AI GTM STAGE

As a courtesy to our speakers and fellow attendees, please take a moment and silence your devices. Please give a warm welcome to the CRO of AI from Stripe, Maya Josapakhvili.

Hello.

Hi, everybody.

Maya here, CRO of AI for Stripe. And 20 years ago, I started a company called Urban Escapes. We did whitewater rafting, kayaking, skydiving. Now, 20 years ago, this is pre-WeWork, so we thought we were being very clever by having, renting out— are the slides going? They're not going. We had really cool skydiving and whitewater rafting pictures up. Do you guys back— BRB. No slides. I promise they were very good whitewater rafting pictures.

We got them?

We got them? Yeah, but my talking points aren't there.

Just go with the flow.

There you go.

Two minutes.

Can you zoom?

Is it synced in?

Can you cut her mic for a second?

Now we need the mic. Good. Start over. Hi, I'm Maya. No, I'm just kidding. Okay, so we would take people kayaking, whitewater rafting, skydiving. That is me in the white shirt when my prefrontal cortex was not super developed. And then this is our conference room. Okay. So anyway, a long time ago, things were different. What I'm going to talk to you about now is how differently I would run a company today if I was starting one and what we're seeing in our data from all the AI companies. So at Stripe, we work with the fastest growing AI companies in the world, which means we have a front row seat to what's working, what's not, and what the best companies companies in the world are doing to navigate this agentic era. And the headline is that the playbook for how to run a company is changing faster than any of us can wrap our heads around. So there's really 4 patterns we're seeing across these companies. Number 1, the top AI companies just build so much faster. And I think we're all feeling that. Uh, 2, They sell globally by default. 3, pricing is evolving faster than, again, any of us can wrap our heads around. And 4, they're adapting their go-to-market motions super fast. So building enterprise sales used to be this decade-long journey. Now it happens in year 1. So we're going to dig into each of these and talk a little bit about what we're seeing from the companies working with Stripe.

All right.

So we will start with the numbers. We saw the top AI companies in the world grow by 120% in 2025 and then in '26 by 175%. That's nearly tripling in a single year. So what we're seeing is growth is accelerating, not slowing down. And the top ones are even more mind-blowing. So I think you're going to hear from a few of these companies today, Lovable reported $100 million in revenue in 8 months, and then 8 months later they got to $400 million. Cursor hit a billion-dollar run rate in under 2 years, and then 3 months later they got to $2 billion. And Anthropic went from $1 billion to a $30 billion run rate in just 2 years. It's wild. And it's not just B2B. So our Link data shows that the consumer adoption of AI has doubled from under 6 million consumers to 14 million in just one year. And they're all spending more too. So the top Link buyers are now spending $371 on AI. That's up from $140 just a year ago. Just to put that into perspective, that's more than the average American spends on internet, streaming, and phone service combined, which means people aren't treating AI like a streaming service they might cancel. This is a critical utility for how they live their lives.

All right.

So we've got 175% year-on-year revenue growth, adoption doubling, billion-dollar run rates in months. What's behind all of it? We'll dig into the 4 patterns starting with speed. So when I first started Urban Escapes, getting to a prototype took months. Okay, and I'm not kidding here, but building a shopping cart was so hard that our first checkout, when you hit buy, told customers to mail a check to to my apartment in Brooklyn, and people did it. Now, just like you no longer need months of engineering to accept a payment, you don't need months of engineering to ship software. This chart shows when that really clicked. So as late as 2024, iOS app releases were declining, and then Agentic coding tools hit, and then they went up 24% month on month. And you see Delaware incorporations actually follow the same exact pattern. So you look at that huge spike when the tools went mainstream. Now, a lot of people are talking about, you know, non-technical founders vibe coding companies. But here's something really surprising. As startup creation has grown, the share of technical founders has actually gone up by 7 points in just 1 year. So a technical founder can now, with AI tools, do in days what used to take teams months. These aren't just side projects, these are real businesses. In fact, since we started embedding payments into developer platforms like Replit or Vercel, we've seen each monthly cohort go from idea to first charge faster and faster. And what you see is it now takes builders less than 6 weeks to get their first paying customer. And I can promise you, none of them are asking people to send checks to their house. Okay, so we're going to make this very practical. What does this mean for you? Now, many of you are already doing these things, but for those that aren't, here's the playbook that's emerging. First, obsess over developer productivity. Collapsing build times is a competitive advantage. The companies that are moving fastest treat it as a top priority. Second, build, sell, and iterate at the same time. That old linear playbook where you could build the product and then go sell it, it just doesn't work anymore. The best teams are doing all three in parallel. And third, when posed with the build versus buy question, the best teams are building and buying. So they're focusing their resources on what makes them differentiated and then buying the infrastructure and building blocks to help them that they don't have to build. Okay. Getting to market fast is one thing, but where you sell matters just as much. So with Urban Escapes, expansion was city by city. New York first, then Philly, Boston, D.C. And SaaS had the same logic. So you'd explore selling in other markets, But the focus was really on nailing your home market until you had true product-market fit at scale. And then you'd go find a GM in Dublin or London, and then you were international. That playbook just does not work anymore. And the data confirms it. So a few years ago, the fastest growing SaaS companies would reach 25 countries in year 1. And 50 by year 3. AI companies in 2025, 42 countries in year 1, and get this, 120 countries by year 3. That means Kazakhstan is now showing up on the list for many AI companies. Now, that was not on my bingo card. And it's not about selling to just one person in these countries, there's real revenue coming from them. Gamma is a really good example. It's an AI-powered slide and website builder. They reported $100 million in revenue in their first year. And although they're based in San Francisco, the majority of their revenue comes from outside the US. That's not the exception anymore. That's the rule. Across top AI companies, we see 48% of revenue comes from outside their home market. That is nearly half of every dollar. 3 years ago, that number was only 33%. So what that says to me is global revenue is not a bonus or just, you know, the extra. It's actually just the baseline. Now, let's take a look at which markets are spending the most on AI tools on Stripe. Not surprisingly, we're seeing strong spend in markets with high GDP like the US, Japan, and Germany. And then there are some countries emerging as hotspots like South Korea, Brazil, and India. The demand is there. So the question is, how do you actually capture it? By meeting your customers where they are. That means offering local currencies and local payment methods. All right, get this. Localized pricing drives 18% higher cross-border revenue. And when you have at least one local payment method added, we see more than 7% uplift. Now, with these stats, it's kind of crazy to not be doing this. So here's a good check for yourself. Put yourself in the shoes of a customer in Brazil. Can they pay in Reais with Pix? If your answer is not, of course, you're certainly leaving money on the table.

All right.

The playbook for this one. 3 ways to pressure test your global readiness. So first, localize prices. Make sure you have local payment methods in key markets. 2, automate your tax collection because no one wants to get in trouble with this. And 3, track revenue and conversion by country and obsess over optimizing it.

All right.

Pricing, my favorite topic. Who thinks about pricing? Who's nailed it? Yeah. The old models just are not working anymore because AI has upended everything we know. All right, so at Urban Escapes, pricing was very straightforward. Skydiving, $300. Rafting, $100. Same experience, same price, done. And the SaaS corollary is similar with seats, right? Same price for everyone. But as we all know, AI pricing is way more complicated. Value is elastic. So while everyone might use the same AI tool, the value they get and the experience they're having is vastly different. Let's take two users. So one is an engineer who sets up before bed a bunch of agents and wakes up in the morning with code ready to review. And the other is my mom, who last week told me very proudly, and I'm proud of her, that she just replaced Safari with ChatGPT on her phone. One is creating tremendous output. And the other has an upgraded search experience. The value and the costs of these two experiences are really, really different. And so the pricing has to be different as well. Now, this change in value isn't new. Every major technology shift has repriced software. For on-prem, you paid once for all the code written up until the moment that it was installed. Installed. And so a one-time fee really made sense. And then cloud changed everything, right? Because software became continuous and auto-updating. So recurring subscriptions made sense. With AI, the value is elastic. And so we have to— again, people get different value from the software and the cost is very elastic as well. And so we just have to think about it differently.

All right.

