Is SaaS really dead? Dharmesh Shah from HubSpot on AI, Vibe-Coding & the Future of Work

We have all the resources in the world. The number of SaaS apps that we have replaced as a result of all this kind of army of resources and talent that we have— we're a software company through and through— is exactly zero.

Okay, folks, welcome to Agents of Scale. I'm here with Dharmesh Shah. He is the co-founder and CTO of HubSpot. Dharmesh needs no introduction. Dharmesh, welcome to Agents of Scale.

Thanks for having me. Great to see you.

Yeah. So Dharmesh, I must admit, for the CTO of a fairly large public company, I end up talking to you more often about culture. Somehow you got deemed the culture guy at HubSpot. And, you know, one of the things we end up talking about a lot on the show is how different cultures are navigating AI right now. And you have a particularly interesting phrase I've heard you say, which is, um, the job isn't to compete with AI, it's to compete with AI. Uh, can you like talk about what you mean by that and like Yeah.

Yeah. So, you know, I had this experiment I ran with an online poll and the question was kind of posed randomly is like, how do you compete with AI? And as it turns out, that's an ambiguous question and it could be interpreted in one of two ways. One is how do you compete against AI is one possible interpretation. And the other one is how do you compete using AI? So that kind of with is, you know, carries those kind of dual meanings. And I would posit that those that are thinking of competing against AI are, that's not the optimal way to think about it. I think the right way to think about it is how do I as a person compete better, grow better using AI? I think that's a more productive, just generally a better approach. And I'm a techno optimist and I'm optimistic about AI, but I generally believe that, and lots of technologies have come before and this one obviously is very different and much more transformative. But I think it just elevates what humans are capable of, and that's one of the reasons I'm excited for it.

Yeah. I tend to think that trait is fairly common amongst founders who are optimistic about the future and what's possible here. Can you share how have you all built a culture inside of HubSpot that takes that approach versus maybe takes a less optimistic approach? I think when you get to a certain scale, you have an organization that's filled with all types of people, and maybe not everyone comes in with that default positive framing of that with word.

No, and that's true. And it's been a, you know, as many things are, it's been a journey in terms of getting people comfortable with the idea of AI. We were kind of in the game early. And I think one thing that helps just on the kind of cultural side of it is you sort of have to tie it back to the things that we believe at HubSpot in terms of culture. And one of those things is around That we are here to create lots of customer value. That's kind of the overall mission is to help millions grow better. But if you kind of double-click on that, the only real way for us to deliver on that is for us to continually get better, is to continually grow, to continually learn. And it's easy to say those words in any organization, right? It's like, okay, fine. But you sort of have to make the investment. And so what we've done at HubSpot is to say, okay, We think AI is going to be a really big deal, both for us internally and for imbuing into our products and our platform. But we need to carve out the mental space. We need to make sure it's okay for people to experiment. It's like, oh, I went off down this rabbit hole and I tried out X on AI or something like that with AI and it didn't quite work. All of those things are not just tolerated but encouraged. It's like, okay, if we're not trying enough things, playing with the new tools and getting excited about them, I don't think we're doing our jobs. And so then we also carve out time. So we had a week earlier this year where we spent and we had outside speakers, I spoke, and just to talk about what's going on, how I'm using it, and just encourage folks to take some time away from their day-to-day and actually invest in learning and playing and experimenting and trying things.

Are there rituals or practices that you all have instituted that have helped with that experimentation, that playtime?

Not yet. And I regret not doing something sooner. It's just been a busier year than usual this year. But it's a good idea. I'm a big believer in kind of taking kind of good practices and turning them into habits and making them rituals and making them something that's kind of part of the fabric of the company. We're not quite there yet. I think you folks have done like a better job of doing that, just kind of baking in some of the rituals. Yeah. It's great.

Well, I think, you know, for us, the— we were sort of— we're able to just like fall back to the founding moments where Zapier started at a hackathon. We have long rituals of like running hackathons internally. And so AI showed up and we're like, well, like, we'll do what we always do, do another hackathon.

Sure.

And, you know, that I think has been a huge accelerator of adoption, at least for us internally, especially in like the non-technical orgs where we open these hackathons up to, your marketing teams, your sales teams, your finance teams and whatnot and give them access to tools that maybe are a little friendlier to them than maybe these, the cogent tools are.

Yeah.

As a result, they're able to start to imagine, hey, here's how this process might change for me, or this day-to-day task might look a little bit different. That has been a huge vector for us. I'm curious for you, Are there— so if it's not a ritual, like, what has been the practices maybe that have worked the best inside of HubSpot at encouraging this? Has it been, you know, these, these like your own like demo show and tell, these internal talks, or is there other, other practices? Or when I come back to the practices, other cultural elements?

Yeah, that works really well in terms of the kind of internal demos, and we've had hackathons and things like that to do internal projects. But what we found sort of the most, you know, we are at heart like problem solvers, both kind of internally and then externally for our customers. And so we'll make a list of, you know, what are the things that we think, and when I say we, I mean the collective we, that are just inefficient or we could do better with AI, they're like, it might maybe works, right? And we have, just like many organizations have, these internal kind of champions that are often a little bit ahead of the game. You know, they've either started sooner or just, you know, have a more, experimental mindset and they just jump in. Or someone comes, it's like, you know, this is sort of painful, um, that we have to do this, this, this, and this. And like, isn't it possible now? Um, and so one thing I do encourage people, um, and I'll— and I mean this literally, they don't take me literally, but they should— is that, uh, let's say you try something with AI. It's like, oh, I tried this thing or whatever, I'm gonna try to automate this or do that, whatever happens to be, and it doesn't work. It's like, that's okay. What you should do, it was— if it was important enough to try, then put a date on your calendar for 6 months from now and try it again. Because this stuff moves so quickly that, you know, often I think a common mistake is that, oh, I did that, I tried that, I tried ChatGPT for X or Cursor for Y and it just didn't do it. And then they said, oh, well, AI doesn't work for that. It's like, well, it doesn't work now for that. But when on an exponential curve, you need to sort of kind of retry it.

