--:-- Joe Schmidt: As OpenAI and Anthropic move deeper into applications, what's left for startups to build? A16Z partner Joe Schmidt and HubSpot co-founder and CTO Dharmesh Shah explore how founders can avoid competing directly with Frontier Labs and build lasting value through workflows, customer context, and data. Drawing on HubSpot's evolution and AI experiments, Dharmesh explains why founders should start narrow, learn quickly, and look for markets that cheaper software creation makes possible.
--:-- Welcome back to the A16Z podcast. I'm Joe Schmidt, and today we're joined by Dharmesh Shah, the co-founder and CTO of HubSpot. We're picking up on a conversation that started with an exchange on X about my Yellow Brick Road essay, the idea that there's plenty of room to build software outside the path the frontier AI labs are racing down.
--:-- Since then, HubSpot's been busy. They've launched Breeze, a CRM for one-person businesses, and they've been rebuilding their own platform around agents. Today, we're going to talk about what's worth building, what an established software company has to change, and what happens when AI becomes the way customers use your product. Dharmesh, thanks for joining us.
--:-- Dharmesh Shah: Thanks for having me.
--:-- Joe Schmidt: Yeah, so excited. It's Dharmesh, you know, LeBron, Madonna. You've got the one-word name, so pretty exciting. This is a fun conversation for me because I wrote that essay a few months ago, really in response to this nihilistic, nothing-matters feeling that was going on in Silicon Valley. And I feel like it's actually still going on. So it's got to be strange running a big public company, a successful public company, through this time period. What does it feel like? I'd love to maybe start just with a market check.
--:-- Dharmesh Shah: On the one hand, it's the old best of times, worst of times. It's the best of times because we have all this energy and excitement that you have when you have this kind of massive transformation. HubSpot's 20 years old now. When the company started, the genesis was, "Oh, there's this new thing called the internet that SMBs should be benefiting from to drive growth." And they weren't yet because it was overwhelming and the tools were fragmented. That's where HubSpot started.
--:-- Fast forward 20 years, there's this new thing called AI that we think small and medium-sized businesses should be benefiting from, but it still feels intimidating. So it feels like a new, revitalized mission—same customer base, just new technology. On the other hand, it does create a lot of the fog of war and uncertainty. For the most part, as a team, we stay focused on the customers. The markets will do what the markets will do. We have our own thesis on how things are evolving and what bets we're putting down. But I focus on the positive side. I've been building commercial software for 30-plus years now, literally every day, and this is the most fun I've ever had in terms of building things. What a time to be a builder.
--:-- Joe Schmidt: No, I totally agree. And it's so strange that the perception is negative when literally everything is working. It's like you fall out of the boat and you hit water right now. Maybe if we go back to some of those formation experiences, obviously with Pyramid and then HubSpot and how you got there. I'm sure back then people were saying similar things: "Hey, it's gotten so much easier to build software, what are you even building now?" I'd love to hear your commentary on the similarities and differences in how you're actually building companies and what you're seeing today with new startups.
--:-- Dharmesh Shah: I'll give you a slightly geeky response because I'm a geeky guy. Having been in the software industry since prehistoric ages—150 years ago—the story of software development has always been about increasing levels of abstraction. When I started, we still had machine language, running on mainframes. Then we had 3GLs, C and C++, and then 4GLs. What we've done is ask: can we get to whatever manifestation you have in your head of whatever problem you're trying to solve, and make that easier for you to do?
--:-- Back then it was really, really hard. Then it got progressively easier. We've had some nonlinear jumps in the level of abstraction. But at its base, this is what excites me about AI. It gives us a new toolset to work at yet higher levels of abstraction. That's the thing that's changed. The problems are still the same—humanity and businesses are dealing with the same issues—but now we have a better set of tools. And I have a software-oriented bias; that's my thing.
--:-- In some ways, it's the same. And we'll go off on a tangent here because one thing that's happened recently is the JEV model released by the TypeSafe folks. What that really takes us back to is: "Oh, it really is just software." Here's another dimension of development that's not a transformer, but it's opening up. We're just putting more tools in the tool chest of builders and expanding the scope of who can be a builder, which is awesome too. A lot of it is still the same. We're dealing with similar issues, and it's nice to be able to work at a higher level of abstraction.
--:-- One reason I think this is a great time—it's always a good time to do a startup, but it's a great time—is that because we're working at such high levels of abstraction, AI lets you iterate so much faster. You can learn lessons that might have taken you six months, a year, or two years. Now it's like: just try it. Build something, get it out there, test it with actual users and customers, and see what happens. That just wasn't possible before.
--:-- Joe Schmidt: Oh my gosh, I totally agree. And I wish you could see all the things we've been building here at our venture capital firm. Even being on the front lines is so much more fun. This leads into a question I wanted to ask you. Historically, especially with the businesses you've built, you would start in a very narrow product surface for a very narrow type of customer. How have you observed this changing with that fast iterative process and the ability to build so much more, create so much more software so much faster?
--:-- Dharmesh Shah: I've been in the startup world directly and indirectly for a long time. I still think of myself as a startup guy. Picking ideas has always been one of the hardest things, especially for first-time founders who haven't been through it. Even after 30 years, I think I'm marginally better at it than I used to be.
--:-- One common mistake entrepreneurs make in idea selection is thinking it's a matter of rigorous analysis—that we're going to have an 18-column spreadsheet that spits out the best possible idea. The number of times I've seen the U.S. Census file... "Oh, there's 18,000 law firms between two and five employees." This is not relevant.