So now is when you want the playbook for how to price with AI, right? Good news, bad news. The bad news is we're still in early innings and the full gold standard has not been stamped yet. But the good news is that some patterns are emerging. So let's look at Replit. A developer tools company that's been around for almost a decade. When agentic coding took off, they pivoted their product, rethought their pricing, and hockey-sticked their growth. So they started with flat subscriptions, and then they layered in credits as AI became core to their product. Subscriptions for predictability, usage for value capture. And the result? They're now targeting a billion-dollar run rate. And they're not alone. Usage-based pricing is fundamental to how AI delivers value. So 2 in 3 companies on the Forbes AI 50 have some form of usage-based pricing. This was less than 50% last summer, and a majority of those are running this hybrid model. So subscriptions to anchor the relationship. Layered with credits that scale with value and usage. All right, knowing you need to be usage-based is one thing, but getting it right is really different. There's 3 steps to add to our playbook here. First, price in your customer's language. Developers think in tokens, enterprises in seats, but they're increasingly getting comfortable with consumption. Consumption so you can do the hybrid. Basically, you want to match your pricing with how your customers experience value. Two, show your customers what they're consuming before they get the bill. Who has ever opened up their electric bill and said, "Well, that is exactly what I was expecting." No, it's always a surprise and that is not a great experience. So you You want to do the same for your— you want to avoid that experience with your end customers. Real-time visibility is not a nice-to-have. It's a critical part of your product experience and how you prevent churn. So practically, this means finding ways to tell your customers how much they're using while they're using it. And then third, sell credits, not cost. Have your customers exchange exchange money for credits once, so they're thinking about value, not dollars and cents, every time they use your product. Okay, and now we're up to our last pattern. Companies are adapting their go-to-market motions much faster. All right, at Urban Escapes, we started with consumers first, and then we built the enterprise business. Because believe it or not, the Googles and Facebooks of the world loved taking their customers on adventures. Now, that's been the norm, right? You start product-led, startups find organically, and then you layer in enterprise sales years later once you've proven product-market fit at scale. I mean, Stripe followed the same arc. I was our first head of enterprise product, and that was just 3 years ago. Every AI founder I talk to now is hiring a CRO. Building out an enterprise motion is not a 3 years from now problem anymore. Take Cursor AI code editor. They launched in '23 self-serve, then they layered in sales-led to land their enterprise contracts. They now have an an incredibly impressive enterprise business that would have taken other companies decades to build. They launched it in '23. But that's not the only shift. No, because that would be too easy. Entirely new motions are emerging too. So for one, channel sales have now become a huge part of the AI go-to-market motion. A lot of people are buying OpenAI and Anthropic through their cloud providers. And now the AI companies are doing the same thing and building their own marketplaces so others can do the same. So you really have to think about channel sales as a critical channel for your product. And then the billion or probably trillion dollar question that's on everyone's mind is how do we all sell to our new buyer The agent. We've spent decades optimizing pricing around human psychology, anchoring on good, better, best. $9.99 instead of $10. Agents do not care about any of that. I think this chart is one of the clearest examples of agents as buyers. So, in 2025, agent traffic to Stripe Docs 10x'd in a single year. The purple line is humans and the pink line is agents, and they're converging. By the end of this year, agents will read more Stripe docs than humans. Now, here's where a lot of companies underestimate the work. Go-to-market is the tip of the iceberg. Everything underneath has to move with it as well. Adding a new go-to-market motion touches everything in your company. Your product looks different for each motion because the onboarding flow is different. Pricing will have different nuances. You have commits for enterprise, sticker pricing for product, and then your org has to be built for it too. We all know an enterprise sale is very different than how you do customer support for the long tail. It's not just a go-to-market change. We're really talking about your whole business model having to shift. Now, most companies will build the revenue infrastructure the same way, one piece at a time. A billing tool here, a tax vendor there, and then you like add in payments when you need to. And when you're growing slowly, a Frankenstein stack is much more manageable. But when you're compressing a decade of go-to-market evolution into months, every seam becomes a liability. So here's how I think about it. Your customers are channel agnostic. The same customer might find you self-serve, graduate to enterprise, and then have an agent turn on a new feature. They are certainly not thinking I'm self-serve, I'm an agent, I'm enterprise. And so you can't think of them as a channel either. So that means your systems have to move with them. So here's the playbook. First, have a clear graduation process. Know when a self-serve customer becomes enterprise and how pricing changes with that. Second, build unified systems. That means one customer object, one product catalog, one data model regardless of how they came in. And third, design for agent readiness where an agent can discover, evaluate, and activate your product without a human in the loop. All right, we've covered a lot. And one thing is for sure, the playbook is being rewritten in real time from how you build and price, to where you sell and go to market. And the question we all have to ask ourselves is, are you driving the change or are you reacting to it? I'll leave you with this. I remember sitting around a campfire and a friend telling me to pursue Urban Escapes and that it was okay to quit my job.

And I did.

It took months to build a website, 2.5 years to get to 4 cities, and a lot of sweat equity along the way. If I was doing it today, I could be done with all that before the fire was out. And that's what's changed. Not just the tools, but the entire starting line has moved. You can build in hours. You can sell in 42 countries on day one. And you can go from PLG to enterprise before you raise the Series A. The question is not whether the opportunity is real. You've seen the data. The question is whether you're moving fast enough to capture it. I think this is the most fun time ever to start a company, whether you're a startup whether you're a startup or one of the fastest-growing AI companies or an enterprise trying to navigate. And our job at Stripe is to be there with you every step of the way. So, have fun building and know that Stripe will be there to help you as you scale. Thank you very much.

Welcome everyone, and thank you for joining us today.

Our program will begin shortly.

Please make your way to your seats.

Joining us now from Google is the AI Developer Relations Lead for Google DeepMind, Paige Bailey, and the Managing Director of North America Startups, Scott Barnason.

Thank you. Welcome. Welcome to the future. That's what this stage is about. I want to say thank you to the SaaStr team for putting us on the GTM stage. I was a little worried they'd put us on Stage D, which would be a Stage D SaaStr. Anybody?

One giggle?

Yeah.

All right, I got two kids, so dad jokes are authorized. But thank you. I'm Scott Barnason. I lead the startups business in North America for Google, which means folks in this room I get a chance to hang out with every day and see what they're building on Google, which is super fun. Paige, what are you doing?

Yeah.

Greetings, everyone. My name is Paige. I'm the engineering lead for our developer relations team at Google DeepMind, and I get to work really closely with Scott and all of the partners that we have coming through Google Cloud to help make sure that we're building the best models and tools for developers.

Cool. So we're going to have basically a conversation on stage. It's like a one-on-one with an audience. And we're going to get through a couple topics. But really what I'm excited about is to dig a little bit behind the scenes of what's happening at DeepMind. So the first theme, what we want to talk about is the new normal, which is everything changes. All the time. So we had the biggest Google Cloud event, Google Cloud Next, a few weeks ago in Las Vegas. We have I/O coming up next week. Next week.

No spoilers.

And so I want to get your perspective. So Paige works at DeepMind. Now, if you haven't been to Planet DeepMind, it's a place in outer space where really smart people live and they build the future that we all get to play with at Google. Well, so how are you navigating this wild time where we're announcing a bunch of stuff? We're about to announce a bunch of stuff. Like, what is it like in the factory?

Yeah, I do. It's a cliché, but we are kind of building the plane as we're flying it. There's a lot of work happening right now to get everything ready for the launches that are coming next week to make sure that we're adding as many new features as we can as possible. Even automating some of the feature releases, which is, which is pretty cool. And then every single day is a new announcement that feels really game-changing. Like, I'm not sure how many of y'all have heard of Aletheia. It's okay if you have not. It's one of our, it's one of our models that's focused on AI for mathematics. And so AI for proof generation, proof verification, kind of furthering the field of mathematics with these generative AI tools. Alethea, today it was announced that it was able to solve a Kirby problem, which is one of the most challenging feats in topology. And it's wild to think that these things that used to take like decades of work to accomplish are now just happening every single day and in every field, not just software engineering, but also the physical health sciences, biological sciences, and more.

So maybe, maybe if we flip it to the perspective of the folks in the audience. So a lot of startups, a lot of founders, probably some investors. We're navigating all of the Google announcements, but of course all the other announcements in the industry. What advice would you give? How do you focus on what matters and then try to pull away some of the distraction.

Yeah, one of the things that we use internally is, so we all have kind of these agents that are running 24/7, including one that's kind of like a social listening agent, and this agent monitors things like Twitter, Reddit, Discord channels, things that get posted on other platforms that are social media platforms, and it aggregates and summarizes all of the releases that are happening across the industry. This is really, really helpful for folks like me who try to stay on top of Twitter, but it's impossible. But it's also great in the sense that a similar sort of pipeline you can use to kind of keep listening about product feedback, to understand what features should exist for your tools that don't currently. and then to also even automate the creation of those features if they're simple enough or they're like simple bug fixes. So I know that the answer to everything for people has always been, or at least for the last couple of months has been like, you should have an agent help with that, but you really should have an agent kind of help you stay up to speed in the AI releases.