Yeah.

And it might just Might just work.

One of the— I'm, you know, I heard a, I heard a story that I think is worth sharing. I don't know if you have much to say about it, but I think it represents how you, you know, approach the role inside of HubSpot. I'd heard from somebody who shared with me that one of the reasons Dharmesh is so effective inside of HubSpot is you can act as like elbow grease to get the org doing more stuff. And they'd shared that they were— I forget what tool it was, but they were trying to use this or that or the other, and And maybe things were getting stuck in procurement and Dharmesh was like, here's my API key. Like, you just go nuts and have some fun on this thing. And like, I think that's another great example of like just, you know, someone in your stature being able to say like, hey, I'm gonna help encourage this innovation and just find ways to, you know, continue to cut through the, I guess, orange tape in this regard.

Yeah, yeah. And part of it is just, you know, and we do this. So Yamini, our CEO, you know, has like a weekly internal communications that she does and she's a daily user of AI, right? sort of leading from the front in terms of just how she uses it in her kind of day-to-day life and encouraging people to kind of adopt it. And I think it helps. It helps when people see their leaders actually not just kind of saying the words that seem to be popular, but actually like digging in and doing real work and making it part of their kind of daily lives. So yeah.

Yeah, I think that's— it's hard to— I mean, it's a It's hard for anyone to want to follow someone who they themselves aren't willing to do the thing. And so if like you say AI is important, you probably got to be one that's saying, hey, I, I, I'm, I'm dogfooding this stuff as well too.

Yeah. And I'm— it's one of the things I found helpful for me personally, uh, because of my excitement around in terms of galvanizing the organization, is that like I'm usually, uh, kind of customer zero or user zero for a lot of these things, uh, partly through my kind of startup activity. So I get early access to them. Um, and so by the time— and they'll kind of read about it and things like that, it's like, oh yeah, that's the thing that, you know, Dharmesh was talking about, you know, a week ago or a month ago or something like that. And yeah, and that helps, um, for just— yeah.

So another thing I want to talk about is, you know, demystifying AI. Uh, HubSpot sells predominantly to, uh, entrepreneurs, marketers, and, and small and medium-sized businesses. So these are folks that are usually not coming at AI from a strong engineering mindset, but they are builders. They do like to make things. What have you learned about helping these folks actually deploy AI inside their companies?

Yeah, I think the first thing is that there's a bar every organization has in terms of what's the highest amount of risk they're willing to take or how readily are they willing to put kind of AI in their organization? And the thing that's worked well for us is to kind of tell folks it's okay to have that bar. You can define, it's like, okay, I'm not going to put something that's going to be front-facing for customers until we've gotten either this CSAT or tried it internally or whatever your kind of criteria are. That's completely okay. What I think is not okay, and we try to talk our customers off of this, is to have some arbitrary line It's like, oh, it's just too risky. It's like, okay, well, it's risky. Why? Like, what is it and what would it have to do, uh, or in order for it to demonstrate that it's ready? And it doesn't have to be a kind of production-oriented— even if it's an internal project, even it's for a small team, whatever your constraints are, you sort of have to like define those constraints. Otherwise, uh, this is just human nature. The tendency is that it'll, it'll never be ready because you keep moving the bar, right?

Yeah.

It's like, okay, well, Fine. Like, if you have something that says we need to see, you know, 500 tickets, uh, you know, handled in an, uh, automatic way with a CSAT above X before we kind of do a wider rollout, that's completely fine. Um, and so alongside with that advice, and this goes to the, uh, you know, sometimes, uh, fear is actually a good motivator, is that if you do wait, uh, until the point where it's completely, uh, risk mitigated— not completely, nothing's ever completely risk mitigated, but, um, you sort of do fall behind. This is not, you know, things are moving really, really fast and you don't necessarily have to keep up with every model update and things like that. But you can't really sit on the sidelines. And we saw this before with the internet, right? It wasn't— it took years, right? Like, but within a few years, like, okay, if you were a business and you didn't have a website, that was sort of kind of unthinkable. And we're sort of at that stage. You should be using AI in some way and you should sort of be expanding that circle of what AI can kind of bring to your organization. You don't have to take a, a leap into the deep end on time t equals 0, but you should start somewhere.

Yeah.

And then kind of increase your confidence level over time.

On that start somewhere, are there like really good starting spots that you've run into with your customers? It's like, hey, you're, you're maybe thinking too big about this. Like, if you actually start in these more practical places, you can start to stack wins.

Yeah, I mean, the, the obvious ones. So at one end of the spectrum is around, uh, kind And this is where generative AI kind of got started, is around creation, content creation, writing a blog post, creating an image, writing an email. All those kinds of things are relatively low risk, right? So let's say you give it a chance at writing a blog post or at least the first draft of an email or whatever it happens to be. There's still a human in the loop. It's you, right? And you control whether you use the output or not. Fine, right? Like, that's an easy kind of low pain kind of thing to try out. Over time, as you move down the spectrum, you bring in other people. It's like, oh, I'm going to try to automate this process that 3 people touched or whatever. I'm going to take one of the steps and I'm going to automate it and see. It's like, oh, it's going to do some lead routing algorithm or whatever, but then someone still verifies that that was the actual right thing to do. Then you get to like the, oh, I'm going to put an agent in front of customers to do frontline support kinds of things. It's like all of those are along a spectrum, but it's easy to do the single-player mode. It's just me and the machine. Let's see what it can do. And I, and I get to be the sole arbiter of whether it was good enough or not.

Mm-hmm.

