HubSpot, Inc. ($HUBS)

Well, we will kick it off. And for those that are listening in, really excited to have Yamini Rangan with us from HubSpot. Thanks very much for being here. I am ecstatic to have a conversation about what's going on at HubSpot with -- around AI. We were just talking about -- there's so much -- the velocity and the scope of this cycle is pretty remarkable.

I don't know, actually, we -- before I get into AI, why don't we just recap the quarter maybe really quickly with you. It was a really good quarter, 18% growth, net adds were above 10,000. Can you just talk about sort of what's driving sort of the business right now? Because I know everybody wants to just get to the AI upfront, but maybe we'll cycle back then jump forward.

Yes. Well, thanks a lot for having me. I really appreciate you taking the time. And yes, we'll start with Q1. Q1, it was a good quarter. From a revenue perspective, we beat guidance. From an operating margin perspective, we beat that, as well as like net adds. And look, we are seeing the same trends and patterns that we have seen for many quarters now, which is that a lot of our customers, our customers are in the 2 to 2,000 to employee range. They are consolidating on fewer customer solutions. And we've seen this now, and we continue to see this going into this year, where if you look at a typical mid-market company with like 1,000 employees, they have maybe a bit of a homegrown system, they might have a legacy system or they might have a point solution. And they cannot get into AI without actually consolidating, bringing all of those insights into the same place. And so I think like we've basically seen that same trend continue, and that strength is both upmarket as well as multi-hub. And again, the trend that we are seeing is that you can't just have marketing and sales and service in very siloed locations. You just have to have all of the data to start leveraging AI. So we continue to see consolidation and multi-hub adoption. And then over -- what our customers are looking at is where do they get started in terms of their AI journey and what is that like -- road map look like. And we begin to see the emerging levers that we have talked about in the business, which is core seats and credits. And the combination is what we saw in Q1. And the combination of all of that gives us this very clear view of what the growth levers are and how we are continuing to drive growth.

Okay. And can you just talk about on the pricing and packaging? You guys have made some changes. Just obviously, you're trying to drive adoption. Customers are asking you, let's see the value. Just talk maybe holistically about how you're viewing that right now and some of the changes you made and maybe why?

Yes. And Kirk, I think like we were just talking about this. If you think about AI, the fundamental shift within AI is that before AI, as a SaaS provider, you provided access to your technology so that users could then deploy it, point, click, navigate, get work done. And therefore, the pricing model that we had in the SaaS world was pretty simple. You either get a subscription fee to your technology or you provide access to a specific role like sales or service, and that came in seats. The fundamental shift with AI is that you are now delivering work, and the work looks like either a ticket that is resolved or it could be a qualified lead that you deliver or it could be a target prospect that you're delivering. And so you're delivering something completely different, which is not tied to the value of the seat that someone has. And therefore, it means we have to have credits. And so you asked, like we're making some changes. I think the industry is making changes. If you ask 20 different vendors what their pricing strategy is, I think the strategy is experimentation. It's not like, oh, we figured it out and we know exactly what we're doing. And so where we are in the process is that we believe that the future of pricing is still going to be hybrid. There will be a set of just table stakes AI capabilities that we will deliver per seat. And that is going to always be the case. And then there will be work that we are delivering that is not based on seat, and that will be credit-based. What we did in Q2 is this. We took our agents -- not all AI features because we have hundreds of AI features. We specifically took agents, which is Prospecting Agent, Customer Agent, and we made those outcome-based because we believe that AI should always be about not outputs, it should be based on outcomes. And so we took our Customer Agent and we said we were going to cut the price, but we are also going to make it based on the number of resolutions. So Customer Agent resolves tickets. It used to be based on the number of conversations. We moved it to resolutions, and we cut the price in half. Why did we do that? Because we truly believe that we are in the very early stages of technology adoption and we want to remove friction. We want people to say, "Oh, this is like a no-brainer for us," and they start adopting. The same thing with Prospecting Agent. Now the combination of the moves that we made will have short-term negative headwinds in terms of net new ARR, which we have been explicit about. But we also believe that these are the kinds of moves that like actually provide the right foundation for consistent adoption as we look at the next few quarters, and that's why we did it.