Cool.

Yeah.

All right. I think we're going to shift. We only have 20 minutes, so we're going to get to the next theme.

Yeah.

Which is maybe not even shown for you guys, so that's cool. This next theme is really about reimagining what work feels like. I think probably everyone who's here, you know, we went through our phase of like, stop telling me about AI, and I think we're all kind of like, stop telling me about agents. But agents are here. Agents are here, and agents are reshaping what it means to be productive, what it means to be employee, what it means to interact and communicate. So, so maybe give a little DeepMind flavor and then I'm happy to reciprocate for how it works for our team.

Yeah, I would love to. I would love to compare notes against this. I know that for our new product releases and our new feature releases, now all of our documentation generation is kind of automated just through these agents that are able to create things like tutorials, create really Engaging documentation to even help generate the blog posts, which is pretty wild. We've also been experimenting with having every time we push a new feature, just as part of our CI/CD pipeline, a video gets generated explaining the new feature in multiple languages and also showing kind of how you would use it, the code snippet, but also having a custom avatar. So you can do really, really interesting things just as part of your regular CI/CD pipelines. And then another way that we've been using agents, which I'm pretty excited about, is everybody has an inbox that is increasingly full. Just sort of helping automatically draft responses for some of the things that feel like P0s. As opposed to you having to think through it, all you have to do is review it, make a few minor tweaks, and then send the email afterwards. So all of those things are super, super like time savers for me personally, but also mean that we can scale out the work and scale out more features and launches too.

Yeah, it's an interesting time. So working at Google, I'd say at least every week, if not every day, there is an agent bestowed upon me. You know, there's like a central team or some number of teams and they're figuring out ways to make work easier.

Yep.

But my team's also experimenting and building their own agents. And what's fascinating is I think we're in this maybe a little bit messy middle where you maybe have a blessed agent that's really good at a thing. And then you have these bootstrap agents that are being shared organically across teams. And then you have people And some people are like, no, no, no, I'm just gonna do my agent. And I actually think that's really healthy. We're in this moment of experimentation and creativity, and so again, navigating focus versus distraction is the hard part. But it'll be interesting to see kind of how things evolve.

Yeah, I also think, I love the description of messy middle, because if everyone remembers like MCP servers used to be this huge thing. Everybody was kind of racing to build an MCP server, and then afterwards it was kind of like, oh no, you know, the complexity level is a little bit higher. Let's move to skills. Skills are just markdown files. The markdown files can help the models make decisions and accomplish tasks. And now it's to the place where, well, skills get stale. Like, how do you evolve your skills? Wouldn't it be better to just automate that? Wouldn't it be better to just automate the creation of skills? So I think eventually we're not even going to need to explicitly ask for some of these things. It will just be kind of like, hey, I built you an agent that does what you need it to do because I noticed that you do the same thing 3 times. Totally.

Yeah.

And the world looks a little bit—

That actually makes me feel smarter.

Well, it makes me very excited about where the tech industry is heading.

For sure.

Yeah.

And speaking of kind of agents in how we're using it. So my team works with founders, sales and customer engineering. I see a few of them in the front here.

Hello.

And our booth is nearby. So if you want to have a conversation, like, hey, what are you guys doing? What can I learn? What can we do with Google? Please, this is my plug. It's over there in the corner. Come hang out with us. We've got boba tea also, which you might want. OK, I'm going to keep moving. And we did that one. Okay, I want to talk about connections to customers and maybe ground this in the concept of a roadmap. So, you know, I've been around a while. In my early days, you know, we're doing a biannual release cycle. It's like, okay, we're going to do the spring launch and like stuff all the features in and then you have 6 months to listen to customers and stuff it into the next launch. Like that doesn't work anymore.

More.

Yeah.

So maybe help me with, again, within the factory, two, two-part question. How do you receive the signals that you listen to, but then also the rationalization of what gets shipped first? It just seems like an impossible task.

Yeah. And I think it's also helpful to separate out between like what are things that should be going into the model, like what are improvements improvements that the model needs in order to be able to solve the tasks that people care about? And then the second thing is, what are the features that you should be adding to your API or to a product like AntiGravity to really kind of nail the things that people need? And they're, they're solved in, in much the same way. So that social listening thing that I mentioned before, you can also use it to monitor for things like product feedback. You can use deep research, which we have within the Gemini app, to kind of say, hey, tell me some of the features that our competitors have that we don't have and should add, and then draft up what a PRD might look like. Okay, cool. Now I've got like Gemini to work on that feature and then like go through the code review process. So a lot of this is getting increasingly automated, like just this always-on. I'm going to listen for the things that needed. I'm going to identify that as something that's a low-hanging fruit. I'm going to automate the creation of the PRD. I'm going to use that to feed into Antigravity to make the change to the codebase. And then I'm going to have AI also test it and review it so that when I get to it, it's gone through a couple of rounds of tire kicking to give me more confidence. And then for For the models, I think customers are kind of the only way that we can build the best possible model. We partner with a lot of folks through the things like our early access programs and then also sharing evals. So if you have a bunch of instances where the models aren't doing what you need them to do, you can send that to the Gemini team as an eval, have them run it as part of a suite of evals each time we generate a new model and then understand if we're closing the gap for the thing that you need it to be great at. And that's really the only way to kind of convince a research team to make model changes is by giving them an eval that they can hill climb against. Yeah.

In that first bucket, maybe share a little bit on— I imagine there's still contention on prioritization. I mean, the constraints maybe have shifted a little bit, like where's the bottleneck? How are the prioritization decisions adjudicated? You know, is it whoever builds code that works and we can prove it, or is it really working back from what customers are asking for? What's the tiebreaker on what gets launched next?

I think there— it's a great question. I think there are two, a couple of different buckets. So for the low-hanging fruit fixes that can be completely just, for lack of a better term, like vibe-coded or nudging models to solve for you, I think it's basically like, all right, if you see something that's broken, it's your responsibility to fix it, which I really love as a Texan. If you see that a button is broken or that there's something strange about a transition from one page to another, then that's something that you can do yourself and get shipped as a CL. And then there are these other more broad-reaching problems that will impact customers that would need to be maybe larger architectural decisions. And in that way, the customer feedback and the signals that we get from both social media and from our field teams are the best way that we can kind of make those changes.

Cool.

Yeah.

Yeah, I think for me, maybe—

sorry—

one of the patterns I've observed is that the companies that are closest to the use case are just able to go faster than they ever have before. And going fast and iterating right now is probably the most defensible use of your energy, right? If you're trying to deliver value and exchange that for dollars from your customers. So it's an interesting time to be building. Okay, we've got 5 minutes left. We are going super fast. We're into our last theme. Our last theme is a continuation. It's really about kind of this combination of AI and agents, the two things that I said we're all maybe getting tired of. We're going to keep talking about it. We're in a bit of— on one hand, this is the most exciting time to be in technology because there's so much advancement. On the other hand, it feels a little bit like survival mode because it's all happening so quickly. When I have conversations— in fact, last night I was at dinner with a founder, and they are building an application in the legal space. And I don't work for this company, But he says, "Hey, would this company, do you think this company is gonna do what we do?" And I'm like, "Well, I don't have any way to tell you that because I don't work there." But best thing you can do is stay laser-focused on your customer use case. How do you think about whether it's getting closer to customers or just kinda augmenting your capabilities personally, what are some things you're doing?

Yeah. So I think that personally I have a bunch of different agents that I use to help accelerate my own work and to also help communicate it out. So we already talked a little bit about things like documentation generation, being able to translate into multiple languages, the different kinds of services or the different kind of tutorials that you have, even different videos. And I love the anecdote anecdote that you shared about how everything seems to be converging. Like, even companies that you would have never expected to have conflicting or converging roadmaps, things that were like infrastructure plays versus SaaS plays. Now it seems like everybody's playing in the same space. So it can be really, really hard to understand what to build next. I do think that the best, the best way to build a moat is to build relationships with your customers too. That's one thing that AI can't really automate away, are the human connections and making sure that people really, really feel heard if they're giving feedback, which is a gift, like being able to get it incorporated really quickly. And I feel like there's this urge to kind of automate all interactions with other humans or to manage everything through agents, when the reality is that one of the most magical things that make us human are real-world conversations, getting to talk to each other, getting to work with each other, and really understanding people's problems. So I think that those will only get more important as more and more of the world gets AI-inundated, is making sure that you build relationships keep human connections and make it feel kind of special and bespoke, even if it is AI-enabled.