And nobody even needs to know. And then you sort of move it. It's like, okay, now I've gotten some level of comfort for these kinds of things. Now maybe it's not just creation, now maybe it's synthesis, right? I'm gonna be able to take a body of information and distill it down into either a different media format or I wanna create this 300-word post of a 50-page PDF that's really important to my industry, whatever it happens to be. So there's these kind of levels of the game that you can move through. And it's usually creation is the easiest one to get started. Then synthesis is likely the next. There's, you have automation. And the other one that's really fun, and it's just short of the customer one, but is simulation, which is very interesting that says, okay, well, if I can describe to AI what my business is, who my customers are, all these things with, you know, some degree of fidelity, spend some time on it as if you were being interviewed by someone, But then you can pose the what-ifs, right? Similar to what we might do in Excel, you know, 20 years ago. It's like, what if I launched a product in this bucket that sort of intersected with our platform in these ways? What if we introduced a new pricing tier? What might happen? And AI is actually shockingly good at coming up with a response to that, right? And it's not going to be perfect, but it's a really good kind of idea generator. And given— and Once you kind of pull out this thread far enough, you can say, here are the types of customers that you can create effectively virtual audiences or virtual panels. It's like you had a customer advisory board of 12 people that have these profiles, and then you put this idea in front of them, and it's like the AI can sort of simulate that. It's like, oh yeah, one of my customers is this. It's the, you know, the partner at a 50-person law firm, whatever it is. And you sort of kind of simulate what the real world might look like.

That's fascinating. Is there a particular problem that you've used simulation for recently and how did that work?

Yeah, I use it a lot for speaking things. It's like, okay, well, I'm going to speak to this audience and maybe I have some sense for the questions or topics I'm trying to address. It's like, how will this be received? Is it controversial? I'll try to— and this is, I think, My most common use for AI, for me personally, outside of agentic coding, which we can talk about, is just as a sparring partner. So not as a writer, but as an editor, as a critic to say, okay, Dharmesh, you said this right here, but it's logically inconsistent with this thing you said earlier, the thing or whatever. Or you said this, here are the gaps in your reasoning, whatever. It didn't quite make sense. Or this was not clear enough, or this is controversial, or this like, for this particular audience, that doesn't feel like that's gonna land as well as you might think it's going to. And I like to treat it very conversationally. It's like, I don't like to kind of attribute human qualities, but it's very good at that too.

Yeah, yeah. Are there particular tools you're using for simulation or is this like, hey, I'm just going back and forth with ChatGPT asking it to—

It's a combination of GPT, but I have like my own intro 'cause I like to do a lot of these things outside of the kind of classic, bot and ChatGPT interfaces, but I just use the API. But it's those core models.

Yeah, we've seen, um, in fact I've started to experiment with this too, where we'll build out GPTs for different people internally that sort of has their, you know, like personality tests, like tendencies, principles that they like to obey by. And then you, like, to your point, you can create like, oh, here's the, the council, uh, internally, and you can ask each GPT how they might respond to a different situation.

Yeah, yeah, it's what What gets interesting with these things, and this is going off track a little bit, but I think it's fun and interesting, is that the more you give these models in terms of the grounding information they have access to, and we've seen this with all forms of retrieval augmented generation. One of the things I found very useful, so I have lots and lots of emails over 30-odd years of being a professional, and I have all of them, and I've done multiple passes at creating vector embeddings for all of them. And so I can then say, oh, well, based on my conversations with Brian, my co-founder, how did we arrive on this particular thing or whatever? What were the kind of sides that we considered? And it's really, really good at kind of taking that kind of prior, like, so I can kind of, because you sort of forget and it's got the historical record because most of the conversations that have happened internally, especially in the early days, were all in the form of email.

Mm-hmm.

Even if Brian and I were sitting in the same room, we'd still send each other emails. That's how we operate it. So yeah.

That's fascinating. Have you tried asking it things like, you know, how have we— like, what have we changed our mind on over like the last 10 years? Or like questions like that?

I haven't asked that. That's actually a good one. What things I have asked is around if I were to try to summarize kind of decision, if I'm trying to onboard new people, or we're kind of revisiting, you know, a decision or something like that. It's like, you know, one thing I'm like a big believer in is that I don't like second-guessing decisions.

Yeah.

But after a certain amount of time has elapsed, I think some decisions are actually worth revisiting.

Totally.

Because the context may have changed dramatically, you know, since the last time we made it. And we don't remember, and sometimes the people weren't there for that initial debate, is to go back through that history as just to educate ourselves and educate new people. It's like, here's how we kind of came to that and be able to kind of write a meaningful kind of article that captures Yeah, this is one of the use cases I'm most excited about for call recording.

So I've fallen in love with this tool Granola, not just because it takes nice notes, like that part is great, but the context of being able to search and ask questions over a large corpus of conversations is incredible. You know, I had it like we've done internally a lot of work trying to write and describe our culture, and I tend to think we We generally do a good job at this. And so one of the things I tried with Granola is I said, hey, can you tell me what like the unspoken culture of Zapier is based on all of the meetings that we're in?

Yeah.

And it writes up, here's what I think it is. And as I read through that, I was like shocked. I was like, wow, they're— this is a better answer than all these like thoughtful things that I had thought I'd written over the years.

Totally.

And so it's just fascinating to see how like these tools know us better than we know ourselves in a lot of ways.

Yeah. One, one, I'll call it a pro tip. I'm not sure if it qualifies to be like a pro tip, but so if you think back, Wade, back when you were originally building, building Zapier, chances are since it's a web app, you were sort of logging events that's like, oh, the user did this or this or whatever.

Yep.