Okay. That makes tons of sense. How does this impact the sales motion for you all? I think driving adoption is just key for anybody. We were talking about earlier, the only way to push back on like existential narratives is adoption and value being delivered. So what's the sales motion that now goes along with sort of a more almost value-based approach to pricing? What are you changing on that? What are you leaving alone?

I love the way you framed it, which is it is value-based sales, right? It is much more about value-based sales. So a couple of things that we're doing. The first thing is that customers want proof of value. Because, again, if you are buying a piece of application, that application needs to be deployed. You train your salespeople. Those salespeople use the product, and then they build pipeline. It's -- that was the old SaaS world. Right now, you're actually deploying agents. And the week 1 that it's deployed, it better write e-mails like your best sales rep. It better qualify leads like your best BDR. And so the difference is that customers now evaluating agents want proof of value that it actually works in their environment given their conditions and their value proposition they need to do. So what we did, again, we saw this in Q1, and we provided trial periods. And during the trial periods, we activate the agents, we make it work within their environment. They can play with it, tweak with it, understand the kinds of outputs that these agents are delivering, and that gives them a lot more confidence that they can scale with it. So that's the change. But if you step back, Kirk -- and you've known this industry for a very long time. I think what you're just saying is that the sales motion itself is changing. It used to be that salespeople go, sell, demo, product, and then they close the deal. And then comes the implementation, and GSIs, partners implement it. Then throw it over to customer success, and customer success will come 6 months later and say, what have you adopted and kind of like improve that model. That model is changing real time. You -- during sales, you have to prove the value. The first week that you implement, you have to show the value. By week 4, that value better have compounded. Otherwise, why pay for tokens, why pay for credits, why pay for any of the things that we're delivering? And so one of the other things that we did in Q1 to set the stage for this is like we pulled our sales and customer success teams together under one leader. And what that enables us to do is to make the handoffs easier, think about incentives that drive the value sales motion, helps set up like the whole organization in terms of delivering value, which is a completely different motion. We're going to look back and say, wow, like the AI didn't just change the technology and the platform. It's actually having pretty significant changes in terms of how you reach customers and how you communicate the value and how you showcase that value. And that is what is happening, and we are like leaning into it. So it feels like a set of changes that we are aggressively pushing, which we believe will have medium-term, long-term impact, but will cause a bit of a short-term headwind for us.

Yes. And when you think about customers getting value, I mean, to some degree, their data has to be in a place where they can actually use it correctly or drive the agents to get the outcomes they want and see the value. So I guess when you're talking to your sales and your go-to-market about this, is it really about, look, you have to understand where the customer is at from a data perspective? If they're not there, let's not push something that's going to have suboptimal value for them in the near term? Is that something you feel like you're -- like when you look at your customer base, you've sort of segmented it so that you understand like who's ready to go, who has more care and feeding to do before they even get to the point where they can take advantage of agentic? Because to some degree, you're not -- like if their data is all over the place, no matter how great your agents are, it's not going to have the outcome they want.

Absolutely. 100%. And the two of us, we were talking about this. Like if you think about the classic adoption curve, you have innovators, and these innovators are okay if the technology is not there. They can string it together. They can work on weekends, they can Claude Code. And those innovators also happen to be on X and LinkedIn, being very loud about what they built over the weekend. That tends to be 2% of the entire customer base. Then you have the early adopters of technology. And early adopters of technology will turn an agent on, will work with what they have, and they will also give us feedback and iterate with us. But then there is actually the early majority and late majority. The late majority folks don't have the data ready, don't have -- they're still in fragmented applications. And AI is not a magic wand that you can wave and say all of the data get to the same place. And so for those folks that are somewhat later in the cycle, we talk to them about data readiness. We talk to them about getting a consistent view of all of their customer information, and that is absolutely critical. Without that, they cannot even get on AI. So I think what is even going to happen is that even if you're today not ready for AI, you know that you have to get ready. And the first step that they're taking is data readiness and making sure that they have the right foundation. So we talk to a lot of customers there. And then for the customers who are ready and willing to kind of like adopt, then we have a very clear road map of where to start with agentic use cases, the ones we've talked about, Prospecting Agent, AEO, which we haven't talked about, as well as Customer Agent, and we get them on that. So I think what is different this cycle is that it's moving faster, right? If you compare this to the cloud adoption cycle. 1999 is when like a lot of these companies like were found and formed. 2009, we were still having conversations about like cloud adoption and getting CIOs comfortable of moving from data centers into like public cloud. I think what's going to happen this cycle is that it's going faster. And so the conversations are really based on the maturity of the organization and where they are in the AI curve, but we are actually working with all of our 300,000 customers to help them navigate the big platform shift.