--:-- What I've experienced—and it's not just a data point of one, even talking to other founders—is that it's almost impossible to know a priori what the quality of an idea is, its probability of success, or what the outcome will look like. The only real way to test an idea is to actually test it. Even if you're successful, the thing you end up with that succeeds will oftentimes not resemble the original idea you started with. There might be some similarity—the gene pool of the original concepts might still be there—but it's different.
--:-- The byproduct of that thinking is: since you're unlikely to know whether a given idea is going to be good or not, the smart move is to get the idea out there as quickly as possible and learn as quickly as possible whether this is a good idea or whether you should pursue something else. I think that still holds today.
--:-- Going back to the AI point, we can iterate faster now. The one thing that's still hard—and I'd argue maybe even harder now than it was—is distribution. Because it's so much easier to build software now, many more people will be doing it. There will be many more options, and it's going to be hard to cut through the noise and get into contact with a customer.
--:-- I'm a big believer in starting out by narrowing the aperture in the early days. Focus in, because that increases your odds of getting through and learning. Mistake number two that entrepreneurs make is feeling like they need to start with this massive TAM: "This can be a $100 billion business." That's great—have the ambition and vision that this could be $100 billion. But tell me what you're going to do in the first 30 days. What's the thing you're actually going to focus on? $1 revenue first. Let's get to $100 billion.
--:-- Even now for me, as I explore new ideas within HubSpot—both for HubSpot Next, our innovation arm, and new ideas within the core platform—it's about narrowing that field of view. I think the people who win are the ones with the shortest iteration time and the cheapest learning time. If you can shorten that loop and learn less expensively, over the fullness of time, you end up winning on average.
--:-- Joe Schmidt: I still think this moment is the best possible moment for high-agency people. The reality is, yes, it's way easier to build software, but there's only so many people who are sitting alone at 10:45 PM when we exchange an email on Saturday night, hacking on their computer, just being that person and building something cool.
--:-- I want to go back to the beginning of HubSpot. You started with—and correct me if I'm wrong—a legal product. Maybe there was a legal product in there somewhere. And somehow that has now become this CRM for small and mid-sized businesses. It's unbelievable what you've expanded that wedge into. I'd love to hear how you thought about that evolution into the SMB marketing wedge and how you expanded from there to platform.
--:-- Dharmesh Shah: When we first started, we were in business school. That's where my co-founder and I met, here on the East Coast. The thing we alighted upon was the beauty of the SMB as a market. It was underserved.
--:-- I'll give you the quick story because it's interesting. We had both been in software for a long time—on the marketing and sales side and the product development engineering side. We had both done enterprise software before, so we knew that. The challenge with enterprise software is it's completely non-fun as a startup entrepreneur. Sales cycles are long, you get revenue concentration, you've got a small set of customers impacting your product roadmap and your vision.
--:-- On the other end of the spectrum, you have consumer markets. The right consumer companies in the software space have very bimodal outcomes. If you're successful, it's a massive outcome. A lot of times you're doing advertising subsidies or figuring out some other way. You can't really charge for the product because "don't be a chump."
--:-- The beauty of SMB, from a pure market analysis perspective—this is two guys in business school—was that it has the beauty of enterprise in that you can actually charge money for a problem you're solving. There's nothing wrong with charging money. But it has the scale of consumer—you're not going to have revenue concentration. You can have tens of thousands, hundreds of thousands, millions of customers. So you have the best of both worlds.
--:-- That was our original thinking. Our original idea that we took to the MIT 50K competition was ERP software—an Oracle for small business. Going back to narrowing the aperture, we thought: "Okay, that seems interesting. We're going through this transition from client-server to the web. Someone should create an Oracle for small business—everything integrated, working together, built for SMB."
--:-- When we went to the business plan competition, we said that's too big. Let's narrow the aperture and call it LegalSpot. The HubSpot company name already existed, but the thing we wrote the business plan around—purely as a way to test the market early on—was to focus on ERP for law firms. Back to your original point, I was looking at all the U.S. Census Bureau data: "Oh, there's lots of them out there. We can multiply X by Y." It makes for great business school spreadsheets and PowerPoint decks.
--:-- We did that, and we have a propensity for testing the waters. One thing we learned early on, while still in business school noodling on the HubSpot idea, was that it's not a fun market to sell into. As we talked to law firms about ERP and what the solution could be, the marketing and the website part was what was most interesting to them.
--:-- Joe Schmidt: Like, "Wait, hold on, can you help me find more customers?"
--:-- Dharmesh Shah: Exactly. We had time tracking and all that. They said, "That doesn't really matter. Can you help me find more clients that are the ideal fit?" So we were down to those two ideas: do we stick with ERP for SMB, or do we do marketing for SMB? We chose—we almost could have flipped a coin. We didn't. We were a little more deliberate. We extrapolated from a minor set of data points, but life is short. As it turns out, lawyers are the second hardest market to sell into, software developers being the hardest. So now we're navigating the SMB marketing product.
--:-- Joe Schmidt: Of course that then morphs into more of a platform and a full CRM. Was that thoughtful from day zero? The reason I'm asking is I think there are a lot of folks out there using some new AI medium—whether it's voice products, which I still think is a huge opportunity, or some sort of support chat interface. Everyone always wants to end up being "the HubSpot for whatever use case." Talk a little bit about how you navigated that product journey and created the right loops to get to more of a platform from this initial wedge.
--:-- Dharmesh Shah: We started off as a suite of marketing applications. To answer your first question: no, at time T equals zero, we did not think we were going to be a CRM platform someday. That was not in any of our PowerPoint decks. But as the company evolved—and this may be a little idiosyncratic to HubSpot, but I don't think so—you have the need for product number two, the second act. We hit that.