And you didn't, hadn't built the dashboards or whatever. You may or may not have been using an analytical tool for just kind of doing the kind of classic product analyticsy type stuff, but you just like, I'm just gonna throw this and you would just dump the data. Because the reality is the cost of dumping the data was super simple. It was a single line of code that you could just write to whatever data store you're writing to. And but then you had that data that you could then later say, oh, well, now I've got this and I can do something with it, right? And I have had very— and I talk about the email example. Another good example is like screenshots that if folks are not already using something that will actually do a permanent storage, like you do the clipboard thing, whatever, I'm going to take a screenshot of this and paste it into an email, whatever. You should be using a tool. There's hundreds of them out there that will actually save it to a local folder. Not that you're gonna go back, but there will be a day that future self will thank you for having that kind of historical record of all the screenshots that you took because you'll be able to do video analysis. You'll be able to find, it's like, I know I looked at this particular version of this app or whatever, internally, externally, doesn't matter. And you'll have questions that you could not answer before that you will be able to answer someday.

Yeah. I, I had heard an interview with Toby from Shopify where he had shared he'd been like, Logging his.

He did this on the 15th level. Yeah, he was logging every keystroke.

Yeah, and I was just like, I mean, it was one of those things where I like I had FOMO all of a sudden, where I was like, like, oh my! Like that was not even a thing that had dawned on me that I should maybe consider doing. But in this day and age, how valuable to have that data set. It's incredible. So. Maybe shifting gears again, back to the more human side of this. Walk me through some of the techniques you've used building AI to make it feel more approachable to this non-technical audience that HubSpot has built up. Yeah.

A couple of things is we have to internalize. So yes, AI is, you know, very, very new and it is transformative and there's all these things and you can sort of kind of geek out about it as we do. But looking at it from like an SMB's perspective, you know, mere mortals that are out there kind of living their lives and growing their businesses, you sort of have to accept and kind of say, this is software. Like, that's what it is. It happens to be smarter software. It can do more, but it's not magical in that way, right?

Yeah.

It's like, yes, it's kind of lifted it. And so, and one of the things I think a mistake the industry is making right now, and we're getting getting better about it is that because of ChatGPT and its popularity, and because we got— we all had that magical moment the first time we, you know, typed in an English prompt and got like a meaningful cohesive response back, you know, it was like life-changing, right? We all had that light bulb moment. But just because that magical moment happened doesn't necessarily mean that's the best possible way for a human to interact with software, right? So one of the things we're doing right now is like, yes, there are certain use cases where a conversational UI makes perfectly good sense. There are other use cases where you can use AI that the user doesn't even have to know that there's AI being used. It's like, oh, well, this thing works pretty well. I don't know how the heck it does it. And they're gonna assume AI. But I think over time, like any good technology should sort of fade into the background, right?

Yeah.

It's like, so like now we have the internet, it's like, oh, all the computers are connected and we have REST APIs. We have all this stuff right now. We don't really talk about that anymore because it's sort of built in the fabric of the way machines work and software works and we work. I think AI, we should be nudging humans along that in a gradual way. We should not say, oh, everything's different now. Everything you ever thought you knew about software is gone. It's like, no, actually a lot of that still makes sense. You still have the same problems that you used to have. We have better ways of solving some of them and help people bridge the past to the current.

I totally co-sign with that. I feel like a lot of the products I'm using right now, there still is a fair amount of configuration involved and that configuration, Feels like it's an unnecessary friction point, uh, in this day and age. I mean, we run into it a lot talking to customers when we're trying to describe the difference between like a workflow and an agent. And every time I have that conversation, I'm like, they don't care. Like, why am I having to have this conversation? Like, I— it should just fade totally into the background and you're just saying, hey, I want to automate this, I want to build this, I want to do this thing, and it knows how to go figure these things out. Yeah.

Yeah, it helps a lot actually to— and we can all do it when we do it. It's like if you use your own products, even if you are an expert in AI, there are some things that are just cumbersome using a particular modality or something like that. It's like, okay, well, yeah, I know AI could do this in a textual kind of like, but if I had this UI, that would be a much more efficient, better way to do this thing. Yeah.

Are there sleeper use cases right now that you feel like are just not more broadly talked about?

Yes, I think we're yet looking at what I think of as dynamic or fluid UIs yet. I think that's the bridge between the two. I think the thing we like about conversational UI is that there's this nice easy thing where I'm not converting the thoughts in my head of what I'm trying to do. I can just tell the computer in English or whatever my natural language happens to be, and the computer goes, does it. The other end of the spectrum is this very static UI that some designer went through and we did a bunch of customer testing, whatever, and we have this. There is an in-between that says, okay, well, it's something that's more natural than the static UI, more personalized than the static UI, and maybe it's some sort of a hybrid dynamic interface, right? That's the thing that I'm starting to play with now. I'll give you a concrete example because it's a fun one. One of the things I'm working on right now personally, is this AI agent for image generation for businesses, right? So right now we know image generation, we— there's a bunch of really good image generation models and most of them are kind of text-to-image, right? You type in whatever image you're looking for and it comes out with an image. It's like, okay, well that's useful, but for my personal use cases, whether I'm doing it for a blog post or need a visual for a slide deck or something like that, it's like, okay, well here's my actual use case. My actual use case is I have a particular style that I like to use for my social media posts or whatever, and I have already 5, 10, 50 images that are along those lines. What I want to be able to do, just like if I actually had access to a designer, which I've had in the past, I would just say I'm going to give it these 5 images. Great. They would come back with— I give a rough idea of what I'm looking for, it'll come back with something. It's like, okay, well, generate an infographic of this or whatever. It's got some text in there. The models are really good now. Here's the leap I had last night from a UI perspective. It's like, okay, So it did the thing with a single prompt. It came up with an infographic that described the timeline of HubSpot's evolution or something like that. If I'm you know talking about our history, and then to be able to do a revision on that to say okay well I can say okay well tweak this and tweak that, and we've seen that now with Nano Banana and others. The leap now is okay well we can do an interpretation of the image that it generated because I didn't give it the text. So it's like oh here are the milestones it put. Change the text for what happened in 2006 from calling it. an IPO to a whatever, right? And what it does though, it presents it in a, here's all the elements of text that were auto-generated, just overwrite the ones that you don't like. Yeah. Not have to describe it and say, change this text to that text. Actually give me a visual, like a modal dialogue that lets me say, okay, well, here are the text that I detected in this image that I generated and let me override and, and kind of tweak it. That's interesting. Yeah.