Yes. And when you think about your customer base, obviously, more -- a little bit more mid-market. If you're talking to a big enterprise, like we got to get our data in order, that could be like a 2-year project or take forever. Can your customers get there if they really focus on it in 6 months, 9 months so that late majority starts to become a targeted customer for you from an agentic perspective within 12, 18 months? Whereas I think if you're talking about big enterprises, that data sort of unification could take a long time.

You're talking about it in the right way. So the -- one of the fundamental reasons why HubSpot didn't acquire through the growth is because when we went from marketing automation to sales and service, we said the fundamental thing is that the customer data has to be in a single record. And so we made a choice a while ago that we were going to invest in platform primitives to help us build faster and innovate faster. But the resultant benefit for customers is that when they adopt HubSpot, they don't have to do data unification across marketing, sales and service. And this is also why we've seen consistent multi-hub expansion over many years. It's because you now have data in -- at least the core customer data in a single place, and that helps. Now if you're not already a HubSpot customer, it certainly is not like a year implementation. It mostly is weeks and months before they can implement, get the data ready in one place and then begin to adopt. So the cycle is much faster. I do think that this provides an impetus even for late majority from the previous cycle to really start getting their data ready so that they can be in a position to adopt AI. And we're definitely seeing and having those conversations with customers now.

Okay. And we talked about it a little bit earlier, but your view on pricing being a little bit of a mix or a hybrid of seats and consumption. Is there a right answer to that longer term for your customer base? Because it's not where you might want to be. It's really where the customers -- you want to meet the customers where they're at. Is there -- in 2 years or 3 years, like is there a way you think about that to be like, ah, it might be like 50-50? Or is it just -- it really just depends on adoption of these agents and how fast it goes?

I think that even 2 years, 3 years, who knows, 5 years, but 3 years out, pricing needs to be hybrid, and it needs to have a seat-based component and a credit and usage and outcome-based component. The reason is that there are so many things within a customer platform like HubSpot that is just table stakes. That just comes with the core platform, and we would not want to nickel and dime or make that all usage-based. So we'll have core seats that every go-to-market employee needs or a specialized sales or a service seat that a specialized team might need and provide a lot of the functionality as part of that. And then we see what we deliver will become more usage-based. I actually don't know the exact split between seats and credit, but we will think that -- the way we think about the growth levers is that there will be core growth levers. The emerging growth levers are core seats, which have a lot of data and AI value, and credits, which will be what we deliver through work, and it will continue to kind of build in that. Again, a lot of like seat compression and how does this actually impact how you think about the pricing. I think in the longer term, there will be some level of seat compression, but that will come from delivering even more value. Like if you take a look at our business, the teams that we have actually held in a headcount are support, for example, where we have deployed Customer Agent to augment our Tier 1. But then we are delivering much better like value through the same individuals. And so I think in the fullness of time, there will be some seat compression, but the narrative of it's going to happen tomorrow it's going to happen in the next quarter, the timing might be off. I think it's like we believe that there is this hybrid world of seats plus credits.

Okay. That makes a ton of sense. Let's talk about AEO a little bit. You're seeing some benefit from it at your top of funnel already. Talk to us about how this sort of helps the broader conversation with customers, what the sort of opportunity is for that and what you all have learned?