--:-- A couple of things happened simultaneously. We were in the marketing software business, and being in marketing software was very, very close to sales and CRM. That was the adjacent category, so it made logical sense. The other thing we learned once we had a bunch of customers: we did a survey of our customers, maybe a couple of years before the IPO, so five or six years into the company, just to see which CRMs they were using. We discovered that 20% were using Salesforce.com. Another 20% were using some other CRM. But 60% were not using a CRM at all. That was insane to us. You're using HubSpot software, you're paying us money—these are not random companies. These are people paying us to generate leads. What's happening to those leads?
--:-- The thing we alighted upon was: these companies should be using a CRM. My co-founder and I are very weirdly simple-minded. We thought, "Why are they not using a CRM?" The category exists. There are great companies with great products out there. We came down to: it's overwhelming. They're intimidated by the idea of a CRM. It's too expensive. They think it's going to cost too much to get going.
--:-- We made the decision that it was a nice adjacent thing. We needed a second product anyway. There was strategic value in saying that over the fullness of time, the existing CRM players are likely going to get into the marketing space. Even if we want to stay in the marketing software category, it's unlikely we'll be able to do that forever. All those things converged around the same time period.
--:-- When we launched the CRM, we addressed both issues. For the 60% who hadn't picked a CRM yet, we asked why. "Too hard." Okay, let's make the easiest CRM that's ever been conceived. We're focused on SMB, so that was a core strength anyway. And we took "it's too expensive" off the table by making it free. Not freemium—free. That's been the philosophy at HubSpot: create customer value before you try to extract it. If you can get out there with something useful, we can subsidize it with the marketing software business we already had.
--:-- As we iterated, we learned that to win in the CRM space, it needed to be a platform. CRM is almost by definition a platform. That was the journey—going from product one to product two, expanding into a strategic adjacent thing.
--:-- The other thing that carries over to entrepreneurs listening: do what you can to solve the whole customer problem. However you define it, whatever circle you draw around it, what's the actual problem you're trying to solve? Even when HubSpot started, we had a very broad base of marketing applications, which goes against every piece of advice I've given: "Do one thing really, really well." We did all the things. We had content management, SEO, landing pages, forms. There were great companies with great products in each of those categories. The reason we chose to do it all is that it was the only way to solve the actual problem customers had: they were intimidated by fragmentation. Everything's all right. We had to do an all-in-one offering. We carried that idea with us for 20 years. We still do the same thing now. What's the actual problem for us to solve?
--:-- Joe Schmidt: With hindsight, everyone always talks about the brownfield approach—going after people on every core system and ripping and replacing—or the greenfield approach—winning new businesses who have never adopted software. The standard thinking on HubSpot is that it was a very thoughtful decision from the beginning to go after the greenfield. The reality is: all of our customers just didn't have it, and we built it. And it's still going to be fine if you do it in that order. You can still be a massive company if you nail it.
--:-- How did you think about building the right ecosystem around that initial offering? That's something HubSpot has done such an amazing job with. What are the key learnings from the early days of building the ecosystem? And now that it's such a different world, with AI impacting the ecosystem, I'd love to hear you talk about that.
--:-- Dharmesh Shah: I've been a consumer of platforms and ecosystems for most of my professional career, so I know what it's like on both sides. This is no disparagement to any existing platforms, then or now, but what tends to happen with ecosystems is that platform providers start looking to attract the developer ecosystem—get people to build apps or companies adding value to the core platform. But over time, either intentionally or unintentionally, they stomp over those developers. They close things up, shut things down.
--:-- The story is 15-plus years old now. Twitter had this in the early days. The original Twitter was a very open ecosystem. "We're a platform." A thousand flowers bloomed. Lots of Twitter clients were built. Lots of people built products and companies on that original set of APIs. It's not exclusive to Twitter—that's just what jumps to mind. But then they closed it down and became everything from developer-unfriendly to almost developer-hostile.
--:-- One thing that's helped us build the ecosystem is that we've taken those lessons to heart. We said, "Let's be the platform we wished would have existed for us back when we were doing various and sundry things." Let's carry that HubSpot brand of being friendly, likable. We're red-blooded capitalists, but we like to think of ourselves as warm-hearted red-blooded capitalists. Everyone's here to make money.
--:-- We try to be, even with our partners, as open a book as we can be. "Here are things we can't promise you we're not going to do, but here's what we're doing. If we were looking for white space, if we were you, here are the things we would do that we think there's an opportunity—things we're highly, highly unlikely to traverse into." That's work. We've built trust across a wide pool of people.
--:-- Building an ecosystem is a lot like building a product. You have to think of it that way. As a developer at heart, in the same way we think about time to value for core customers, we have to think about time to joy for a developer. How fast can someone get from "Oh, I got something working with the HubSpot APIs" to "Oh, I made my first dollar working within the HubSpot ecosystem"? Can we compress that? Can we reduce the time to joy for that class of customer?
--:-- Joe Schmidt: No, totally. And congrats on building such a thriving ecosystem. Maybe transitioning a little bit into the essay and the conversation we were having around it. I'd love to hear how you think about the things you're maybe concerned about with regard to the workflows the labs are going after. There are many things inside the HubSpot experience that those horizontal products are trying to pull in. How do you think about the area you want to defend and where you can build the most durable value?
--:-- Dharmesh Shah: A few thoughts. I'm going to open this with a very, very biased point of view, given I'm the co-founder and CTO of HubSpot. That doesn't necessarily make me wrong.
--:-- One thought, speaking broadly before we zero in on specifics, is that for those on the outside looking in, it feels like—and this is your Yellow Brick Road, which I love the essay—there's this well-trodden, shiny, bright road pointed to the Emerald City. How could you not walk down it? It feels empty at the time. But you don't know there's a fleet of tanks coming down that road, and they're getting faster. It's not that they're even slow, because they've invented this new thing that makes them go fast.