So you could, you could put kind of more of like a classic editor in front of you where you can just come in and be like, I'm gonna, physically change the text on this like I would in any other tool.

Totally. And the editor itself, right? So, and what you don't want to do is overload. It's like, oh well, you might change text in an image or whatever. It's like, no, there will be situations where there's a lot of text in the image or whatever. That feels like a good time to give the user the option of doing this. There might be other cases where there are 4 discrete objects or there's 2 people or there's something happening. Yeah. Here are the natural kinds of things that you might want to change. Yeah.

That makes all the sense in the world. That should definitely exist.

Yeah, it should exist.

So I want to shift to another topic. You often share these anecdotes around like what you and your son are doing and how you all are just like watching him like learn new stuff. So, you know, you'd shared this story earlier about watching your son learn to program in English. Tell me more about what your takeaways and learnings are, extrapolating maybe his experience to a whole generation of folks that are maybe coming of age right now.

Yeah. He started right now, he's— okay, so I'll take a step back. He's a big reader, loves fantasy fiction, and that's on an aspiring author someday. His early use case with AI, and he's been exposed to it super early outside of the normal things like writing and whatnot. But it was building a text-based video game similar to, if you've been around, like Zork and the Inspector's Guide, Infocom games, right?

Yeah.

It's like text-based adventure. So he built his own to do that and did surprisingly well. But then he kept pulling on that thread. So what he's doing now, and he's got a very intricate kind of 2,000-plus word prompt. Um, so he's got this kind of fantasy world that he's built out that he describes. It's like, here's my world, here's the context, here's, uh, the laws of power, here's how people accrue magical power, here's what— like, here's all the things, right? The kinds of things that can happen in this world, in this kind of imaginary world. Uh, and then what he does— and this is sort of, uh, where I got my simulation idea from— then he's like, okay, well, what would happen if someone came along and had this particular power, or if these 2 people you know, kind of butted heads and he sort of is testing his world effectively. It's like, oh, well that would be like, it would create an imbalance and this particular one just like win all the marbles kind of thing because it's completely, you know, non-winnable once this happens, right? So he kind of tweaks it. So that's his kind of primary use case and this is what he tests new models with is, and what he's solving for in terms of his model selection is what follows the instructions the best. Not whether they can write words that are particularly poetic or whatever. He just wants to create this simulated world and be able to poke at the things, whatever, and try things and evolve it. Then he's going to use that as the basis for the book that he will one day write, which I think is an awesome use case.

That reminds me of being in comics or reading comic books and stuff as a kid where you'd always have these arguments with your friends on who would actually win, Superman or Batman or da-da-da. And like actually building a world that can simulate some of those things and say like, oh, you've created a character that like is, is too overpowered. Like it needs an Achilles heel or—

Yes. You know, that's a common thing, right? Like most of these systems will degenerate into, well, will some particular attribute become just so powerful? And we saw this in, in, uh, you know, World of Warcraft back at the time was like, there were certain things like if you were a particular jerk about it, it's like, it was just like an unwinnable, uh, you know, combination, uh, just too powerful. Yeah.

Yeah. That's fascinating. I'm, I'm sure there's like a, I, I would have to guess this technique is already in practice for like game designers and things like that.

Oh yeah, it has to be. It just makes too much sense not to be. Yeah.

What, you know, how, how are you seeing this generation of folks that are now entering the workforce? What, what skills are they bringing to bear? Like, how are they using things the same or differently, um, compared to maybe others, uh, in the past?

Yeah, it's interesting. I think it's a little early. What's happening, you know, the, um, at HubSpot specifically, you know, our kind of— we kind of skew a little bit, um, earlier because we hire lots of people earlier in their career. So maybe the average age is 30-ish, give or take. Um, but, you know, we're just seeing this cohort of what I would think of as like the Gen AI folks, um, kind of coming into the workforce. Uh, and I'm not finding a radical difference yet. Um, and it's So, you know, all things being equal, the person that's kind of been in the workforce for 5 or 10 years is probably as likely to be super excited about AI and playing with all the tools as someone that just came outta college. I don't know why that is yet. Maybe it's just at, you know, the time that that happened and it's like, you know, hasn't sort of absorbed in. And it may be different, you know, in, in 5 years, you know, once my son's of, of kind of working age. But yeah, I haven't seen a, a dramatic difference in, in that regard yet. I am, It may be that the dominant factor is not what age you— whether you grew up with AI or had early exposure to it versus just the— are you just naturally curious? That might be the dominant variable versus the time-based one. I don't know yet, but yeah. I'm biased about being a little bit further in my career, but yeah.

Yeah. I've been asking a lot of folks this question too, and I think I've drawn a similar conclusion to you where You know, amongst my small sample size where I'll ask folks who hire a lot of interns, hire a lot of folks, you hear responses sort of all over the spectrum where on one hand you'll hear stories of like, you know, these folks are using AI like you've never seen before, to, you know, these folks barely know how to show up to work on time and stuff like that. And as best as I can gather, it's just like any other generation before where there's some folks who are very curious and apply themselves diligently, and there's some folks who you know, use college to have a good time and prioritize other stuff and, um, you know, everything in between. Uh, and so it feels like, you know, at least at this moment in time, like it's the same skills that you always wanted. It's grit and determination and curiosity and hard work and all that other stuff. And there's a chunk of people that have that.

Yeah. One thing I will share though is, um, just another kind of observation. So if you think about, you know, startups versus, uh, you know, scale-ups and, uh, you know, grown-ups, very large companies, You know, the one kind of difference between them, I think, is, I don't know if I've ever written this, but I think it would make a good LinkedIn post, is around the advantage that startups have over, you know, massive companies is not the presence of any specific thing. It's the absence of all the weight and baggage that the larger companies kind of accrued over years.