Yes, absolutely. Look, there's a huge shift that is happening. People are not searching for information, clicking on blue links, coming into website. They are, in fact, going to LLMs and asking questions, but those questions tend to be much deeper. The average search was like 4 to 5 words, and the average question or prompt within LLM is like 23. So it's a deeper search. It's much more intentional. And when you get a very targeted answer with a citation, then people convert more. So that's the broader shift that's happening. If you look within HubSpot's customer base this year, the content leads or the leads that you got from searching to your website, that's down 27%. That's like big. Now we've seen this because we were even more inbound-focused. We've seen this for the last 2 years. And certainly, from a HubSpot perspective, we have gone through a diversification of lead sources from content leads to social. We are much bigger on YouTube as well as podcasts as well as newsletters. So we've diversified the sources and we have experimented with AEO. So it's not like sometimes people confuse that, oh, SEO is getting replaced by AEO. SEO is getting replaced by a diversification strategy across multiple channels where your content needs to show up. So content is important, but where it shows up is much more diversified. And that's what we've seen internally. And we've built all of that into a playbook called the Loop that we launched at our conference last year. We've built that into our products. And at Spring Spotlight, which was in April for us, we launched an AEO solution. And that AEO solution basically can give you visibility. So you want to know how your brand shows up and how does that show up compared to your competitors, the share of voice, it can do that. The second thing it will do is to say, look at your share of voice and say your share of voice today is 62%. In order for you to get it to 70%, here are 5 actions that you need to take from a content perspective. And where we are going with it is that you can seamlessly take those actions. And so at some point, in the next couple of years, if a B2B marketer does not have an AEO strategy, they're going to be left behind. So we do think that this is early and there's just a lot that is happening in the industry, but it's an exceptionally important part of your marketing channel strategy, and that's why we are leading the market with our AEO solution.

And what's the -- remind me, I probably know this, but what's the monetization strategy around that? Is this become just a central part of sort of the Marketing Hub in general? Or -- and this is sort of becomes more -- something that just brings more people in, you're helping them evolve? Or is there an opportunity to sort of take price a little bit as part of this as they get better, frankly, hopefully, better outcomes to go around it?

It's two parts. The first is that we launched a stand-alone solution, which means that if you don't have anything else and you don't want any other marketing solution from HubSpot, you can start with it. And the good news with that is it will start showing your share of voice and give you recommendations and you can just adopt it to get going. And we think of it as a front door if someone wants to look at this and they can start with that. Now if you want to take content actions and you want to be able to drive much better multichannel content, then you do need Marketing Hub. So you would buy stand-alone and then upgrade into Marketing Hub Pro. That's the path. So it is a stand-alone solution. The second way we think about it is it comes included within Marketing Hub Pro as well as Enterprise, which means it increases the value of our current solutions. But what we are finding early adopters do is that the base case, it will come with 25 prompts in 3 LLMs. Now if you look at our AEO strategy and how we've been able to get our visibility up, we have hundreds of prompts that we are tracking. And those hundreds of prompts give very different share of voice metrics and very different recommendations. And so as Marketing Hub Pro+ customers begin to use AEO solutions, they are going to need more prompts and more visibility, and that will also consume credits. That's the same mechanism of included credits plus packs that you can buy. And so that's the second form of monetization. So that's -- you buy it stand-alone or you use it within our current products, and then you continue to use more of the prompts, then it will consume credits.

But you're aligned very closely with value to this?

Absolutely. I mean, I think that's -- it almost goes without saying, but in the prior cycle of SaaS, you were providing access to a technology that could add value. In AI, you're providing value. So everything has to be based on products delivering outcomes. And then you can monetize it, but you have to start with a very clear view of how you're delivering outcomes and value.

Yes. Nice thing about AI is it's very binary. As we were talking earlier, if you -- either deliver value, people will use it. If you don't, they don't have to turn on. It's -- I think it's very -- it's a positive over the long term for a lot of people that are delivering value.

Yes. Absolutely.

Obviously, a question I'm sure you get all the time is -- talk to us about sort of the interplay between the HubSpot platform and the model intelligence underneath it that will help power some of the answers and frankly, be sort of the brain behind sort of the orchestration layer that you have. What's your view on sort of the interplay between you all in Anthropic or an OpenAI? I think there's a better together scenario for most of these -- for these companies and you all, but can you just explain that a little bit or go through the way you think about it?