--:-- That was the premise of your essay, which I wholeheartedly agree with. Yes, they're going to barrel down that road. It is a shiny, nice road, hence the trillion-dollar valuations and soon-to-be market caps. But there's so much opportunity outside of that.
--:-- Here's a framework. Here's a mistake I think entrepreneurs make on both sides of it. When they dismiss it, they say, "Oh, well, ChatGPT doesn't do this, or Claude doesn't do that. We're going to fill this gap." Those are the ideas that make me nervous. There are two things misguided about that thinking.
--:-- One: if you can just append the word "yet" to it and it still holds water, that's a problem. "Oh, ChatGPT can't do that yet." Does that ring true? If it does, that's a problem.
--:-- Thing number two: what I think of as measuring the product versus the company. "Oh, ChatGPT can't do that." Okay, but can OpenAI, the company, do that? Maybe it's a new product, a new product line. You have to think through how wide that road is.
--:-- Joe Schmidt: You know, like, okay, maybe the road itself is getting wider, fine. So maybe stay off the road, but maybe also stay a little bit outside. It's like, okay, if all things being equal, you can just be a little bit further away. So even as they expand, and it's somewhat predictable. So that's kind of thing number one: make sure you either pick things that the frontier model companies have a structural constraint that they just can't do—regulatory-wise, whatever it is—or have what I think of as an incentive constraint. Yes, they could do that, but the odds of them doing that are really, really low.
--:-- Joe Schmidt: I'll use HubSpot as an example. Let's say any of the frontier model companies—I'll pick on Sam Altman since I'm an investor in OpenAI and I know Sam. He could wake up tomorrow and say, "You know, that CRM industry looks awfully tempting. And that three-plus billion dollars of revenue that HubSpot has? We're going to make it our mission, or somebody within the company's mission, to go take that revenue." Are they smart enough? Yes. Do they have ambition? Yes. But it's that very ambition that makes it highly unlikely that the team at OpenAI or Sam or anyone—same thing with Anthropic or any of the frontier model companies—they're unlikely to do that. But that's what it would take to be able to.
--:-- Joe Schmidt: So my thesis, and this kind of winds back to where do you find the pockets of opportunity that are off the Yellow Brick Road, is this: one thing we know for sure is the frontier model companies have invested hundreds of billions of dollars building, doing the pre-training. Their whole business effectively now is taking the capex they've spent and the money they're continuing to spend and saying, "How do we sell inference? How do we sell tokens? How do we monetize the thing that we've built?"
--:-- Dharmesh Shah: Yep. Awesome. A tale as old as time.
--:-- Joe Schmidt: And they have two choices. One is they can say, "We're just going to go directly to everybody on the planet and sell them tokens." They'll do that by virtue of the consumer product. And even on the high end, they'll say, "We're going to go directly to Fortune 500, JP Morgan, and we're going to sell them tokens because there doesn't need to be an intermediary. We'll have an army of FDEs that can help us do that." That's certainly a path, and they're walking down it.
--:-- Joe Schmidt: Another possible path is that there are SaaS companies and products and startups that know their particular domain, know their market, have some distribution, or have something that allows them to sell tokens and be a reseller of inference, for lack of a better term. This is not a binary choice. They can do both. They can say, "For these kinds of things—for a consumer—we want to own that. We don't want other people doing a ChatGPT equivalent. We're going to own that. But there's going to be lots of opportunity to say, 'Okay, think of AI as infrastructure in the same way we thought of AWS and cloud computing infrastructure 15 years ago.'" It's the exact same thing.
--:-- Joe Schmidt: As entrepreneurs—once again, I'm biased but not wrong—I think it's less scary now to build in and around the AI arena than it was in 2023. The reason I say that is back in 2023, if I were a betting person, I would have said they are just so far ahead. It was like they had figured out this magic elixir. I had no idea if anyone was ever going to be able to really solve it and pull off all the things, including the capex spend that was going to be necessary. I was thankfully wrong about that. Thankfully, from an ecosystem perspective, because where things stand now is that models—I'm not going to say they're a commodity—but part of the virtue of them being as common as they are is they are infrastructure. They're going to trend toward both a pricing perspective and a fungibility perspective.
--:-- Joe Schmidt: Right now, things like OpenRouter exist because it's easy to swap out models. Even within HubSpot, most SaaS companies are doing some form of use case decomposition and model routing to figure out what the right model is for this task.
--:-- Dharmesh Shah: The byproduct of that is that you can sort of swap out what you need. So there's less reason to be scared because you don't have this supply chain power where one company controls the one crucial ingredient for you to build a product and solve the problems you're solving. We don't have that right now. We have open source. We have lots of great commercial products. So as long as you stay off the Yellow Brick Road, it is completely possible now that you're not going to get margined out because you have options. There's not just one company providing the one key ingredient to building an AI-native company.
--:-- Joe Schmidt: No, totally. And I think the other piece of this, which you kind of touch on, is that of course, with enough focus, but it's also just that the advances or the advantages you're going to build as a startup or as a HubSpot or whoever it is—if you're not building a foundation model, you're actually building high-quality software. I think that is the layer you're going after to provide value to your customer. There are all these other great pieces of infrastructure that you can now package up and give to those customers. The way I think about it, and I argue in this essay, is there are loops, there are tests, there are things you can do to try to identify these pockets. But then hopefully you're building some sort of increasing return to scale or compounding advantage at this layer. I'd love to hear how you think about this—HubSpot or any of the things you're working on—as it relates to the loops you try to build in these products or the advantages you're trying to build.
