Growing Better In The Age Of The Builder | Dharmesh Shah | UNBOUND 2026

Wow! Greetings, Unbound! Thank you. Thank you. For those of you that don't know me, I'm Dharmesh, and like Yamini and Duncan, I also work at HubSpot. And for those of you that know me, you know that I'm an introvert. Which means two things: I'll be looking at the ground a lot. It helps me think, and because I'm not in my natural habitat here, this is me in my natural habitat on LinkedIn. Thanks to you for those that follow me, and sorry, Yamini, for all the LinkedIn posts. So to kick things off, I thought I'd share a quick story with you. It's a bit embarrassing, but it's just me and. 13,000 of my best friends. So here it goes. Imagine this. It's a cold, crispy evening in Boston, just a stone's throw from where we are right now, if you're really, really good at throwing stones. And it's not just any evening. It's New Year's Eve. It's New Year's Eve 2007. 19 years ago. The iPhone was just 6 months old. HubSpot was 18 months old. And only one of them had a viable business model. So, I'm happily married to my lovely wife, Kirsten. Hey, sweetie. And Brian is happily single. So you might think, surely they have some New Year's Eve plans. And you're right, we did, with each other. And please don't call me Shirley. So we were 2 wild and crazy guys in Boston on New Year's Eve in 2007, 19 years ago. So what were we up to? A nice, quiet, gromantic dinner at Via Amara, an Italian restaurant in Boston. And Brian and I had a gromance going on back then. And a gromance is basically a bromance with a board deck. So that evening, we talked about our plans for growing this tiny little company. And we talked about a lot of things at that founders' dinner. The wine and ideas were flowing. The wine was exceptional. It was a Barolo from Italy. The ideas, less exceptional, but they got better as the Barolo kicked in. HubSpot turned 20 years old this past June. Thank you. Thank you. So it's no longer in the teenage years, so it's not looks-maxing and jaw-maxing and all those things. And reflecting back on those 2 decades, I think one of the reasons it worked is because back then Brian was customer zero and I was builder zero. We built the product for Brian. We built the product for HubSpot to grow ourselves. And so this made the gap between the customer and the builder Tiny. It was a proper romance, so we would talk every day. We would email every night, trying to build build this company. So I've been a builder now for thirty plus years. Now I know that sounds like a long time because it is. It's it's a long time. And back then, being a builder. was reserved for a select few. A lot has changed. Now everyone can be a builder. Even me, you wonder? Yes, especially you. If you're in this room, if you're watching on the livestream— hi, by the way— whether you're a RevOps person, or you're a sales rep, or you're a marketer, or you're a customer support magician, you can be a builder. Now, when I say builder, I don't necessarily mean builder as in you're building commercial apps or vibe coding your own CRM. Please don't do that, friends. Don't let friends vibe code their own CRM. Thank you. Yeah, let's— And but you can build as a GTM builder now. You can build a wide variety of things on the HubSpot platform. You can build automations. You can build custom apps that sit and live right within the HubSpot UI. And now with HubSpot's new agent builder that Duncan talked about, you can build custom agents all on the HubSpot platform. By the way, I have a quota in terms of how many times I promote HubSpot while I'm on stage here as part of my employment contract, something I get to do as founder. So let's dig in. Let's get some kind of basic AI fundamentals down. on your path to becoming a builder. I'm going to try and demystify things that shouldn't be, uh, that mysterious. So we're going to first chat about what I call horsepower. This is the AI engine. It's the AI model. Next, we're going to talk about the harness, and you'll hear that word a lot. The harness is the car. The AI model is the engine. And then finally, we'll talk about— The data. The human person driving the car. So let's dig in. Let's first start talking about horsepower. This is the AI model. A lot has happened in the world of AI models since we were last together about a year ago, and I can paraphrase my sentiment around what's happening with AI models and what's gone on in the past year in three words: What the frontier? There has been so much innovation happening in AI models over the last year. It's mind-boggling. So models have become smarter, like much smarter. So if you think back to November of 2022, that's prehistoric times now, but that's when ChatGPT launched. Then November of last year, not a lot of people talked about it, but it was a big moment. Claude launched and it sparked this revolution in AI models around reasoning. The models got much, much better at reasoning and doing long-term tasks. And then 2 weeks ago, OpenAI launched GPT-6 Astra. It's an unbelievable moment, September 2026, and a better person than me would have resisted the temptation To highlight the un in unbelievable on the screen there. I am not that better person. In my defense, at least I didn't point out that the word inspirational starts with in. The word incredible starts with in. Didn't point that out. Now, GPT-6 