Hmm.

All the complexity, all the steps, all the processes and things like that. So absent that, most of the large companies have lots of resources, lots of things going for, they have brand, they have customers, they have all these things. And I think the same exists in this kind of AI conversation that we're having is that the people with experience, the one thing that the newer folks have as an advantage is the absence of the preconceived notions. So if you are a senior engineer that's been doing this for 25 years, you may have this kind of preconceived notion that there is no way that software could do this or it could never possibly work because of all these nuances. Yeah, I know it can, complete this line of code, but this, this, this, this, and the other thing. So they'll sort of kind of stop themselves because they're a little bit more skeptical than someone that's coming just fresh into it because they haven't had that kind of built-up kind of experience, right? So that's the one possible thing that may kind of be a kind of counterweight. But yeah, but as long as you're kind of curious and willing to put that aside, I think you're—

Yeah, that makes sense. Like, the new generation gets to benefit from this blank slate. You know, this is an area where I occasionally get jealous of the new startups where I'm just like, oh man, they get to do it all from scratch and sort of reinvent from first principles all these things. I'm curious for you, how has HubSpot approached that problem of like, hey, we do have like years and years of way of working. Now, to your point, some of that's very good because you have lots of customers and a reputation and all this stuff that's very positive, but you also have things that will need to be reinvented and reworked. And so how have you approached that aspect of this?

Yeah, you sort of have to, um, you have to do both, right? So we have, um, kind of existing things that's been working really well, existing kind of products and platforms and processes for building those products and platforms. And we're doing, um, I think the, um, the obviously right things to do as far as features to build and kind of upgrading the, the core product and platform. Uh, and that's going, uh, really, really well. And we are simultaneously sort of reimagining that says if we were to start with a kind of blankish slate, what would the world look like? So we have like a series of bets that we make that are kind of independent of the core product and the core product teams that are kind of proving out a particular thesis that we think the world is going to look like X based on— and here are, you know, 5 possible things that could happen over the next, you know, 5 or 10 years. What are the bets that we want to have on the table? Like what needs to exist? Can we go create those things? Can we try that? Can we prove or disprove that particular theory? And so that's something we've had in place for a little while, not that long a while. Yeah, so.

Are there any particularly, you know, contrarian approaches you think you all are taking that seems to go against what, you know, a traditional take you might find on LinkedIn or X or something like that?

People are talking about this. So if you look on like the core product, you know, we have hundreds of engineers. And the thing I think we're getting really good at is now we're thinking through how do we weave AI into the actual process? Forget the code generation, forget everything else. It's like, okay, well, when a PR gets submitted, it's like, okay, well, you know, we have these kinds of approaches to how, you know, how many people review them, how do we do it, what parts of the code, all this stuff.

Unit tests.

And so that I think we're doing really well of saying, okay, well, we get it now. The reasoning models are capable of producing, you know, vibe coding an entire module or doing this or that or the other thing. The thing we're kind of really digging into is thinking of, and I don't think people have written about this yet, I haven't seen it, is around one of the advantages that companies like yours and mine have is that we actually have a codebase. And the codebase historically has been used to produce product, right? And so we iterate on the codebase and we build new products. And the way we all do these things is that some portion of the codebase that already exists is like a framework shared services thing that we can build new features that will leverage the stuff we've— lines of code we've written before. That's, as you know, a tale as old as time. The thing that's different now is don't think of code as just something you leverage and, you know, shared services for building new products and features. Think of the code as data. If you think of the code as data that says, oh, well, part of the code is we have 1,000 unit tests or something like that. It's like, okay, well, maybe in a future world someone's going to try to vibe code a CRM, let's say. It's like, okay, well, what advantage do we have that maybe they don't have? It's like, okay, well, we have access to the same models. It's like, okay, well, the thing we do have is we have an existing body of code tested over 19 years or whatever and maybe The languages are different, maybe the approaches are different, maybe the architectures are different, but there's something in there that's actually pretty useful for the same reason that large language models are trained on all this kind of public internet to create new things or whatever. It's like, like those lines of code are useful, the unit tests are useful. Then we can say, oh, well, maybe everyone's vibecoding their own personal CRM, but the only company that can do it is us because we've got the unit test to prove that can still work in high fidelity with what existed before. That's something that's relatively hard to pull off in another way.

You went a direction I was maybe going to go next. There's this debate that I see pop up, gosh, it feels like daily now, which is, is SaaS dead? Is somebody going to just vibe code their CRM in the future? You started to hit on why that may not be the case, but I'm curious, how you all think about that, um, uh, that debate that seems to wage on, uh, perpetually at the moment.

Yeah, I had a— I have a biased, uh, but not necessarily wrong opinion on this. Uh, you know, will people just be vibe coding their apps and is SaaS going to be dead? I don't think so. Uh, and here's, here's why. It's like, okay, so HubSpot is 9,000 people. We have hundreds of engineers that have been doing this a long time. They are professional developers. They have moved products into production. We have all the resources in the world. The number of SaaS apps that we have replaced as a result of all this kind of army of resources and talent that we have— we're a software company through and through— is exactly zero. The number of SaaS applications we plan to replace, uh, using our army of developers that are really good at shipping product or whatever is exactly zero. And the reason is, is that, okay, well, even if you could, let's assume That even today or someday very soon, you could vibe code a high-quality production application that does X, whatever X happens to be at a particular point in time. Why would you do that as a business person? I'm like, I want to spend every possible calorie and every engineering hour, every product— like, I want them in service of our customers, not for eking out, like, driving our cost for the expense tracking system or to rewrite Slack or whatever. It's like, That just seems completely misguided to me just because you can, even from a technological perspective, which we could argue whether that's actually possible or when it's going to be possible. But even if you could, if you stipulate that you could, why would you?

Yeah.