--:-- Dharmesh Shah: Yeah, it's compounding loops all the way down, Joe. Of course. That's the entire name of the game, before AI, after AI. But a couple of things I'll say. I'm going to rewind. I pushed one idea on the stack. I'm going to pop it off and talk through it.
--:-- Dharmesh Shah: One of the questions we get back on the "if you were an entrepreneur evaluating ideas" front is, "Okay, well, how about us?" Today was OpenAI's Dev Day, by the way. It's September 30th, right before this call.
--:-- Joe Schmidt: Data recording.
--:-- Dharmesh Shah: What's that?
--:-- Joe Schmidt: I said we're dating the recording for the audience. So you now know. Time stamped.
--:-- Dharmesh Shah: Exactly. You know exactly when this is being recorded. So HubSpot launched—we were the first CRM company to do so—our connector to ChatGPT over a year ago, and we have a similar one for Anthropic. One of the questions we get sometimes is, "Oh, doesn't that sort of diminish the long-term value of HubSpot? If everyone's coming in through ChatGPT or through Claude or something like that, isn't that eroding the long-term value?" I understand the reasoning. My counter to that is that the interface matters, but it matters to varying degrees. There's a spectrum.
--:-- Dharmesh Shah: Let's say on one end of the spectrum—I'm making this up as I go—you look at something like Uber Eats or DoorDash. They have very compelling apps, but the app is trivial. What's actually valuable is the network they have, the partnerships with the restaurants, the logistics so a driver can show up, pick up food, and deliver it. That's the actual value, and that's the part that's hard to displace. There are other things where the interface is the entirety of the product. That's all there is. If it weren't for the interface, there would be no value.
--:-- Dharmesh Shah: I think most business applications actually sit more towards the Uber Eats kind of thing. Not in the sense that they have network effects like Uber Eats and DoorDash do, but in the sense that the actual thing that's valuable is the work that's happening. The interface is a means to an end. The only reason the interface is important to HubSpot is because marketers and salespeople are trying to get something done and deliver on certain outcomes. The interface is the way for them to communicate with a computer. Before AI happened, HubSpot's claim to fame was, "We understand the problem you're trying to do, and we've built the interface that's super simple to help you get there." But that was not the entirety of the value. One can argue where on the spectrum HubSpot falls, but I can guarantee you it does not fall where the interface is 99% of the value.
--:-- Dharmesh Shah: Everything kind of comes back around again. When we saw the original SaaS development moving to the web-based application world, we had the same kind of thing. We had client-server apps running on our desktops back in the day, and then we had this browser thing come along—Internet Explorer and later Mozilla. Now my software is running inside of a browser. But as it turns out, Google did not own the entire application stack. Lots and lots of software companies got built. The interface became the browser because that was a convenient way to get to the actual product you needed. The browser was a better way than installing something on 19 million computers.
--:-- Dharmesh Shah: We have a similar thing now. It's different because the AI is doing cognition work. It's doing actual smart things. It can do more things. But if you just look at ChatGPT and Claude, they're effectively the new OS. They are the new browsers. This is the way people are consuming value. There will be companies like HubSpot that plug in. We saw this from DevDay last year, and we continue to see it. They're going to increase the richness of what's available within ChatGPT. They launched a kind of GPT store, a marketplace for people to plug in with a plugin architecture. That makes complete sense to me.
--:-- Dharmesh Shah: So my fundamental philosophy, which I think holds true for most people looking to build in the B2B world, is: don't overthink the interface. Figure out what's the best possible way to achieve the outcomes your customers are looking for. Then the other value you can bring is around what's the economic best way, what's the lowest latency way, what's the lowest friction way to accomplish those outcomes. An LLM is very, very good at lots of things. It's the most brilliant technology we've had in the history of humankind. But one of the things I've gone on record saying is: anything that can be solved with a structured query, deterministically, solve with a structured query deterministically. Don't use inference to solve something that a query can solve. It doesn't make any sense. If your question is "How many customers or how much revenue are we getting from Ireland?" you shouldn't have to troll through 8 million Slack messages and emails and infer the answer. Yes, the data is probably there somewhere, but that should be a database query. That should not be solved through inference by throwing a 300 IQ model at it.
--:-- Joe Schmidt: No, totally. Absolutely. And I think there are many companies navigating through this as they try to understand their own pricing and consumption ideas, which could be its own podcast, I think. Maybe last question on this, because I think it's really interesting what you're talking through. The architectural decisions you made when you originally built HubSpot are probably—with the benefit of hindsight, I don't know how you would revisit those. I'm so curious to hear how you think about that because, of course, the way you designed it originally was for a human to be navigating all of this information. You're storing that in a way that's intuitive to a human. Now we have these unbelievably powerful models that can go look at things in very different ways. How do you think about the way you may have wanted to start differently, or what you're doing to address that now as an executive at HubSpot?
--:-- Dharmesh Shah: Yeah, a couple of things. I'm a classically trained computer science, software development nerd forever. One of the things I'm a big believer in for large-scale software applications is this notion of frameworks and starting with a set of primitives that you evolve over time. That's the thing that gives you compounding and leverage. Across every company I've done and worked in, if you can get those core primitives right, what you end up doing is remixing and recomposing those primitives in novel ways based on how the market evolves and how the product evolves. That part has been true at HubSpot since the beginning.
--:-- Dharmesh Shah: The thing we built out was this core operating system or operating framework on which we built the marketing automation and things like that. When we went to product two, three, four, most of the value we needed to bring to market already existed in the form of primitives. For instance, we started with marketing automation: "Here's this deterministic workflow to do marketing—send an email out, wait three days, wait for them to click." When we got into sales and CRM, our marketing automation primitive was actually a primitive. We could just reuse all of that technology, and we would have more on day one of our sales automation product than anyone likely had that had been adding for a few years. That part of it, I think, still holds true.