Yes. I mean, look, you know the narrative better than I do. It's like there is a very, very powerful technology which is LLMs, but the narrative has been like it wipes out all categories of software. And from the beginning, we've been saying that we think that LLMs and HubSpot are complementary. And now 3.5 years into it, our conviction is even more deep in terms of how complementary they are. And you can look at this in multiple ways. The first way is like LLMs are powerful technology, but completely trained on publicly available information. And we have private information about a company, about marketing, about sales and about service. And so you have to make those two work. We can take the capabilities of the LLM, apply the context that we have, which is what do you do within marketing, sales and service, and then train the output so that it can be like much more targeted. So that's like one. The second is everything that you get from LLMs are probabilistic. And everything that we need to do is a combination of probabilistic and deterministic. I just did my sales forecast. If they gave me a probabilistic answer, I would not be happy with a forecast that is probabilistic. I need a deterministic answer. I can go through that for many, many things within sales and marketing. So you need the combination of something that's probabilistic with deterministic. The third thing I would say is LLMs are today, very, very single player. I can write an e-mail, I can write a blog. But if it's a marketing team that's working on a multi-language, multi-region campaign, it is a multiplayer mode. It's not a single player mode. And we bring that technology and make it available for multiple players. And so for many of those reasons, we take a powerful technology and we apply it to the domain of marketing, sales and service and really help our customers get to outcomes. And that's why I think that we'll continue to work. And one of the leading indicators we see, Kirk, is that we were one of the first ones to launch connectors, connectors with ChatGPT, with Claude, as well as with Gemini. And what we are finding in the patterns of users that are using those connectors is that they ask a lot of questions, but then they take a lot more actions within HubSpot. And the level of engagement in users with connectors is actually going up, which is a positive, which shows that there is like value in both. And I think that that's the pattern that we are beginning to see consistently.

And I think you guys have always been -- you've always been fans of sort of an open framework that you're going to have agents that want to come in, access some data on HubSpot through MCP or APIs, whatever it might be. How do you think about -- what's sort of the fair monetization strategy on that front? And I think a lot of companies are still figuring this out, frankly. But I guess if a CFO wants to go get data, HubSpot to make a decision from a CFO perspective, you might not have a sales you might not have a Marketing Hub or a Sales Hub for them. If he want to bring information from HubSpot up to a dashboard to do his job, how do you think about the value you're bringing to that scenario? Because I think a lot of the incremental new use cases that are being built on Claude that you see are things that aren't necessarily -- they're not trying to replicate or replace what you are bringing. It's sort of a net new workflow. And I'm sure you want to be a part of that, right?

I mean, multiple thoughts there. First, we always care about being an open ecosystem, and that's been the posture from day 1. The second thought is that, again, contrasting between the SaaS world and the AI world. In SaaS, we delivered great user experience and great developer experience. So if you're a salesperson, you log into HubSpot, you had great experience on the graphical interface. If you are a developer, you had APIs to basically connect with HubSpot, extend, integrate all of those. I think what is different with AI is that now we need to have agent experience. That means an agent can sit on top of HubSpot, either talk to HubSpot using MCP or talk to HubSpot with APIs or talk to HubSpot with CLIs. And why CLI? It's not like anything new. It's just that agentic coding was trained on command line interface. So it actually does a more efficient job with command line interface than actually APIs and other things. So the fundamentals are the same. It's like better user experience, better developer experience and now better agent experience. So then the question becomes, how do you monetize it? We think about it in twofold. One is that we deliver data. And we've always had APIs that deliver data. We monitor the API usage. And unless you are kind of like bulk extracting every minute in real time, you don't trip that. And so if a CFO wants to really extract data, go ahead, do it. We think that, that is not going to be the place of value as AI begins to develop. Here's why. You can, from HubSpot, basically say, I want all of my company information and I want to compute which companies have a propensity to close deals this month. You'll take company data, deal data, sales conversation data, you'll compute all of that outside of HubSpot, and you'll come up with a metric, which is called propensity to close. That's what you do with our data. Go ahead, do it. It will be inefficient, and you'll burn a lot of tokens and inference [ cost ] doing that. The better way is that you can come to HubSpot and say give me accounts that are -- can have a high propensity to close, give me accounts that are high propensity to churn. And we would have done all of that because we have an insight layer, what is our growth context, and you can extract that and make a better decision as a CFO. What are you going to do as a CFO? You shouldn't be doing the first one. You should be doing the second one. And the way we think about our API monetization is if you take data, go ahead, take it. And you'll get charged the same way, but we will be able to extract more value if we deliver more value through the insights. And that's what AI will do, and that's what our strategy is becoming. It's like 2-part, data API and insights API. And when we deliver the insights API in high volumes, we're going to be able to extract more. And that becomes a newer way in which we can monetize the agents interacting with us. And that's how we're thinking about it. Look, this is all evolving. We're thinking deeply. We're experimenting. We're investing in all the right places, and then we'll see how this develops.

Okay. All right. Yes, please.