For sure.

Yeah, that makes sense.

Yeah.

I think one of the interesting things for me personally, there are these obvious things that never change, right? Like that in-person building relationships, that's never going to change. At least I don't think it will. For me, if you ask, hey, what's the most important thing? It's focus. And prioritization, and I use Agents just to look at how I'm spending my time and whether that relates to my priorities. That sounds so obvious, but it wasn't actually that easy to do a year ago.

There's also internally at Google, there's a tool called the Snippets Machine.

That's what I use.

Oh, amazing. Excellent. So it's a collection of tools that help you understand understand how you're spending your time each week, which is rock solid, that will generate kind of an overview of what you spent your last 24 hours or week or month working on. We'll also send you a reminder of who to thank, like who has been helping you over the course of the last week or month that really deserves a thank you. And then it will also generate a meme for the work that you've done if you'd like to select it, which is pretty wild. But I use that all the time too. It's so, so useful. And it's also a good reminder because you can connect it back to OKRs. Like, is all of the work that you're doing directly tied to a company OKR?

100%. Okay, we're to the key takeaway moment. I'm going to put you on the spot. If you were starting a company today, why would you build it on Google?

Yeah.

And what would be the thing you would encourage folks to focus on?

Gotcha. So I would build on Google primarily because we're the only company that has the fully integrated stack from like TPUs to machine learning frameworks to the cloud infrastructure required to run both to even up to the highest level of these SaaS components. I also think it's pretty magical that Google not only cares about SaaS but also about the physical world. So we've got Waymo, we've got robotics through Intrinsic, we've got a whole bunch of different robotics projects through DeepMind as well. And all of the different ways that are being stress tested across the entire company, across the entire world, build a really compelling platform that a lot of folks like Anthropic and, you know, xAI or SpaceX AI or what Well, you know, like things, names evolve rapidly as well. Our, our, and Meta are building their companies around too. So that's, that's what I would do. And then for, for all of you all in the audience, the thing that I would focus on is building not where the gaps exist in the models today, but building where, you know, a world where AI models are $0 and they're are local, and they're fast, and they're cheap, and they're perfect, because that's the world that we're going to live in, and very shortly.

I love it.

Yeah.

Well, thank you all for joining. Thank you, Paige, for the sneak peek behind the curtain—

Thank you for the great questions.

—of DeepMind.

And as I said, if you would like to connect with us or the team, Google for Startups over in the corner.

Thank you very much. Awesome.

Thank you.

Please give a warm welcome to the GM and Head of Ecosystem from Canva, Anwar Hanif.