--:-- Dharmesh Shah: What we're doing now is saying, "What is the framework we need for the next 20 years of HubSpot, now knowing that we have this AI world?" You touched on one thing: the data is no longer just this structured data in a relational data store, which made sense for the kinds of deterministic queries you're doing. Now it's both structured and unstructured data. It's not like this data didn't exist within the organization, even within our customers' organizations. You just couldn't do anything with it that was meaningful at any price back 10 years ago. But now you can.
--:-- Dharmesh Shah: So now we revisit the primitives and say, "Okay, we still need a structured data store." I will happily debate this with whoever wants to debate it. There is still room—whether it's relational or something else—for some persistent mechanism for storing known facts and data, some sort of system of record, if you will, alongside this system we think of as a system of context, which is this unstructured data. That's the thing we're building out. The way I think this is going to evolve is this notion of—I use this term, others use it too—a "second brain." Can we put all the things that are happening in some sort of graph structure? Can we store this in some sort of context graph, another term that has been used? That's the idea: can we take this and use it in combination with the structured data store, draw inference, drive agents, drive outcomes, and get work done?
--:-- Dharmesh Shah: One thing that's interesting in terms of AI—you didn't ask this question, but it occurred to me as we were chatting—is that one of the things that's really beautiful about agentic coding now, particularly true for incumbents, is that yes, it allows us to produce software faster. We can test ideas faster, build products faster, which is all awesome. But the other thing you can do is go deeper and do larger projects that you might not have otherwise considered.
--:-- Dharmesh Shah: What's possible now is that we have a system of record for code—GitHub. Imagine you say, "I wish younger me had made these choices when it came to those primitives, or I would have introduced this as a primitive that's a reusable service, because we didn't know that then, but we know that now." In theory, it's actually possible to say, "I want to play that back. I want to take this body of code that we used this data structure for, this language for, this framework for, this architecture for. What would happen if we spent $200,000 on tokens to just redo that?" It's like the infinite possible universes thing. "Let's jump to that branch of the universe where we made a different choice and see what happens. Where do you wind up?"
--:-- Dharmesh Shah: One of the advantages companies like HubSpot have is that we have what people used to call unit tests and now show up as evals. Can we actually measure the viability of the thing we just did? Does it still pass the test? Does it still produce the outcomes? Does it still do all the things it needs to do? We have that now. We run this test when new models come out. We run our battery of evals across new models and say, "Here's a tradeoff. Here's the economic value. We lose about three percentage points of quality, but we gain this much in terms of latency or economics." We can do the same thing with testing these kinds of theories. "If we did rewrite this whole thing in Rust"—I'm making that up—"what would it look like?" You can test it now with small bodies of work. Anyway, it's exciting.
--:-- Joe Schmidt: I think this is such a fascinating topic. I wrote a different essay, I think right around the same time I wrote "The Yellow Brick Road," maybe a little bit before, and I actually used Workday as an example on this front. "Hey, you have to kind of start over to think about being AI-native, and here's all the advantages you have, but here's all the disadvantages." The benefit I think HubSpot has is it has an amazing founder who's still very, very involved in the company, of course, running the business and driving these changes. I love that example of trying to think about these alternative universes you can build. Are there other processes you're trying to create inside of HubSpot, or teams you're trying to create, or destroying bureaucracy? I'm just curious to see what you're doing as a founder right now.
--:-- Dharmesh Shah: Yeah. We have a group within HubSpot called HubSpot Next. The idea is it's effectively an innovation lab. It's a startup within a startup. We'll have any number of these going on at any particular point in time. The motivation—and we truly do treat it like a startup—we did a variation of this when we built the CRM. It was like, "Okay, we're going to build the CRM thing." People said, "Wait, what's that? We're in the marketing software business." We cordoned off the team. They had none of the rules. You didn't have to build it on our stack. You didn't have to use the brand. You didn't have to do anything. Just go solve the problem. We've carried that idea through now.
--:-- Dharmesh Shah: I've championed and led several of these projects over the years. The one I'm working on right now, literally every single night, is a project called YouSpot. It's a one-person CRM. It's a CRM for solopreneurs. That's the market. It's going back to narrowing your aperture of focus. The HubSpot Next projects are pursuing either a new product category, a new product idea, a new market, or testing a new technology. Going back to learning your lesson cheaply: it's much easier for me to test this idea for a second brain—"Maybe a graph model works better, maybe you plug in all the social media and everything coming in"—in YouSpot because it's a much easier, faster-paced way to test it. The product now is exactly 30 days old.
--:-- Joe Schmidt: Congratulations.
--:-- Dharmesh Shah: Yeah. And it's been fascinating learning. The answer to the original question is: can something like that work? The answer is yes, it does work. I think other entrepreneurs should take away from this that the notion of having a second brain—going back to the Yellow Brick Road conversation, avoiding being run over—the thing you shouldn't do is say, "Oh, ChatGPT and Claude are really bad at memory." It's going to get better. Don't solve that problem. Don't solve the horizontal use case. But say, "Within this class of customer, here's a class of problems they have that would be benefited by thinking about this notion of having a second brain. What would the ingestion points be? How do you pull it together? What are the actual agents you would produce or the work you would accomplish?" I think that underlying primitive is actually a powerful one. I think it applies.
--:-- Joe Schmidt: Yeah. And now that you—I think YouSpot is so fascinating. I think it's very valuable for that, particularly for this type of user. It's what they want, of course. How do you then take those learnings and productize them at the mothership, if at all? And what processes do you need to put in place to make that happen?
