how do you hold onto them, especially as you run into things like the new OpenAI voice APIs? Do you go direct? It's probably way more work to try to use the underlying APIs off the bat, but these platforms are clearly low bar. And then the question is, can you keep raising the ceiling so that you can hold onto customers forever.

Harsh, you were making an interesting point earlier about how the apps that people have built on top of LLMs has changed from early 2023 when it started until now.

Voice, which we were just talking about, is a great example of this. I think even if you went 6 months back, it felt like the voices were not realistic enough yet. The latency was too high. It felt like we were probably a ways off having AI voice apps that could meaningfully replace like humans calling people up and here we are. And yeah, I was just zooming out, thinking back to the first YC batch where LLM-powered apps first came in. It was probably winter 2023, almost 2 years ago now. And the apps were essentially just things that spat out some text and not even like perfect text.

The rocks could talk. That's about it.

Yeah, right.

Sort of more like copy editing, marketing edit, email edits, it was just kind of more just incremental.

Yeah. I had a company, I mean, the one that sticks in my head is a company called Speedy Brand. And what they did is make it very easy for a small business to just generate a blog and spit out content marketing. It was a very obvious idea and it wasn't perfect, but it was pretty cool at the time. And that's what we've talked about a bunch of the show, but that was like the ChatGPT wrapper turned out around that time. So, hey, this is what an LLM app looks like. It's just a ChatGPT wrapper. It does very basic, spits out some text. It's going to get crushed by OpenAI in the next release.

And it did.

Yeah.

Well, I don't know if that one did, but that first wave of LLM apps mostly did get crushed by the next wave of GPT.

But I feel like we've had this sort of boiling of the frog effect where from our perspective, it's sort of like every 3 months things have just kept getting progressively better. And now we're at this point where we're talking about full-on vertical AI agents that are going to replace entire teams and functions and enterprises. And just that progression is still mind-blowing to me. 2 years in, which is still relatively early, and the rate of progress is just unlike anything we've seen before.

And I think what's interesting to see is, we discussed this in the last episode, is a lot of the foundation models are kind of coming head to As you said, there used to be only one player in town with OpenAI, but we've been seeing in the last batch this has been changing. Claude is a huge contender.

Thank God. It's like competition is, you know, the soil for a very fertile marketplace ecosystem for which consumers will have choice and founders have a shot. And that's the world I want to live in. So people are watching and thinking about starting a startup or maybe have already started. And they're hearing all of this, how do you know what the right vertical is for you?

You gotta find some boring, repetitive admin work somewhere. And that seems to be just like the common thread across all of this stuff is if you can find a boring, repetitive admin task, there is likely going to be a billion-dollar AI agent startup if you keep digging deep enough into But it sounds like you should go after something that you directly have some sort of experience or relationship to. That is a common— there's definitely a common thread I've seen in the companies that I'm seeing promise with. And another one just pops into my head, Sweet Spot. I think I mentioned on this before, they're basically building an AI agent to bid on government contracts. And the way they found that idea, and this is a year ago, was they just had a friend whose full-time job was to sit there on a government website, refreshing the page, looking for new proposals to bid on., and they were pivoting. They're like, oh, that seems like something an LLM could do. A company from a recent batch which pivoted into a new idea that's getting great traction, they're basically building an AI agent to do process medical billing for dental clinics. And the way they found the idea was one of the founders' mothers is a dentist, and so he just decided to go to work with her for a day and just sit there seeing what she did. And she's like, oh, all of that processing claims seems really boring. An LLM should totally be able to do that. And he just started writing software for like his mother's dental clinic.

So I guess, I mean, in robotics, the classic maxim is, uh, you know, the robots that are going to be profitable and that are going to work are going to be, um, dirty and dangerous jobs. And in this case, for vertical SaaS, look for boring butter-passing jobs. Well, with that, we're out of time for today. We'll catch you on the Light Cone next time.