And that's for us at scale software company that has the resources. If you are the other 90% of businesses that are out there, it makes no sense for a 100-person law firm to say, oh, well, why would I pay Zapier to do X or HubSpot to do Y? When I could just, you know, vibe code that. It doesn't make any sense to me. I just, I have a hard time. So it's easy for people to sort of pontificate and theorize online. It's like, oh well. But I don't think— I've talked to customers all the time. I talk to salespeople at HubSpot all the time. I don't think we've ever lost a deal where the customer said, oh yeah, I'm not buying HubSpot because I've chosen to go vibe code my own. I've not heard that. Not that it doesn't happen or couldn't happen, but yeah.

Yeah, I, I like where I kind of think I, I have a similar point of view. I'm like, look, there's lots of things that, you know, in my life that theoretically I could do it myself, but I still choose to do it. Like, I, I could, you know, uh, I, I can make my own food. I do make my own food, but I still go to restaurants a lot.

Uh, yeah.

And so I think on that hand, you have one of these things, but also software, to your point. This is, um, a place where if there is rock-solid production-grade software out there that you can buy at a very economical price. Yeah, that is always a better exchange than like, you know, putting engineering time and attention to it. And so I've always thought—

Yeah, one additional argument I have to get out there, uh, is, um, you know, we sort of, um, kind of over-index on the value of the actual lines of code. Yeah, what the actual lines of code are are a manifestation of the domain expertise and iteration across many, many years from really smart people talking to customers that are And that's all they thought about for a long period of time. And now it's embodied in what we call this kind of codebase, right? That's— and the reality is, even if you could reproduce that codebase in a particular point in time, you're sort of like— part of the value I think, you know, software companies like yours and mine bring to the table is that we spend all day, every day thinking about these set of problems. It's like, and we try to solve them to the best of our ability with the tools that are available.

Mm-hmm.

But it's like, Okay, well, yeah, like even if I could cook, I'm never going to be able to cook as well as, you know, the best chef that's out there or whatever. It's like, I think it discounts this idea of kind of a domain expertise that you build. And it's like, okay, well, a law firm's not going to think about building software for automation, or, you know, it's just that's not what they do.

Yeah, I think the heuristic I've kind of settled on is if you can buy the software, you should.

Yes.

If it's not available for purchase, Then okay, and it is valuable to you, then great. Vibe coding now opens up a new way to build internal tools, things that you couldn't get access to, etc., before. And so from that point of view, these vibe coding tools are incredible. It's like, oh my God, I agree. The marginal piece of software that just doesn't have like huge amounts of economical value but is important to you at this moment in time, now you can have it.

Yeah, yeah. And Just, you know, kind of, I guess, closing the point out a little bit is that, so we've had, you know, low-code tools and no-code tools since even before the AI days. And they have been super functional in creating departmental apps and internal things or whatever, because like on the edges, there's not going to be a SaaS application that you can buy that does exactly this thing that you need for this particular purpose within this department of, you know, your organization. I think what this does is expands that particular sphere of what's actually possible now. expands the scope of what's possible, but you're still likely not going to build production apps that need to be maintained over a long period of time. They're going to have customers that you need to support. That's the thing. If you're doing it for 4 people, maybe now those teams that the bar was here and maybe now it's lower because it's cheaper and faster to produce these things now with what's available. But yeah, anyway. Yeah.

It's hard to build great software for a fleet of sales reps. You need something that encodes well-earned domain wisdom over many, many, many years.

Yeah.

Let's shift gears to talk a little bit about AI as kind of a career amplifier. I've heard you talk about how, hey, AI is not going to take your job, someone using AI will, but then you kind of add back in AI is going to take your job but give you a much better one.

Yeah.

I'm curious on that front, what are you seeing the new jobs look like for the people who have really embraced this? How does their work look different today versus what it did, you know, 2 years ago?

Yeah, um, so, and this goes back to, I guess, some of the agent— agentic coding, um, things that we've talked about, is that— so I think we're moving from this, um, kind of age of, I'll call it specialization, where, you know, um, you know, designers did X and developers did Y, and, and then you need to have, you know, all these people kind of come together in order to kind of accomplish one particular kind of internal objective goal or project that you might have. What we're seeing now is that, um, curious and determined people can do a lot more than they were, uh, previously capable of, right? So now you can say, oh well, you know, Susie in HR wants to automate this particular kind of process, um, and it's not— she's not gonna rewrite Workday, which we use. What she's going to say is like, okay, given the tools that I have in the— it's like, I don't even know anything about APIs, but I'm going to go— it's like a new employee onboarding thing, whatever. I'm going to build up this kind of internal tool that will take X hours out of, uh, and human labor out of this kind of process for every time we hire a new employee, which happens, you know, often enough. It's those kinds of things. And so I think what happens over time is that anyone, uh, my— I'll say this with conviction because I believe it with conviction— is that if you are determined and curious enough to automate your way out of your current job, it's like, oh, I used to do— I spent 80% of my time you are going to invariably get promoted.

Yes.

That's what's going to happen, right? Like, within our organization, someone came to me tomorrow, it's like, Dharmesh, you know, I did this thing or whatever, it's like, and I thought it was just a waste of time, so I kind of, you know, built the script and used whatever and did this. It's like, great, we'll find something bigger and better for you to do. Let's find bigger problems for you to go solve and tell me how I can help, right? That's what would happen. Any reasonable, rational organization, which I think most organizations are, That's what would happen. It's like, oh, well, thank you for like automating that. And now we're just going to fire you. No, that would be idiotic, right? Like, this is a high-leverage person. This is, this is what we dream about is these kinds of people.

Yeah, I will speak confidently now for listeners out there. If you have been fired because you automated your way out of a job, drop me an email, drop Dharmesh an email. We are happy to take a look and add you to our organizations. I would, I would take that 10 times out of 10 and twice on Sundays.

Yep.

I'm curious, are there specific roles or functions that you're seeing transformed the fastest with AI? Like, are there, are there places that look the most different now inside of HubSpot versus maybe 2 years ago?