Hello, SaaSer. Hi, my name is Anwar Hanif. I'm the GM and head of ecosystem at Canva. And so the title of this presentation says Playbook intentionally because what I wanted to share with you today is not just theory. I want to share with you some of our experiences working across agents and AI assistants for the past 2.5 years, which I think in AI terms is sort of like dog years. It feels like 23 years at this point. So let's, let's jump into it. So Let's, let's be honest about what's in the air right now. You know, the SaaSpocalypse headline. So when I talk to founders and ecosystem leaders, I keep hearing the same questions that keep coming up over and over again. Will agents replace my product? How, how do I build integrations that agents can find? And then how do I build my own agents to become tools for other agents as well? With agents becoming widespread at Canva, we asked a slightly different question. The question we asked ourselves is how do we ensure that agents are able to reach out for us, our service, when they needed, when they needed us? The instinct in the industry is to be so sort of binary and the options are one of two, either build your own agents or become a tool for agents. We didn't necessarily see it that way. Instead, we asked, why don't we build our own agent platform and also become a tool that agents use to, for their needs when it comes to design? When we looked at this world of agents, there are 3 shifts that guided us in our thinking and in helping us evaluate and figure out how we wanted to position ourselves. The first shift is behavioral, right? Humans browse UIs. Agents, they end up using— calling APIs, right? If your value is locked behind a graphical user interface, it means that it's impossible for agents to find them. And if you go back to that earlier question of how do we make sure that agents reach us when they need, when they have the need for our capabilities. The second shift is distribution. See, when we've seen this before in the world of mobile and the companies that really won in the world of mobile are the ones that really understood understood where their users are and they met users where they are on this new screen. Agents in a sense are like this new screen. They are this new surface that you need to be thinking of as a first-class citizen. And the third shift is really where all this gets really interesting, right? This is about value. And a human might choose a product because they have, it has a beautiful user interface or a friend recommended to them. But the thing is, agents really don't care about any of this. Ultimately, what an agent needs is it needs something that can get the job done when it needs it, on time, consistently. That's all an agent cares about, the result. And this is where domain expertise becomes the true differentiator and a moat, right? So for example, when we looked at creating value. What we observed is that with AI, the, the, the barrier to entry in creating design has been lowered. So what we did is we, we understood that producing designs that were on brand, that were structured, that were editable, that were publishable, that's where the real value is. And that's where we invested our energies in making sure that we're positioning ourselves in this new world of agents. So let me tell you why we decided to make that bet, right? So many of you might already know that Canva is, Canva makes design simpler, right? And the way we do this is by bringing together all the creative tools and productivity tools that you need to get your work done all in one simple platform. This is rooted in our mission to empower the world and allow, allow anyone to design as well as to publish anywhere. And from day one, it positioned— it pushed us to think about Canva not just as a product but as a platform. And this orientation towards ecosystem and interoperability is what positioned us well in this world. We embraced AI as it accelerated our mission. AI, in fact, today is embedded in every pillar of our mission and every part of our product and our organization has really embraced AI, and that makes all the difference for us. Today, more than a quarter billion users use Canva every month to create designs, making us one of the most widely used platforms on the internet today. And across our users, the demand for AI has dramatically increased. With over 27 billion uses of, of our AI capabilities till date. That makes us the third most popular AI platform used by users today, according to Anderson Horowitz. So our goal is to connect with every tool that you need to get your work done. And that means bringing all the tools, all them that millions of our users rely on to create hundreds of thousands of designs every single month. We host over 1,700 apps and over— and we have over 4,000 developers building apps on Canva today. And last month we had one of our biggest product launches in our history. We launched Canva AI 2.0, which is our own agentic a conversational platform that transforms Canva into an AI platform that has the ability to create designs. Users can describe a goal or an outcome, and we have agents that can take care of that for them with minimal human input. Now, what powers all of this is what is unique. It— all this is powered by our foundational design model. This is a one-of-its-kind in the world. In the world today. Our model understands the full architecture of a design, the layout, visual hierarchy, brand, and what it does is it enables Canva AI to create fully structured editable designs that can be collaborated on. Now we, we already had built an ecosystem of APIs, app developers, and developer tools. but agents required something new. So not only need, not only do we need to connect with other platforms, we also needed to make sure that our design capabilities are directly accessible within AI assistants. So we launched, so last year we launched the Canva MCP server, and this enables us to have one standard by which we are universally accessible across multiple AI assistants and agents. Agent platforms where our designs can be created, edited, and collaborated on by our users. And this is primarily the way in which we've been positioning ourselves along with, you know, and working with AI assistants today. So when you look at how agents handle creative work, there's a consistent drop-off when it comes to the visual layer. Agents need a tool like Canva to produce an on-brand publishable visual communication. Presentations, social media, social media content, docs, all of these flow through Canva via the MCP, which makes sure that when users or when agents are in their flow, they can continue to remain in their flow and use us to, to create the visual assets. Traction that we've seen has been pretty tremendous. So what we've seen is that over the past 2 years or so, we've been the number 1 productivity app on ChatGPT. But we've also seen our MCP connector usage across ChatGPT, Copilot, Cloud seeing 30% to 40% month over month every month, month over month. What that means is like there is a tremendous not just adoption in terms of usage, we're seeing users coming back using the tool over and over again. Now, this willingness of, of us to integrate with these LLMs goes way back to 2023. We were the first custom GPT within ChatGPT's custom GPT store back in the first OpenAI Dev Day. We continue to work with OpenAI by creating a deep research connector. We launched an interactive app within ChatGPT and others. And, you know, last, last year we started working with Anthropic with interactive capability within Claude. What Canva AI does today is it allows us to have connectors that integrate with Slack, Notion, Google, and many other data resources so that users have context within Canva when they, when they use our AI capabilities. So as we've seen the, the number of places that Canva integrates with, we also scaled what's possible with all these different AI connectors. And this is, you know, this is where developing our own AI capabilities really helped us, especially as we started to work with other agents and other AI assistant platforms. So for example, when we launched integrations to make it easier for teams to deterministically create on-demand, on-brand content for you based on sales templates, we made it available for our integrations across multiple platforms where content can be autofilled that are on-brand and adhering to sales templates. Users can also apply their brand, their brand kit to generate social content that is on brand across multiple platforms as well using Canva. One of my favorite features is Magic Layers. What it does is it helps us convert a flat design generated on any platform and make it completely editable in different layers and manually tweak them within our editor in Canva. Now, and these AI agent ecosystems, as these AI agents ecosystem scale, so do our role in this ecosystem. For example, Claude, the Canva app in Claude is one of the most popular apps integrations on Claude. This puts us in a position where we can collaborate directly with Claude Bard, and we were able to work with them on the launch of Claude Design, where users have the ability to create their initial drafts of their design within Claude Design and bring that over to Canva for editing. Millions of people have accessed Canva through these, through their favorite AI assistants and generating designs based, used on them. What I'm gonna talk about next is a distilled insights from all the different experiments that we've done on these different platforms and the lessons that we've learned by bringing our capabilities to our quarter billion users. So lesson number one, and I think this is a true admission of some of the challenges that we face. Number one thing is you need to treat your MCP setup as a living system. The reality is this space is very, very new, right? So we got our architecture wrong the first time. We had to learn, iterate, and build things over time. Our early structural decisions, like where we placed our shared tooling, created problems that we had to unwind later on. We moved really fast to gain that first mover advantage, but that speed came with attacks. Infrastructure choices that felt fine for launch became real constraints over time as we started to scale. Things like hosting decisions that needed that needed to be revisited sooner than we expected. And then the good news is that we started to see a dramatic increase in usage. We've had hundreds of thousands of designs being generated, tens of thousands of users in the first couple of weeks. And that helped us make sure that we continue to invest in the space. Second lesson is that just because you're using MCP doesn't mean that you don't necessarily need to have that API design hygiene. MCP is only a surface, and an underlying API that is poorly scoped or ambiguous doesn't disappear when you build MCP on top. It just becomes harder to make changes or find debug issues when you start seeing them when your agent starts to fail. The investments that we made in our Connect API, which is our REST API, really helped pay dividends down directly as we invest, continue to invest in MCP. The corners that we cut, it started to bite us. So we've been constantly thinking about how do we scale while making sure that our architecture is designed to scale and expand along with increased usage. Learning number 3, agents can be really, really picky. Now the agents can use your service in all sorts of different ways, Many of which will actually surprise you. Building an experience for agents requires us to think about, take a first principles approach. Agent experience needs to be seen as a first-class citizen. Retrofitting a human experience doesn't necessarily always translate for agents. Huge learning for us. It feels pretty obvious at this point, but we had to kind of learn that the hard way. Number 4, engineering trust at the protocol level isn't a one-time task. It's an ongoing thing and it compounds with every new platform that you integrate with. OAuth setup, Okta configurations, credential management. These are like consistent friction points that we started to see across every single platform that we work with. These are the kind of blockers that slows everything down and like, and starts to add up as we start to integrate with more systems. Today we treat our trust architecture as a first-class product feature, not just as an infrastructure. And if an agent— the thing is, if an agent doesn't work well with your service, they're just gonna shut you down and they won't come back. So you need to be, be careful about thinking through these things. Learning number 5: the agent landscape doesn't award a reward generalists, right? I, I cannot emphasize this enough. Canva's bet was that to be the undisputed visual design layer, we needed to make sure that when an agent creates something visual, a sales deck, a social post, or a presentation, they call Canva. So for example, when E&Y built their agent sales solution with, with Snowflake, they didn't build their own agent Design Engine, they reached out to us. The question that I would ask everyone in this room, I know it's a pretty loud room here, is what is that one thing that your product or your company does that's really, really well? Like what does your company do that no one else can do as good? Because in this agentic stack, the tools that do one thing exceptionally well are the ones that will actually really, really do, really succeed. So we believe that the agent era is the biggest thing in the next decade. You probably heard this enough in this room. At Canva, this is the moment that we've been building towards for the past several years. We've designed our ecosystem to create a compounding effect and a feedback loop from every surface and interaction that we have. Every time an external agent reaches out to us to create a design, our platform gets smarter. Every time our user uses our agent in our own platform, it gets smarter. So that those external agents are feeding one side of our intelligence layer and our own agents are feeding the other side. What this creates is that that elusive flywheel. And once that flywheel is spinning, it gets harder and harder to displace because the platform that has the, the most learnings is the platform that can create the next best design. So the winners of this AI-Agentic wave will be building the tools that agents rely on to execute. That happens when you build continued value on your own platform as well as make your platform accessible to agents. So if I were to leave you with some advice, the advice would be integrate early, iterate quickly. Lots of lessons to learn. This space is moving really, really fast. Invest in what you do best. Invest in the things that you're really well known for. And then if you're building agents or any kind of SaaS product and you want to, and you want to, have visual components created as part of that, come work with Canva. We are happy to help work with you and we are happy to help you, your users, as well as your agents create visual designs. So with that, what I would really urge you to do is to learn more, go to canva.com/developers. And I don't know how we're doing on time, but we may have some time for questions as well. Wow. I don't think we have mics. Yeah.

Not working.

Okay, sorry, the mics isn't working. So I think if I were to repeat the question, the question is, obviously I spoke about a lot of the wins, but like what are the some of the failures and lessons learned along the way. I think one of the— to be honest, this space is so new. We walked in eyes wide open knowing that we don't know what we don't know, right? So when we— the way we see it is many of these things aren't failures. I'm just going to use the old cliché. These are all learnings, right? We went in, we had some assumptions around the agent experience. We thought this is how users would be using us. We thought this is how agents will be accessing our capabilities. But even the first time we were testing it internally, you know, there were some issues with some of our tools not being available. And it was impressive to see how the agents were able to figure out all sorts of permutations and combinations to get access to some of the content. I'll give you an ex— the one example is, you know, our search interface for searching content. Content from Canva was not working. And the agent automatically figured out that it can export a specific presentation into a PDF and bring that over to the agent itself, read that PDF, and then give me a summary. Simple things like this, but it kind of shows you how you might be designing for a certain linear experience. But in this world of agents, you're talking about like nonlinear ways in which these agents are gonna use your services. So like, so, and we, Across the, we had issues with scaling, we had issues with performance, there are delays, latency delays because of MCP overhead. All these things for us, the way we see it is we learned from it, we've been building against that and then that's really helped us scale to 3, 4, now a lot more platforms with the reach of over a billion users and more. I don't think the mic's working, so—

There's a mic that works here if you want, and I'll go after you. Thank you for amazing product and speech.

We use it every day. Awesome, thank you. Thank you for being a user.

I wanted to ask, we're a very small team and we want to know what's the sequence to get to our users to use our platform with AI.

And we've built a quick MCP that's connected to our own API that's internal, but it's not safe enough to give the public.

And then we started building our internal agent to talk to our product within the thing. And then we're like, oh no, we have to build an API. What's the steps to get to the market faster based on your learnings and your failures and what you changed along the way so that we can let our software talk to agents and let our users use it in creative AI ways and compete in the market against others that are maybe lagging behind?