--:-- Dharmesh Shah: The way we've done it, which we've kind of stumbled over the years and tried various mechanisms, is to pull on that original thread of "this is a startup within a startup." We do a couple of things. One is, if the thing gets to scale—and we've had this happen multiple times before. There was a product I launched a couple of years ago called Agent.ai. It was a marketplace for AI agents back in the day. This was one of my "accidentally competed with Sam Altman" stories. Although it was focused on go-to-market, it was too horizontal. But there were two parts to it: the agent AI marketplace and an agent builder that said, "Can we let mere mortals actually build these mix of deterministic and non-deterministic agents?" What we discovered is that half of the idea was right, but it actually belonged and needed the full HubSpot team behind it. So we did effectively an acquisition. We took that, and now we're going to make Agent Builder a product within HubSpot. That's what it lives at right now.
--:-- Dharmesh Shah: In the YouSpot case, one possible outcome is that we're building the CRM for one-person companies. We got these things right, those things wrong. Maybe we just absorb it into HubSpot and it becomes a product of HubSpot. Maybe we say the product itself still probably needs to be standalone, or maybe the market doesn't work. But the idea of a second brain maybe works, and here are the things and the suggestion points that we think work for SMB and multiplayer team-based mode. We just sort of let the core team decide what they want to pick and choose. It's a little bit like they're looking at an acqui-hire. It's literally like that. "Here's what we can buy." There's a low, low price of—
--:-- Joe Schmidt: It's included.
--:-- Dharmesh Shah: It's included, yeah.
--:-- Joe Schmidt: It's not free, it's free.
--:-- Dharmesh Shah: Exactly. It's free. We try to have an open mind about it. The nice thing about doing it the way we do it is that—you might think, "Why can't you just do it within the core business?"—those degrees of freedom are super important to be able to run the experiments at the rate you want to run them. One of the things I've long said, being a startup guy, is that the issue incumbents have—and I refer to HubSpot as an incumbent, though I still think of it as a startup—
--:-- Joe Schmidt: Don't worry.
--:-- Dharmesh Shah: —is that it's not the absence of something. It's not that startups have something that larger companies don't. The issue is the baggage, the weight that the organization has. If they can find a way to shed that, then you get all the advantages of being larger. You have distribution, you have customer relationships, you have evals, you have 20 years of a code base that you can use as data to inform the next product you build. All this stuff is actually an asset. So if you can shed the liabilities, and HubSpot Next is a way to say, "We're just going to pretend we have none of those liabilities"—which we don't, as a startup—we can just go off and do our thing. But when it comes time to lean into the asset, it's like, "Oh, it would be helpful to have the HubSpot brand here, or can we get some advice on distribution over there?" We can do that.
--:-- Joe Schmidt: Yeah. It's so easy right now, especially with some of the national discourse, to think negatively about AI and the impacts. But the reality is, all the data I've seen, it's so exciting. Growth is unbelievable. I actually think YouSpot is a really good example of this potential growth. When in the history of mankind would it have been a good idea to go build this super individualized CRM for one person? It would have been almost impossible to do that historically. But now we're able to do that. What is that then going to let those one-person companies do?
--:-- Joe Schmidt: Like that is, we're going to see growth unlike I think anything we've ever seen before. And it's just so cool to dream. So maybe I'd love to just hear you talk a little bit about what you're seeing, like these initial use cases, maybe from Uspot or what you're seeing from growth. It's just, it is really amazing time to be alive.
--:-- Dharmesh Shah: Yeah. So what's interesting, I knew this to be the case, but I wasn't sure to what degree. So as I mentioned, one thing that I did differently with Uspot versus prior HubSpot Next projects that I've done is I monetized really early. And part of it is not to cover COGS or something like that. That's one of the lessons I wanted to learn—I want to learn the pricing lessons and some of the things that we want to learn more about monetization, pricing, and what value people ascribe to this. That was part of it.
--:-- The thing I've been really surprised by is the willingness for people to put a credit card down on a consumption-based model. It's like nothing I've launched over the years at HubSpot. This has been, in that regard, the most successful. So there's this signal out there. And part of it, I think, is just driven by sheer numbers.
--:-- Let's bring the U.S. Census data into this. In 2025, we had the most number of new company formations that the United States has ever seen. I think it was like 5.6 million, closing on 6 million, which is great. This year, we're already at—by August, we were already over 4 million. So it's going to be another record year. So it's not like we had this one-time spike and now it's settling back in. That's fact number one that's super interesting.
--:-- Fact number two that's even more interesting is the percentage of those new company formations that are expected to remain one-person companies is the highest it's ever been. And it's increasing. The U.S. Census Bureau actually tracks this data. They put every new company into two buckets. One is that this company is predicted to hire at least one employee within the next two years. And the bucket of companies that are unlikely to hire an employee in the next two years—that percentage is now over 70%. So it's exciting.
--:-- There's this new market. For the entrepreneurs that are out there, it's entirely new—just like SMB was back when everybody sort of pushed back on HubSpot. It was like, "Oh, you can't make money in SMB. It's so hard. No software company has ever, other than Intuit, made a multi-billion dollar business in SMB." But you can. It's big enough. There are going to be painful parts to those kinds of markets. But anyway, I'm excited for that. I'm excited for the economy. I'm excited for all the entrepreneurs that are going to be starting these millions of businesses. And then excited for the entrepreneurs that build things for that market. It's exciting all the way around.
--:-- Joe Schmidt: No, I totally agree with this. We talk about aspects of this all the time. I think one of the areas we talked about has been vertical software in the past. How did you make that work? You made vertical software work with a low ACV or monthly cost, but then you made it up because you had payments, interchange revenue. I just don't know if people are fully appreciating the increase in opportunity size. There's just so much more you can do, and so many other ways you can offer value and create value for your small customers.