Astra is the smartest model ever made as of 2:00 AM last night. Well, actually 5:00 AM last night because we had a fire alarm in the Westin and I had to check one more time. Long story. It's not important here. So, by the way, it's not that I'm checking for late-night models. It's because that's when they drop. It's not because I have a problem. Okay, it's fine. I have a problem. That's not when they drop. They drop at 12:00 PM Pacific time. It's not a big deal. Now, Astra is winning by most of the AI benchmarks. It's great at 3D models. It's great at physics. It's great at math. It's great at most sciences. It's great at coding. It's great at lots and lots of things and beats all the benchmarks in terms of any of the models that have come before. But my personal benchmark for AI models, it's one of the hardest ones. Can it write a decent dad joke? To figure this out, I built an app called dadjoke.ai. Which does exactly what it says on the tin. It generates dad jokes using AI, and for the first time now with GPT-6 Astra, AI can generate dad jokes that are decent—not good enough to make it into this presentation, but decent. Now some of you are wondering: He's the CTO of HubSpot. Why is he building something like a dadjoke.ai? 3 reasons. Number 1, there are rumors that have been confirmed that I'm at least partly human, so I get to have fun too, for some definition of fun. Number 2, dadjoke.ai sits at the intersection of 2 things I love: fully grown dad jokes and great website domains. Dadjoke.ai, great website domain. And I have to point out that that previous slide sits at the intersection of 3 things that I love: dad jokes, domain names, and Venn diagrams. Thank you. Thank you for the— it's the very like Modern Family opening credits vibe kind of thing. I was proud of this one, I'll say. But there's a third reason humor works. Humor works. Levity is leverage. A joke is just an idea with better distribution. All right, so back to the models. And here the innovation has been no joke. We've had a bunch of brilliant frontier models come out over the past year. The AI companies have been leapfrogging each other every few weeks. In this time since I've been up here on stage, 3 new models probably launched. So the horsepower problem is solved. We have the most powerful technology that humans have ever built being upgraded every month or 2. So let's talk about the elephant in the room. I know there's a lot of of anxiety around the pace of AI innovation. It's been all over the news, all over social media, and I'm not going to pretend it away. I think the risks are real. The worries are well-placed. Now, the good news is this past weekend, Dario from Anthropic, Sam Altman from OpenAI, and even Elon Musk, all arch rivals all agreed on Saturday that we need to pace AI development more. We need to invest more in safety. We need to be worried about possible catastrophic outcomes from all this development in AI. So, yes, it's— we're not out of the woods. We're not out of the woods, but at least we're walking in a more positive direction. And the fact that those 3 came together for the first time around something this important, I think is important. So we talked about the elephant in the room. Let's talk about the elephant and not in the room. What do you call an elephant that's not particularly pertinent? Irrelevant. Yep. I almost cut that one, by the way. I almost cut it. But those with young kids, that'll work well. Trust me, I've tried it. I've tested it. All right, so this presentation is about AI, not 5 years, 10 years from now, what you can do with it now. And the first practical thing is something Yamini brought up earlier today. Outcomes are more important than optics. Solve for the actual outcomes, not how you look using the kind of latest and greatest frontier. And lots of people talk about the state of the art. This model just released, and now it can do this, and now it can do that. My personal advice is don't focus and obsess over the state of the art. Focus on the state of the practical. What can we do with the models that already exist that are super powerful? And every time you have a task that you want to use AI for, don't throw the biggest, baddest model at it. Doesn't make sense. Don't use a fire hose to water a houseplant. So I had a couple more, like, on my way up here. Don't use a samurai sword to slice your cheese. Okay. Don't send in a search and rescue party to locate your TV remote. Okay. Don't hire Tom Brady to toss you a beer at the game. All right. So I think I chose wisely with that first one. All right. Good to know. So the model that you pick should match the task that you're doing. And the metric you're trying to move. So this is my first 10-minute to-do. Name your number. Write it down. What's the metric you want to move? Is it your email open rate? Is it your pipeline? Is it the time you spend doing X? There has to be some metric you're trying to move when you're using AI. And every time you pick a metric that you want to move, a baby CFO smiles. You should do it. All right, so we've talked about AI models a lot. That's what provides the horsepower. Let's talk about the engine. That's the harness of the car. Now, without it, if you— all you have is an engine, you're not going very far, right? You're just kind of revving the engine. So what the car does is it provides the wheels, obviously, a