So it's a great question. I think the way— I'm a little biased because we have an established product user base and there's a bigger product, but I think the way we think about it is some of these things you just kind of have to get to the market. You know, like, there are a few things that are like absolutely essential: trust and safety, security, we want to make sure that is taken care of. Once that's taken care of, how do we expose features outside, what capabilities do we make available to our users, we're a little bit more open to experimenting with that. I think the advantage, so we have two APIs, we have multiple APIs but two categories. There's the MCP APIs and there's the REST APIs. With the REST APIs, what we've seen is that once you put it out there, you can't change it because you've got developers developers building statically calling your API. So you can't— any changes would disrupt the integration someone else has built. With MCP, it's a lot more flexible that way because you can change things and the AI agents can figure out and figure out and the way, as long as you describe your tools the right way, they can figure out how to use them. So what we've been doing is we've been focusing a little bit on MCP being sort of the front runner of capabilities that we introduced. And then our REST APIs being a fast follower. So the approach that we've taken is we start with the hypothesis, hey, this is how we think, like we have a whole catalog of capabilities. We wanted to see which capabilities actually made sense for the workflows that we are seeing people use within AI systems. So based on that, we said, okay, we're gonna make these capab— this subset of capabilities available via MCP. Like drive some of those use cases, see if people are using it the way we think, and then learn from that and then adapt. And what we've seen is like some of our assumptions were valid, some were not, right? And what that helps us is it helps us inform like what do we need to build more? How do we insert ourselves into assistant workflows, agent workflows? And we've seen differences depending on the kinds of customer, like some might be enterprise customers, customers using an assistant, some might be regular consumers. So we basically have an active monitoring system that figures out all the different patterns of usage and then helps shape the way we expose capabilities to our users. So I think the shortest answer is try it out. Like, don't be too shy. Have that API discipline and like focus on trust and safety, but like just get it out there and try it out because you're not going to find out until you do. And these assistant platforms are also good distribution to get out there where the users are. So we've been very happy working with Frontier Labs on that.

Thank you very much.

Please welcome the Chief Revenue Officer from Owner, Kyle Norton. And joining him on stage is the Chief Strategy Officer from Cloudflare, Stephanie Cohen. Hi everyone! So great to be here. Kyle will be out here in a minute. I thought I would start though by giving you a little bit of context. So we're going to start with who is Cloudflare? At Cloudflare, we help you build, run, and secure agents and applications. Actually, Cloudflare powers more than 20% of the internet, from the world's largest companies to individual bloggers and developers. And our mission is to help build a better internet. And it's a really interesting time to be doing that because the internet itself is changing. Some of those changes are really obvious, but for some of them, you cannot believe that actually already today on the Cloudflare network, more than 50% of HTML page requests are non-human. If we continue at this same rate of growth, by the end of this year, that number will be two-thirds. So let's just pause for a minute. In the very near future, almost all traffic on the internet is going to be machines. And with that change in traffic, the fundamental business model of the internet changes. The first companies to get hit are publishers, people that have news and content as their business model. They're seeing a massive drop in traffic from humans. And that's because people don't read the original source anymore. They only read the derivative. Think about your own personal life. When was the last time you clicked on a source in an AI summary? But it's not just the publishers. The financial services sector is also seeing an impact. How many of you have asked a chatbot for financial advice? Or ask them about your own healthcare. These stats show how often an AI user agent tried to access a healthcare site in just 30 days. And the number for the healthcare site is almost 7 times greater than the publisher site. So you hear a lot about how the publishing industry is changing, but it is every single sector that's being affected. And it's happening to you. Your business to? Is that a bot on your site, or is that a customer? Is your revenue model subscriptions, or is it determined by agentic interaction? But you don't have to play a passive role in this, and that's you as a human and you as a company. You have agency about how you show up in this agent-enabled world. All you need It's transparency, context, control, and then the tools to optimize and to monetize. Okay, let's start with what transparency looks like. The first thing is knowing what agents, what bots are actually on your site. This is like a retailer wanting to understand who's walking through their front door. And it's also true inside your own corporate environment. What data are agents trying to access? And this is part cybersecurity, but it's also part marketing. Once you have transparency, then you need context so you can make an informed decision about which bot, which machine, which agent is helpful to your business and which one is harmful. Training is very different from from inference, which is very different from indexing, which is of course very different than commerce. So now you have transparency, you have context, but now you need control. You need to be able to do something about it. And the best way to instruct a machine is not by putting a do not disturb sign out. It's not by use— by using something like robots.txt. It's by controlling it at the network level. You also need to make sure that inside that corporate environment, all of your data is controlled through what we would call an MCP gateway. The best way to think about this is that we have been doing zero trust for humans for a very long time in the corporate environment. This is zero trust for agents. And for the vast majority of companies, Not only do you want to allow these AI agents, you wanna optimize for it. And that is absolutely true for Cloudflare. It's true for our website, it's true for our blog, it's true for our developer docs. And why? Well, I think the answer is really simple. Because when you ask your coding agent how to build that agent or application, we want the answer to be Cloudflare. And we use tools like Markdown that make it easy and efficient for AI to read your information. It makes no sense for a machine to read a piece of, a piece of the internet that is optimized for a human, which is HTML. It should read what is optimized for a machine. Markdown is 80% less data than HTML. But there are certain companies that Content and news is their business. They depend on ads, they depend on subscriptions, they depend on affiliate fees, and they need a sustainable ecosystem to emerge. And actually, all of you need that too, because there is no AI without high-quality original content. And so they need to create sustainable scarcity so that they can do deals. And this example is one that People Inc. did by using the Cloudflare tools. But interestingly, that monetization model is not just for large publishers. Every non-human interaction on the internet is a chance for commerce. What would be annoying and friction for a human just works for an agent. When every person can be a developer, And every agent can be a customer, and every interaction can be commerce. Every company must start from the position of being agent-enabled. At Cloudflare, we want to be that platform that helps you build for that agentic internet. I'm super excited to see what we can all build together. And Kyle is going to come out and ask me some questions.

Hey, that was great. Thank you. Um, so all really interesting information, but what about me? Uh, and so I think everybody in this audience is in this B2B tech space. So what, what do we need to take away as action items for building our companies in, in this world?

So the internet actually hasn't changed for a really long time. So we had the PC era, we had the mobile era, we had the social era, and now we have the AI era. But in all of those transformations except for this one, the business model of the internet stayed the same. And for this one, the business model has changed. And so what every company needs to think about is what job are they trying to get done for their end customer? And then what is the right business model to support that? And that last point I made about interactions is one of the most important ones. It used to be that you did something like subscriptions. Because it would be a lot of friction if every time I tried to do something I had to pay for it. Very annoying. Inside my company would be annoying. Outside of my personal life, it would be annoying. It's no longer annoying for an agent. And there's a whole host of things that are being built to make that easier. So it doesn't mean everything is moving to per interaction, per outcome, but things that didn't make sense before now make sense.

And, and what was the Cloudflare fit there? You, you described a pretty interesting way that you guys are positioned to solve this problem. I think, I think that's worth sharing.

And so if you think about Cloudflare today, more than 20% of the internet sits behind Cloudflare, and millions and millions of free customers— our free customers dwarf the large customers. And so that allows us to do a ton of things. One, we see traffic patterns and we see bad things that happen on the internet at really internet-scale breadth and depth. But we're also at that link layer where agents have to traverse. And you want to make sure that if you want agents to get to you, you want that to happen fast, reliably, and securely. And if you don't want it, you want those deflected. And at the same time, you want to make sure you're building agents that can traverse the internet that way. And we're the only ones that do— can do both of those on the exact same platform.

Okay. And, and so, uh, one of the conversations that's gaining momentum is agent-to-agent commerce. Agent-to-agent procurement? Like, at what point is somebody going to show up on a B2B SaaS website, choose that product, and buy it? What are you seeing in your data to talk about that prevalence, and where do you see it going?

So there's tremendous demand for where the world is headed. So I'll give you an example. A long time ago, there was a protocol created which is called 402, which was instead of a block, it was said payment required, and it was never used because the payment rails didn't really support it. And so we've revived that protocol with Coinbase and Stripe. If you actually look at the Cloudflare network on any given day, there are 1,402,000,000 responses. So that— I call that a billion voices saying, please pay me, like, I'd rather not block you, please pay me. And that just tells us there is so much demand. And so that's in more of that, that content side of the equation, but we're seeing the same thing on the business side. But there's a whole host of things that you need to do to make sure that this works correctly. There's a whole host of cybersecurity. The minute you deal with, by the way, everything going on with AI and the speed at which the attackers are, are happening. But on top of that, once you have money involved, there's, there's fraud and everything else. And so there's a whole host of protocols that lots of people across a lot of different sectors are working towards so that we can make it easy, simple, and safe. And a good example of that is there's a protocol called WebAuthn, which is an internet-scale protocol that, that we use, which is a cryptographic verification of a bot. It's just a handshake that says I am who I say I am, because all the best-known bots in the world get spoofed all the time. And the last thing you want to do is be making payments to what you think is the Salesforce bot, and it's ultimately something else. On the internet.

What are the other primitives that need to exist before agent-to-agent procurement can, can exist? You've talked about security, identity. What are the other pieces of infrastructure that need to get built?