Marketing is, is way up there, um, because the marketing has this, uh, kind of rhythm to it, especially like content creation and analytics and all these things. Uh, so our marketers— and we happen to have a particularly forward-thinking marketing leadership at HubSpot with our CMO and then Kieran Flanagan, who's one of our VPs. Well, yeah, he was Zapier, so you know him. Anyway, so marketing has changed a lot and they're experimenting on the edges and doing super cool things in terms of repurposing content and building the flows and things like that and automating big pieces. Engineering has widely, but What we're finding now, I think the tools have gotten just good enough where across pretty much all— I don't think you would find a single team or department within HubSpot now that they're not using AI in some way. I just don't think such teams really exist. Yeah, and some of it is they're off buying tools if tools are available. Otherwise, we have an internal dev team that does internal dev apps, or they go off and You know, build their own thing to automate. Yeah.

What— so, you know, it's interesting you brought up marketing because I do agree that that is a function I feel like that has lots of opportunity. Though I've spent a lot of time with marketing teams over the last year and I feel like as a function it's going slower than I would have thought. I'm curious, like, why do you think that is? And like, what advice do you have for marketing functions that actually want to go do this but maybe are just not Yeah, so this is personal opinion. Sure.

My thought, and we've talked about this, we as an industry have talked about kind of blurring of lines between marketing and sales. I think over the fullness of time, marketing looks more and more like sales, and sales looks more and more like marketing. And so I think what happens in the world of marketing is, okay, well, let's not stop at the lead gen or whatever. Like, how far can we carry this ball from the point that we got someone's interest through whatever mechanism we did, brought them to the top of the funnel, And how can we kind of take them further, as far down the funnel as possible, without any friction in the process? I think that's where it's all headed. Now, we as an industry, and HubSpot gets both the credit and the blame for this, content was a big part of marketing for a long, long time. That's becoming less dominant a factor now because of the switches to how SEO and search engines work, and we're shifting. But those people are not going away, the intent, right? So yes, you may not go to Google to do your search. You may go to Perplexity or ChatGPT to do your search now. But so now, but we still, those people buying those products and looking for those services are still doing those things. They're just coming at it from a different angle. And so now the marketing has to sort of figure out how do we increase the odds? And that's the AEO stuff is like, how do we show up in the citations or whatever? But then it's like, okay, well, how do we update, you know, our websites, our collateral, all of it so it serves this thing that's probably deeper in the funnel? Like, it's not the web pages, it's actually answering the questions they're likely to have now that they're further along. And all that stuff is just new, and it would not have been possible. It's a good thing, is, you know, it's like AI giveth, AI taketh away. So it has taken away in terms of top-of-funnel traffic, but what it, what it gives us is the ability to personalize the information that we make available At the point the person does show up on our website or in our funnel, because we have the ability now to create highly personalized, highly relevant, and it's not just about producing the English words. It could be a video-generated screencast demo. It's— yeah.

Yeah, it's pretty wild. Folks show up now and the intent is much, much higher. They're much more serious and way further along. And to your point about marketing being more like sales and sales being more like marketing, Both of them, I feel like, are a lot more like, you know, product experts at the end of the day, where it's like you actually have to teach and educate people because your customers are coming in already that much smarter on these topics.

Yeah.

And so if you're showing up, you know, and you're like, oh, I need to go hand you off to another person who actually knows this stuff. And so we're noticing how that's changing the expectation level of what we have in our sales org and our marketing org, where there's just a level of like expertise that's required On these things where before you could maybe get by on like, I, oh, I'm really good at running a particular sales process, or I know this channel inside of marketing really well. And it's like, okay, you still need those things, but you also have and, and now, which is challenging. But then coming back to your point, AI taketh and AI giveth away. Well, the, the place where it gives is it actually makes it easier to get smarter on these things.

Yeah. And yeah, the thing I'm like most excited about, and I'll take it back to an earlier part of the conversation as well, so imagine this hypothetical prospect for one of our companies and they're out there doing the Perplexity ChatGPT thing. They do the search, they do the due diligence. So they're super smart, super clueful. They're really, really deep. What we can do now once they come into the funnel is like, oh, well, we're, we sort of know, you know, we have some amount of information on their company. It's like, okay, well, here is like historically what we would've done is said, oh, we'll find matching case studies of customers that look sort of like you. Mm-hmm. Maybe now instead of just surfacing information that already exists on our website, in our case study library, whatever it is, we'll actually generate a custom, like, here is you and what it might look like to buy our product. By the way, here's a screencast of you like doing your website in HubSpot or whatever. Oh, and by the way, we pulled your company colors and your fonts and your typefaces and your brand style, all of that which we found. And it's all embedded and it's a literally a custom demo that would've taken someone 4 months. to try and build. We've already done it. By the way, here's a project plan for what your migration would look like because we know that you use this, this, this, and this. These 3 integrations we have covered. This one you're gonna use Zapier for. Oh, by the way, here's like all of it. So imagine if you had, you know, an infinite number of super clueful people that knew both them and your company. It's like, that's all that's possible now, right?

Mm-hmm.

It's not necessarily super easy, but it's gonna get easier over time, but it's actually possible now.

Yeah, I think this kind of comes back to the simulation thing. It's just, yeah, it's like you could actually just— this is what it would look like to use this.

This is what it would look like.

Yeah, it's incredible. Uh, Dharmesh, thanks for joining the show. We're stoked to have you here. For folks who aren't already following you online, which I know they probably all are, where can folks find you?

I'm @dharmesh everywhere— LinkedIn, Twitter, pick your place of choice.

So yeah, awesome.

Almost impossible to avoid.

Dharmesh, thank you so much. Uh, enjoyed having you on the show. And for those who aren't listeners, I hope you like, subscribe, and we'll catch you next time.

Thanks, Wade. Good to see you.