steering wheel, maybe a GPS to help you get from point A to point B. So the harness is super important. And although the models get all the headlines, that's what we see. It's like, oh, XYZ company launched this particular model. It's the harness that actually gets the results. That's, that's an important piece of the AI system. So what the heck is an AI harness? Glad you asked. So it's just software. And what it does is it provides the AI model with the context, with tools, like access to a CRM, to skills that it did not get out of the factory. It knows about lots of things. It knows about Newtonian physics and Shakespearean sonnets, but it doesn't know about business. It doesn't know about certain things. It doesn't have those skills. And finally, the AI harness provides some guardrails. What's the model allowed to do and not allowed to do? So you are already using and familiar with harnesses. Very popular ones are ChatGPT, And Claude, both of those are harnesses. You're not interacting directly with the models. You're interacting with the harness. If you're a HubSpot customer, HubSpot Breeze is a domain-specific, purpose-built AI harness. That's what it is. And there are 2 advantages that Breeze has. Number one, it has access to all the models. And so it can figure out, based on the task at hand, you don't have to worry about picking the model. You don't have to worry about, oh, This model just came out. Should I be using it to do X? Breeze takes care of all that for you, so we can find the most efficient path to accomplish a given go-to-market task. And number two is Breeze has built into it, because it's from HubSpot, 20 years of domain expertise. How do you drive outcomes in marketing, sales, and service? So if you haven't used Breeze yet, you should. And if you have, you should use it again, like Duncan said. It's been completely reimagined. Now, the thing to keep in mind is that models depreciate. They go down in value, not because they get worse. It's because better models come out. And as the new models come out, the older models are less valuable. On the other hand, context appreciates. It goes up in value over time. Because let's say you did the work and you have all your contacts sitting in HubSpot. And it's there, and now a new model comes out. Well, the value of that context is even higher now with a higher IQ model. You're going to be able to get more value out of that context. So be investing in context. That's what gives you compounding returns. So if you've ever wondered, why do some companies and some people tend to get better outcomes with AI than others? Do they have better models? And the answer is no. They're using the same models you and I do. But the reason they get standout results is because of the harness and the context they provide that harness. And think of the AI model as Jerry Maguire, just wanting, like, help me help you. Help me help you. Just give me some context. So give the AI a CRM it can talk to. A CRM that has APIs, MCPs, CLIs. And if those didn't make sense to you, that's okay. Your AI knows what those are. And HubSpot speaks all the AI love languages. It does. It's headless but not brainless. All right. The second 10-minute to-do: write down your own personal instruction manual. How should people and AI work with you? You don't have to overthink it. Just describe who you are, how you like to work, things you would say, things you wouldn't say, how you would evaluate things. Write it all down. Mine says I'm an introvert. Don't ever ask me to call someone. I like to keep things relatively light. I like humor. I like Venn diagrams, that kind of thing. And once you have that instruction manual, put that manual in, just paste it in. to any AI you use. Sometimes it's called a custom prompt. Sometimes it's called a system prompt. Just paste it in. Once you do this, every interaction you have with AI will get better. Which now brings us to the human. That's you. It's most of you. I think most of you are humans. Some of you are managers. And that's awesome. You build, organize teams of people, teams of these carbon-based life forms. And that's awesome. And some of you, like me, are individual contributors. And you're building experiences, you're building products, you're sort of closer to the work. You're awesome too. Now, if you're a manager, there's good news. All the skills that make you a great manager and leader translate directly into the world of AI agents. You're great at setting expectations. You're great at reviewing work. You're great at giving feedback. All of those things are important when working with AI agents and when trying to drive outcomes. So you are even more valuable in the world of AI agents than you were before. Yay for you. If you're an individual contributor, you can now start building out your digital team with AI agents, and you can become a super contributor. So whether you're in RevOps or a sales rep, you can become a super contributor. So yay for you too! And by the way, like I don't mean like yay for you too as in the band. I mean yay for you also. Yay for you also. I probably didn't need to clarify that. So here's the takeaway. Pick one recurring task that you either do or you delegate on your team and write down the job description for that task, something you would give to someone that was a new hire. Once again, don't overthink