So one, there's this 402 protocol that I talked about, and the beauty of the 402 protocol is it's made for machine-to-machine payments, but it's also payment rail agnostic. And so there's a lot of talk about things like stablecoins and crypto. And we've announced, say, how we think about stablecoin, but also the regular payment rails work for a lot of this. And so you need a payment rail agnostic protocol. X402 is one of them. And then you need to make sure that you can sign up millions and millions of merchants, because one of the things that's really hard about what's going on is if you're a really large company, you can deal with the millions and millions of announcements that seem to happen every single day about every new thing going on with agents and how it works. But if you're a small and medium-sized business, you actually need a lot of help doing this, which is why we spent a lot of time partnering with Visa, MasterCard, and American Express as examples, because we got to help the millions of merchants that are— have commerce today move to this new world.

Is this potentially what brings blockchain and stablecoin to like real prominence as a part of the financial infrastructure of the world?

We think there's an exciting place for, for stablecoin in this world. Just so people have perspective, the way that the internet operates is at a much greater scale than the way the payment systems operate. So the Cloudflare network has 500 terabits per second of capacity at peak. We generally provide process about 100 million requests per second. If you look at the largest payment companies, they're 10,000 times smaller than that. Swift, which does messaging in the financial services sector, their peak day is like 68 million messages a day. And so there are a lot of things being built, but there's a lot of— we have a lot of room to grow before every single request on the internet is going to be commerce.

So it It feels like that this could be the forcing function that either brings the traditional finance system along or really makes crypto and blockchain the thing, the promise that we've all heard for so many years. It could get us there.

I think there's a lot— if you take yourself back 20 years ago and you think about how the business model of the internet got created, there's a whole list of things that got created in a certain way because It would have been too much friction to do it a different way. And I think so much of that stuff has to be reinvented. And I think you're going to see a lot of companies that exist today reinvented, and then you're going to see a lot of companies that do something very similar to what companies today do, but they do it in a totally new way.

OK, interesting. And so one of the things that I think about is my customer journey and losing control of the customer journey because so much So much of the buying process happens in a chat window, happens through crawling and agents, and so do you guys have a perspective on how I can communicate the right thing to the world of agents to bring people either, like actually get click-through and site visits, or control that narrative and that customer journey that seemingly just like lives in the, in the abyss right now?

Yeah, so it's super interesting because many people know Cloudflare as a cybersecurity company or a content delivery company. And we have a dashboard that shows you exactly who's trying to access your site. And the site owner decides who accesses their site, but they've been in a mindset of I block or allow, good, bad. And the reality is that those are just customers for most companies. And so there is a ton of information just for the company that is helping deliver content and traffic to your site. There's a ton of information there on how you should think about what's working, what's not working. And there are companies like Cloudflare, there's others that can give you an internet-scale view. We have a particularly unique view given the scale of the Cloudflare network. And so yes, there's like an app for that, there's data for that. That's one. is one piece of this. The other piece is, you mentioned this point is like, it's my customer and I want it to be my customer. We come from a point of view that it's the site owner's information, their business, and so that they have the tools to have control. Mm-hmm. And they— everyone will decide to do something different. But in the protocols that we've been working on, for example, the WebAuthn protocol I talked about, which Visa embedded in their Trust and Agent protocol. In that protocol, it's very easy for the merchant to get the customer's information. So they ultimately, you know, maintain the loyalty, maintain the relationship with the customer. And so there's a whole host of things that we have to deal with, whether it's customer experience or security, but it's also making sure that this continuity of customer relationship stays.

And so do, do we need to think about having a completely different experience experience on our site that is optimized for agents? You were talking about the difference between Markdown and HTML in terms of information density. Do I need an About Us page in HTML and a Markdown page separately?

Yeah, so I— silly stuff, for example. So agents really like FAQs. Yeah. And so if you, if you are optimizing for a human, you're probably not putting an FAQ like in every single part of your site. But the reality is that that's how an agent wants to to ingest information. I don't have a crystal ball on exactly where the world is headed, but we have a pretty good view of the internet. And this idea of a human-optimized internet, that is one where we look at with our two eye— eyes, and then you pushing more information directly to AI agents for various different ways to do that so that they're getting the information that you want them to get and they're digesting it in a way that is more in line with how people are going to either ask them questions or they're going to answer the questions.

And so how is Cloudflare evolving their own go-to-market with all this going on? You're living in the center of it, so there's sort of no better person to ask. How are you thinking about innovating and modernizing your GTM all the way from marketing through to success with, with this all happening?

One of our main philosophies around this is that if we don't create scarcity or change, we're never gonna change actually the fundamental way that we organize the company. And so one of the metaphors we use is redesigning the factory floor. And so if you— inside of Cloudflare, we've looked at the jobs to be done. What are the jobs that need to get done? Meaning what are the outcomes? And then how today would you design it? And one of the things you have to do is you have to pull away the old way of doing it. Because if you keep doing it the old way, you're never going to do it the new way. I was in a conversation yesterday. We're talking about demand gen and webinars and finally, why are we doing any of this? What information are we trying to get out there? Who or what or machine are we trying to get it to? And what's the best format to do it? And there's no doubt in my mind it's not the current way.

And so what's an example of taking it away? Take away the old way?

Yeah, so we, um, announced last week, uh, that 20% of our organization was leaving Cloudflare. And the reason for that is because we believe that we need to redesign every single way that we do things. Um, and they're amazing people and they got us to where we are today. But if we don't create the right organizational structure, the right level of scarcity, we're we're not going to get to the right place. And we're a company that's growing. That's an intersection of cloud, AI, and cyber, kind of everything that's going on in the world. But if we don't create the right conditions, your, your technology is your org structure. So if you don't have the right org structure, if you're not fit for change, someone who is newer and faster is going to beat you.

And so then you're thinking about replacing those roles with people that are more AI native, like built around new roles that maybe didn't exist before.

It's not about like this person versus that person. It's about, is like, does that job function need to exist? Yeah. Right now. And if you have that job function there, someone's gonna do that job. But if that job function is no longer there, then you're gonna figure out another way to do it.

Fascinating. I feel like these are some of the bold choices that many companies have struggled to, to take the plunge on. And so what else in, what else in go-to-market are you thinking about fundamentally different, especially on the, on the sales side? Like what, what happens in funnel, uh, today that maybe you didn't do a couple years ago?

Yeah. So if you think about the Cloudflare platform, we have many different personas that buy from us. And for a long time it was super hard to explain why we sold to the CISO and to the CIO and to the CTO and to the individual developer. And it was all very confusing. And actually now it is so much less confusing. And so we have talked about our business in acts for a long time. It's an easy way for the market to understand the various products that we're in. But actually now we build everything on one single platform and that makes perfect sense, right, to AI. So this whole way of talking about Cloudflare has fundamentally changed because it all is just built on one platform, which makes perfect sense to your AI.

AI agent. And these lines are blurring so much in the organizations you're selling to. Well, this is great, Stephanie. Thanks for doing it and appreciate it. My pleasure. Thanks, everybody.

Please welcome the Chief Revenue Officer from Owner, Kyle Norton.