Can I throw off? How do you think about kind of the headless approach, right? I guess, subquestion to what you were talking about. You're now opening your existing product [indiscernible] you see in the future, you're leading the headless program, right? And maybe the problem is [indiscernible], so kind of flipping it?

Yes. I mean, look, headless is exactly what I described, which is an agent that can now have an experience running on top of HubSpot either through MCP or through API or through CLI. That is exactly what it is. Now the question you're asking is, are you okay with that? I think absolutely. Like I said, we think that the future is our users using our agents or developers or agents developing on top of HubSpot and building, extending other agents. And we'll be able to monetize it either as data API or insights API and access to our platform. And so our job is to deliver just best-in-class first-party agents, the agents that are in core marketing and core sales and core support that we have the domain expertise to, like our Customer Agent, Prospecting Agent, we want to deliver best-in-class agents there. And then we want to deliver a platform where other agents can sit on top and build additional automation, additional workflows, ability to extend to other agents. And when we do that, we have -- our thought process is like we'll monetize it in a very different manner. And so there will be a balance of this. And again, think about the customer adoption journey that I talked about. If you are on the bleeding edge, then you're using agents and you're building on top of platforms, and we get to monetize it differently. If you are in the later majority, then you're not doing the headless development. What you're doing is actually adopting first-party agents that have been tested over and over again. And I think that's our strategy. That's why the simplest way that we think about it is like agents run HubSpot and agents run on HubSpot, which is like first-party as well as the headless ecosystem approach that we're just talking about.

Can you talk about -- actually, I'm curious, how does that sort of concept start to impact like the thought process around like Starter customers? Meaning, if you think about it, you could start having Starter customers just sort of have an agent that they build, but then plug in a hub. Does it actually open the aperture? I'm kind of curious, you guys have done an amazing job bringing small customers in and then helping them grow with you. Does AI, I guess, change that playbook at all at the low end of the market in terms of sort of that adoption? You guys have had lower-priced products that have brought them in, given them quick value. Just talk about that playbook, I guess, in a more agentic world.

Yes. I think that it's very -- AI is making more go-to-market folks builders. That's really the shift that we are beginning to see, right, which means that if you adopted HubSpot as a smaller team, a 5-person team and you started with marketing automation, you started with sales, the fundamentals of what we've provided are workflow sequences. Now it's like agentic automation and extensions, right? And you can continue to use those and grow. I still think that there is a certain level of scale that you get to where the Starter is not going to have the scale that is required and you need to like upgrade, and that's been the path for us. We've had a play of bringing -- having a very compelling free product and then bringing it into Starter, converting the Starter volume into Pro. And I think what AI is going to allow us to do is that you can have a much more smoother curve because you don't have to jump from Starter to Pro. You can actually use a lot of the agentic capabilities, pay credits and kind of like extend the power of the platform before you kind of like jump to Pro. But I think what it's also doing is that your ability to hone in on a use case and deliver value in minutes is really critical. And what we have been talking about, which is SaaS, much more about delivering a product that people begin to use and then get value. AI deliver value immediately, and that's the transformation that's happening within our Starter product as well, how quickly can you get to starter use case with an agent and begin to get value.

You guys are obviously using AI a ton internally, showing some good operating leverage. What's the thought process there? You all have always sort of really focused on and invested in R&D. You've always been a heavy R&D company. Can you kind of do both now? Is this sort of the perfect playbook for you to some degree where you can invest in AI, but also continue to get that kind of operating leverage out of AI internally?