it. Now, every time you bring on an AI agent, provide that job description. It will increase the odds that you get the output And the outcomes you're looking for. So here's how you grow in the age of AI. The growth-eroding move is to fight AI, resist it, push back. The growth-limiting move is to ignore AI. Ah, it's fine for Susie over there or Joe over there to use it, just not for me. The growth-boosting move is, of course, to adopt AI, to use AI. But the growth-compounding move, the catapulting move, is to compound AI. And that's what we're going to talk about in terms of becoming a builder. So here's how you do this. Another 10-minute to-do. Pick something you do every week. You've been doing it. Sometimes you use AI and push AI further or start using AI to do this thing. Doesn't have to be super hard, just slightly harder than the things you've been already using AI for. And then do that week after week. Sometimes you'll try AI, it will fail, but sometimes it will succeed. And every time it succeeds, you are creating compounding value for yourself. You're pushing the boundary of what AI can do for you. Now, after you've applied this for a few weeks, you'll go from, I'm great at my job, which I'm sure you are, to, I'm great at getting my job done with AI. The first is a statement. The second one is a strategy. Now, let's talk about these super contributors inside of an organization. So, they have their digital teams. There's an equivalent of this. If you take the organization away and represent them like this, they're a solopreneur. They're a one-person entrepreneur. And solopreneurs are speedrunning the future of work. All the things that solopreneurs are doing now is going to show up in your workplace Next year. Watch the solopreneurs and you can be the ones that bring that into your workplace in terms of what's happening out there. By the way, this idea of a solopreneur is not a thought experiment. It's not like, oh well, that seems like really sci-fi, it's going to happen in the future. It's happening now. So new business creation in the United States hit an all-time record in 2025. All-time record. 5.6 million new companies were started in 2025. And this is not slowing down. So far this year, we've already had 4.3 million businesses started. So we're going to break the record again this year. Now, what's more interesting than the number of new companies starting is the shape of those new companies, because the percentage of those companies that are one-person companies, not expected to hire people, they are one-person companies, has been growing. The data is not out yet, but in the last month, that number has bounced to 85%, 86% of the new companies formed are formed as one-person companies now and are not expected to hire in the first 2 years. Okay, so this is the rise of the solopreneur, and these folks are not building little hobby projects. The number of solopreneurs, one-person companies, that exceeded $1 million in revenue between '23 and '25 doubled. The number of companies making $1 million in revenue with one person, the number of them exceeding $10 million in revenue tripled in that same time frame. So these are real solopreneurs building real businesses with real revenues. Okay, I'm gonna take a breath so I don't pass out on stage. You're welcome to take a breath as well, by the way. It's allowed. And so every few years, I get to share something with you on stage that's new from HubSpot that I'm personally, like, super excited about. And today, I'm ready to share with you a new project I've been personally working with the HubSpot Next team on. And we don't talk about HubSpot Next a lot. HubSpot Next is a set of innovation projects within HubSpot, and sometimes they're out to test new products, new product ideas. Sometimes they're there to test new technologies that are out, and sometimes they're there to test new markets. This new project does all 3— tests a new product with new technologies for a new market. It's called Uspot. Good thing that's on my shirt. That's— yeah, explains. Thank you. And Uspot is the AI-native solo CRM. So let's break that down. What does it mean really to be AI-native? Because this term gets tossed around a lot. Uspot is built as a domain-specific, custom-purpose AI harness. So this is not a CRM with AI inside. This is an AI app, an AI harness with CRM inside. So you can use it just like ChatGPT, just like Claude. In fact, it supports both the latest models. You can swap between the two. It just happens to know a lot about CRM and has all the skills and has all the data. And what does it mean to be a solo CRM? Well, the HubSpot Agentic Customer platform is built for companies from 2 to 2,000 employees. We have 300,000 customers now. If you're one of those customers, thank you for your support. You are awesome. YouSpot fills in the gap at the bottom there. It is built for one-person companies, the solopreneur. And it does not replace HubSpot, by the way. HubSpot platform does a lot more, and it's built for go-to-market teams to achieve go-to-market outcomes. YouSpot is for managing data for the solopreneur. Because as a solopreneur, for those that have been— it's like Duncan's dad. They're trying to hold everything in their head, right? Right? It's like you are the CRM if you're a solopreneur. It's like all your email and your social posts and all the things, like the