--:-- Dharmesh Shah: Yeah, and this is my counter to the "SaaS-pocalypse" idea. The reality is that—and AI, by the way, for the record, I know you know this—it is software. Just so you know. It's not an alien or a new energy source. It's software. Really smart software, but software.
--:-- What has really happened now is this: If you look at the class of problems that software was designed to solve—I'm going to take one minute because it's actually a good story. I only have a handful of good stories. When we were first starting, in the first year of HubSpot, when we were out raising venture capital, we said, "Hey, we're building this marketing software company. Here's what we're going to do." And almost universally, across all the VCs we talked to, they came back and said, "Well, there's not been a marketing software company in the billions."
--:-- So we had two strikes: the SMB strike and the fact we're doing marketing software. Our argument at the time was this: If you look down the P&L of any size business, every line item on the P&L has a hundred-plus billion dollar software company name attached to that line item—except for marketing. It's not like people don't spend. The reason is that marketing historically has been an arts-and-craftsy kind of thing. You have this amount of spend, you spend a little bit on creating the creative, but then you just spend it on ads and billboards.
--:-- Our thesis was that as a result of the internet, the marketing industry is going to look a lot like a problem that software was designed to solve. That's why a software company didn't need to exist five years prior to HubSpot starting, but now someone will build a massive marketing software business. And I make the same argument now: Before, there was a class of problems that software wasn't really equipped to solve. Now there is a massive set of these new problems—problems that existed, new things that we can do with AI that didn't exist before.
--:-- So the opportunity set is way larger. Yes, some software companies will die, no doubt. Just like some software companies did not make the transition into SaaS and cloud-based things. That's the circle of life. But in aggregate, the opportunity is bigger than we've ever seen. If you're anywhere near software, have the agency, and are entrepreneurial, there has never been a better time to build a software company. Go right now. Go do it.
--:-- Joe Schmidt: Yeah, I totally agree with you. This has been unbelievably fun. I'm so grateful for your time. I'm going to end with a quick lightning round for you.
--:-- Dharmesh Shah: Sure.
--:-- Joe Schmidt: I'm not very quick on my feet, so there's only a few. These are softballs. I'll start with a very easy one, I would assume. Dunkin' or Starbucks?
--:-- Dharmesh Shah: Dunkin'.
--:-- Joe Schmidt: Okay. But I don't drink coffee, by the way.
--:-- Dharmesh Shah: Oh, wow. Okay, well then, what do you drink instead? Just only water? Are you a cyborg?
--:-- Joe Schmidt: Just—yeah.
--:-- Dharmesh Shah: Wow, wow.
--:-- Joe Schmidt: I gave up caffeine like years ago, so anyway.
--:-- Dharmesh Shah: Even more impressive, Dharmesh.
--:-- Joe Schmidt: Okay. Instinct or Muse?
--:-- Dharmesh Shah: I'm sorry, what?
--:-- Joe Schmidt: Instinct or Muse—you know, on the new agentic shopping experiences?
--:-- Dharmesh Shah: I'm going to have to say instinct.
--:-- Joe Schmidt: Okay, interesting. If you had to be an animal, which type would you be?
--:-- Dharmesh Shah: I would still be a mammal. Maybe an elephant.
--:-- Joe Schmidt: An elephant. I like that. Okay, good. What's the most Boston thing about HubSpot?
--:-- Dharmesh Shah: Almost everything.
--:-- Joe Schmidt: Good answer.
--:-- Dharmesh Shah: Yeah. We're very pragmatic. Solve for the customer. I'd say we're sort of a hybrid between some of the best of the West Coast valley thinking in terms of ambition and aggression, but some of the pragmatism of the East Coast. Heads down, just work at it, grind.
--:-- Joe Schmidt: I love it. And then finally, what's the best day you've ever had at HubSpot?
--:-- Dharmesh Shah: About a week ago. Things are starting to fall into place, both in my head and it's just like, okay, I can sort of see how this is unraveling. It's hard to explain. This is not something I'm going to be able to say, but yeah.
--:-- Joe Schmidt: Oh, that's so cool. I mean, after however many years, the best day was a week ago?
--:-- Dharmesh Shah: Yeah. And I'll say this more broadly, and I mean this like all things: The best time I've had being a startup guy in the history of HubSpot, which is 20 years old now, is the time I'm having right now. The reason is—and people are like, "Oh, that can't possibly be true"—it's because I have all the benefits of scale. By the way, I have no direct reports at HubSpot. I've never had direct reports. I'm an IC. But I have all the benefits of scale. I get a seat at the grown-ups table. I can make long-term bets. I can do things. But I have no other downsides. I can do HubSpot Next projects like Uspot or whatever. So I have the best of both worlds. And it's so much fun right now.
--:-- Joe Schmidt: Yeah. It's awesome. Well, Dharmesh, you are a generational entrepreneur, a wonderfully kind human, and a generational thinker as well. I'm so grateful for your time today and for your engagement on all these topics. So I look forward to the next time. Thank you.
--:-- Dharmesh Shah: See you, buddy.
--:-- [End of interview. Host transitions to closing remarks.]
--:-- Joe Schmidt: A16Z and subscribe to our Substack at a16z.substack.com. Thanks again for listening, and I'll see you in the next episode.
--:-- Joe Schmidt: As a reminder, the content here is for informational purposes only, should not be taken as legal, business, tax, or investment advice, or be used to evaluate any investment or security, and is not directed at any investors or potential investors in any A16Z fund. Please note that A16Z and its affiliates may also maintain investments in the companies discussed in this podcast. For more details, including a link to our investments, please see a16z.com forward slash disclosures.