Absolutely. And look, I think that we got on to this internal transformation as early as we got on to kind of like the product transformation and delivering value for customers. And what is not very obvious is that the way we build an AI product is like fundamentally different from the way you build a SaaS product. And in order to do that, what the journey that we've gone through is started with copilots, GitHub, that kind of stuff [indiscernible] coding. Even with the technology and the level of capabilities that we see within the market, it hits a limit very quickly because it's not optimized to your developer environment. And if a HubSpot developer needs to build with the agentic coding, they need our libraries, our skills, our ability to test code and review code and get it through that process. And so what we have done is we've optimized it. The industry now calls it like harness. Everything is like a cool harness. But what we've done is containerized it, optimized it for the internal developers. And so that gives us a lot more in terms of the pace of innovation. And we saw from the time that we actually started leveraging Cursor and Claude Code and the jump between that and our containerized optimized environment for HubSpot developers, the productivity jump is like massive. And that is because of all the things that I talked about. And what we believe in is like we have now an agentic execution platform internally where building the next agent and the next agent becomes like much easier and faster because they are pulling from the same libraries, from the same skills, from the same optimized environment. And as the number of skills that our developers build increases, the ability to build an agent on top of that becomes easier. And for our customers, it results in a very consistent interface with agents, agent training. Like if you train 5 agents, you can see it in the same place rather than adopting 5 different agents from 5 different companies, training it and looking at the context differently. So I think like the internal transformation that we have done in the R&D will pay off in multiple years, but that is the leverage. The other thing that you asked about is like how does that change our own operating leverage. And on the go-to-market side, any and every one of this, we've been like bleeding edge adopters. We have trained, we've experimented and we have scaled with that. And we are seeing certainly, areas where we are getting efficient. I talked a little bit earlier about Tier 1 support, where we've not hired anybody since 2024. And we have taken that and used that headcount in other places to get like leverage. And that's one of the reasons why last year at Analyst Day, we provided a midterm target for our operating margin of 20% to 22%, and we reached that a year earlier, and we've increased our operating guidance this year. And I think when you come to our conference this September, we'll provide you the midterm and long term. And we feel that even though we're spending a lot in terms of agentic coding and our -- you'll see some pressure in gross margin, we feel confident that we can continue to drive operating leverage as a business. And that's because of how much we have done internal adoption and transformation with AI and how we continue to do that.

Okay. Any last questions? I have one more.

I guess part of the bad thesis behind the SaaS [indiscernible] is customers will start [ backloading ] their own solutions. Are you able to track when you have a customer churn? [indiscernible] Are you able to see like what percentage of your churned customers are going [ in-house ]? Is it something you're tracking or [indiscernible]?

Absolutely. And no, this is -- again, you cannot prove otherwise until it gets proven by itself, right? And so -- but we track our customers, why they buy, why they don't buy, and vibe coding is not coming up in conversations. I talk to customers every day of the week, and no one is saying, I'm just going to start like vibe coding. I'll tell you the 3 customer conversations I had today. One was a transportation company, which is like all transportation logistics. Another one was a homebuilder, and the third was a skin clinic. None of them have the capabilities, nor do they have the interest to vibe code. That's number one. The second thing is vibe coding is all about coding becoming easier, but not integrations. We have integrations with 2,000 other providers. So you vibe code CRM, and then who's going to connect it with the ERP, with the accounting system, with the project management system, with name your next tool? No one has that capability to build all of those integrations. And then that vibe coded developer now gets like an even better offer from another person and leaves, and then it's all gone. And then finally, the vibe coding person has not enough domain expertise. We know what's happening in AEO. I can tell you that our best people in AEO will learn so much this month compared to the last 6 months because the industry and the domain is changing so rapidly. So how is the vibe coding person going to be able to do all of that? So look, I know the narrative, but coding has gotten easier. Delivering value out of applications is still very, very challenging, and we don't see that within our customers.

[indiscernible]?

I mean, I don't know if that...

[indiscernible]?

Yes, it's minimal. It doesn't register is the way I would say it.

And no change over the last 3 years?

No.

Do you think the lack of that is also a factor of -- like you have so many B2B customers, meaning these aren't sort of B2C companies, these are -- if you're a law firm delivering that -- like the time it takes to go vibe code something and if you deliver something to your own customers that doesn't work, there's reputational risk for a lot of your customers?

Certainly. I mean this week, I had a conversation with a midsized bank. And I was talking to their CRO and their CIO. And the midsized bank is like, first of all, vibe coding did not count. But most of what their conversation was how can we deliver trusted output that we can scale and grow with and how can we make sure that our data is safe and it's governed and it passes all of the security compliance and SOC. And how do we make sure that we are future-proofing this. Even if they have the capabilities, like a midsized bank is not going to like want to vibe code it on the side and custom build it. I mean, look, throughout the history of technology innovations, there's always this moment of like custom building can replace something else. It's turned out that it is harder than just custom coding and building. There's a lot more to going beyond that.

Just because you can do it, doesn't mean you should in many cases.

Absolutely.

So we'll leave it there. Thank you very much for your time. Appreciate it, Yamini.

Thank you.