handoff from marketing and sales happening in your head. Figuring out whether a new company fits your ideal customer profile or not, in your head. Remembering to follow up with that prospect that does fit your ideal customer profile, once again, in your head. It can be overwhelming. So let's take a peek inside, YouSpot the product. At the core of it is a thing I call the second brain, and it's just like it sounds. This is where YouSpot stores all the people, the projects, the PDFs, the posts, and even things that don't start with the letter P, all in this interconnected data store that geeks might call the context graph because it's represented as a graph. It's a context graph. Thank you. And the reason I paused— I don't have to explain this. If you have to explain it, then it's not that it didn't work. Okay. So this database hydrates with all the accounts you have connected to it. So it brings in your LinkedIn data, your Twitter data, your email, your social posts, everything, people commenting It brings it all in into this second brain, and it uses frontier models to sift through it all to figure out what's going on, what's important, what matters, and be able to do the work you need to do as a solopreneur. I built YouSpot for me. It's the product I always wanted to manage my own professional relationships. So let me tell you how I use it. Every day without fail. Every morning, YouSpot sends me an email that says, here are the things going on in your world, things that have come into the second brain. These are important comments that were left on social media. Here are new companies that you might want to explore. Here are new domain names that are available that you might want to buy. And it brings that all in. And here's the really cool part: I respond to those emails. I scan the emails like, okay, And then I reply to it and I say, yeah, you told me this, but that's not really a good customer fit. You told me about this domain, but it's in manufacturing. That's not really my area of interest. And every time I respond to that email, it goes back into the second brain and it gets better. This is compounding value. So every morning, the email that I get from the second brain is better than the emails that have come before. And I promise you, you should try it out. You will start getting value, and you will be impressed with the quality of the results that you get back. So YouSpot is available now. You can try it out for free. Go to youspot.com, or even cooler, send an email to signup@youspot.com with your personal instruction manual. Here's who I am, here's my company, here's what I'm trying to do, and it will set up the second brain for you and start sending you emails once you're signed up. Okay. Start for free. All right, back to founder dinners. I'm happy to share that this was not a New Year's Eve dinner with Brian. We're not total losers. This was Valentine's Day. And we sort of talked about the things around scaling and what's changed and what's not changed. And I think the big difference now is that before, Before the gap between customer zero and builder zero, that customer-builder gap was very, very small. Now that gap is zero because you can be the builder. Everyone can be a builder. And you're wondering, wait, me? I'm not technical. I'm not an engineer. Yes, you. You are the builder. And AI does not replace GTM builders, as it turns out. That's an entirely new category and a new role. It actually removes the obstacles to becoming a GTM builder. All right. A recap of the 4 to-dos. So don't forget when you get back on Monday morning, name your number. What's the metric you want AI to move? Write it down. Number 2, write your personal user manual. How should AI and people work with you? Write it down. Number 3, define the job you want AI to do as if you were hiring someone for that job. Important. Just write it down. And last, try one harder thing with AI than you're used to. Just don't keep doing the same rote things that are familiar and that have been working for you. Push yourself a little bit more. 4 things, 40 minutes will be game-changing for you, compounding value over time. All right, winding down. So HubSpot's mission for 2 decades is to help millions grow better. Sometimes we do that by delivering a world-class agentic customer platform. Sometimes we do it by connecting you to thousands of solutions partners that can make that platform come to life. Shout out to all the solutions partners out there. I love you, love you folks. And sometimes we do this by convening events like Unbound, so that we can all grow better together. When HubSpot started 20 years ago, we were inspired to democratize the new technology then, which was the internet, and to help millions of small businesses leverage the internet through something we called inbound marketing. Now, 20 years later, we're inspired by the ability to democratize AI, make it accessible to all of you so you can get the most value out of it and drive outcomes in the outcome era. It's about democratizing AI and helping you be undaunted, unconstrained, and unbound. Thank you. It's always an honor to share notes from the journey with you. Uh, this is truly my favorite time of the year. Like, after I leave stage, that's my favorite time of the year and, and interact with you after this is, uh, done. Enjoy the rest of Unbound. Thank you so much.