Hey everybody, thanks for coming out. Uh, all right, we are going to talk about all things GTM AI, and I'm going to lay out 5 decisions that I think every company should be considering and making as quickly as possible. Uh, before we jump in and we talk about all the cool stuff that Owner's done in AI, uh, I want to actually share a big AI fail because it is relevant to this deck and I think, uh, properly sets the, the landscape of what we're all dealing with. You know, we're going to talk about a bunch of really cool results and things we've built, and yet in building this deck, I probably spent I spent 3 hours messing around with deck generation tools and other platforms to see if I could one-shot the creation, and then I ended up spending an inordinate amount of time editing stuff. And so AI is all the rage. We've got all sorts of cool outputs and tools, and yet there's still a long way to go. And even somebody who talks about this topic a lot, I am extremely fallible and I'm still learning myself. So with that, I'm gonna hop into a little bit about Owner first and then we'll dive into the topic. So we're a vertical AI company that, and we serve independent mom and pop restaurants. So you can think of us as HubSpot plus Shopify for little tiny takeout places. And so we'll do everything from their website, email, and text message marketing online ordering, custom branded app, a bunch of other tools to help them grow online. We're approaching $100 million ARR. I've been with the company for about 4 years. I started around $2 million ARR. And we serve this mission passionately. Like our mission is to arm the Davids of the world to fight their Goliaths, the big huge corporate restaurant company companies who are squeezing out independents. Our job is to give them the tools that all these big companies have and let them find success for themselves because we all very much believe that this is like the local tapestry and heart of the neighborhoods we live in. So I want to start with our bona fides and some of the outcomes that we've been able to draw with a very aggressive application of AI across go-to-market. And so a few of the ones that really catch people's eye is at the bottom right, you can see we've got a 20x close one to OTE ratio. So 150K rep is bringing in like over 2 million in ARR per year. And that's an average number. That's just not, that's not just like one or two special reps. And we're also not selling tokens. Like we're not one of these companies that you, you know, sell an account and then they just burn a zillion tokens, uh, and, uh, like monetize in that way. We're selling, you know, traditional SaaS with a bunch of AI into it. Um, also compared to our benchmark peers, because we sell to SMB, which is a, a challenging market, um, the ARR per rep is is about 4x what I see from our direct competitors. And so this is the output. This is like what we've spent all of our time and energy doing, driving incredible efficiency into this motion by making reps better, more productive, scaling marketing with all of these investments. Even on the outbound BDR side, a topic of much discussion, our outbound BDRs as of last month, per BDR close over $100,000 in ARR per month. So they are generating $100K in close one per BDR across that team. And again, that's an average. And so when people say outbound is dead, obviously we disagree with that, but the way we do it is really different. We're gonna get into a bunch of that today. So before I lay out these 5 questions, I wanna frame us in this context. This sophistication ladder. And so you can use this as a way to think about where you sit in the, the maturity curve of adopting AI in your go-to-market. So level 0, most people aren't there anymore, but that's like random reps using ChatGPT as a replacement, as a search engine. And, you know, maybe they copy paste some stuff in, rewrite this email, make it make it better. Uh, we'll call that like level 0. Then many companies are here at level 1. You know, they, you have people, maybe individual reps, maybe your, your RevOps or, uh, uh, those teams building, you know, custom skills, some custom GPTs, things that are reusable. You have to, um, Slack your, your, uh, teammate the Markdown file and then they put it into their own system. And now we're starting to get, a little bit more leverage from AI. But a lot of people who are here also feel like we're doing all of this stuff and the outcome isn't there. I'm not getting the output that I want. And we'll talk about specifically how to progress the organization from level 1 to 2, 3. No one's really at level 4. Level 4 is like a recursively self-improving system that inspects itself and builds new tools to get better. Owner is not there. I talked to a bunch of the smartest people in the space, and most, I haven't found any evidence of this yet, but that is the end state which we will end up in, in the not too distant future. Level 2 is where you start to see go-to-market engineering teams. For us, it's an applied AI organization, and they are automating end-to-end workflows. They're building tools that just get a job done. Getting the job done. And that could be manufacturing pre-call research that you bring to the rep, or it could be lead scoring, but these are workflows that are running with some level of full automation. You've got agents, whatever we feel like defining as an agent these days. And so a lot of companies are finding themselves in level 2 and starting to get that real positive feedback. Win rates rise, per BDR and AE production improve, retention rates, and we're starting to get some real value. And level 3 is what we'll spend a lot of time on today. How to get to level 3 is, I think, what is in front of many of us today and is very possible. And so this is infrastructure and like true leverage. When you're building systems systems and tools that can be plugged in across the organization and make everybody better. And so this tends to be centralized, and so it's a deployment platform, so the app that your sales manager designs on his weekend can be brought to a full production capacity and launched across the organization. It's shared skills that everybody can, can plug into. It's a context library that makes everything that you build, uh, better because it has a, a core set of understanding about your business, your, your product, your company. And we're getting like real leverage here. And, uh, this, uh, sophistication ladder was inspired by Brendan Short who writes The Signal, like a, a very forward-thinking, uh, AI practitioner. And, uh, what he's noticed in, in, in spending time with so many of these, uh, companies is that this This gap is widening fast. The companies who are firmly in level 3 are starting to widen the gap compared to their peer set because this is exponentially valuable for us as organizations. It's not like a 10 or 15% lift. You're starting to see the doubling of output on a per-person basis, or things being done without touching humans anymore. And so level 3 is like really where we want to get to and, and because it is, you know, competitively existential for us. All right, so I'm gonna lay out the 5 decisions that I think people should be thinking through, uh, today. So decision 1 is whether or not to have a centralized versus a decentralized model for AI and go-to-market. Like who owns the mandate for, for AI innovation? Number 2, build versus buy. Everybody saw that. Jason M. Lemkin's talking about it right now. I'm gonna talk about where to start in a decision-making and a prioritization framework. We'll talk about the people stack, like what do these teams need to look like? What are the skills that you're looking for? This is a really important decision for organizations to make. And then lay out, lay out a framework to think about whether or not you should be building things that are assisted, you know, copilots versus like fully agentic workflows. Um, and you know, what role do humans play in this whole thing? So these are the 5 decisions which, which we're gonna unpack. All right. Decision 1 is, uh, centralized versus decentralized. And, uh, the decentralized approach is a mandate to your entire company that everybody's gotta be AI native. Everybody's gotta be building stuff, everybody's gotta be doing things and, and building their their own Claude skills and apps and Vibe coding. And, and this tends to be like a let a thousand flowers bloom and, and see what happens. And so there's a lot of benefits to this. You, you build a lot of AI literacy in your organization having a fully decentralized approach. However, the, the downside is these tools and skills and workflows tend to get locked away in small groups of early adopters and it doesn't deliver to you the, the broad impact that you want. And what I see is a decentralized approach often has companies stalling out at level 1 or level 2 of this maturity curve. I will caveat this and say, because, uh, the companies that counter-argue this point are generally the ones that sell AI. If you are a rep at Claude, you have to be fully AI native. That's a part of your job. You have to be able to sell the product. And so if you're selling tokens for a living, or if your product is a wrapper around tokens, then a decentralized model, uh, can make more sense with centralization bolted onto it. So that, that's the important caveat I'll say, because I'm gonna make the argument for a centralized approach if you haven't already figured that out. Um, the centralized approach is a small group of experts that are building this for your teams and bringing things into production. And we'll talk about the best ways to do this, but the key here is that the quality bar is so much higher, and you— there is a compounding nature to an approach where it is more centralized, because the expertise is there. They're building all of these, all of these things into one cohesive package, as opposed to it being disparate. And this is how we get to level 3 and level 4. The other thing that is a real problem with a decentralized approach, and this is a quote from Adrian Rosenkranz, who is the CRO at Webflow. I have a podcast called the Revenue Leadership Podcast, and so if you like this stuff, this is all I talk about. And what he said a couple weeks ago was, you know, a director had brought him like a really cool app that he had built, and it was interesting, but the question that Adrian asked was like, well, is this helping you be on more customer calls? Is this helping you actually move the needle, or does this feel like work, 'cause it's exciting and fun and we're building stuff, or is this like AI performance theater? And Stuart Butterfield has a great quote. He talks about hyper-realistic work-like activities, and things that feel like I'm doing my job, but aren't doing my job. Like, my— your job as a rep is to be in front of customers and help them make great decisions. And you playing around for 3 hours to build a Claude skill that writes you, like, sort of mediocre emails is, like, not a part of that core role, and it's probably not a great use of their time. And so this decentralization can feel like it, it is empowering everybody to go do lots of interesting things, but oftentimes it actually takes people away from their core responsibility, so much so that the value can be negative. Um, this doesn't mean we don't want ever to get interesting ideas from the bottoms up. Of course, uh, that is a great place that innovation comes from. But my argument is that you need a central team that is bringing these things into production because in our experience, experience in working with our Applied AI team, what the Applied AI team builds isn't just like, you know, 20, 40, 50% better than what a rep might build. It's like 5, 10x better. Like, I'm pretty savvy. I build a bunch of my own stuff for personal workflows. And when I see an output that YG, who is my partner in sales, Applied AI, what he builds is just like infinitely infinitely better, and I could have never imagined getting to that level of sophistication and, and building it. And so why have 20 people build something to a, like, mediocre level of proficiency when you could have those ideas centralized and build something that is just, like, way, way better? That, that is my core argument for centralization. And there's some things that are required. You need central resources. They need to be empowered to go build. We sell to restaurants, so we say, a common thing we say at Owner is like, let them cook. Like you have to let this applied AI team go and chase really interesting new and novel ideas and give them a lot of agency to deeply embed in the functions that they partner with. It requires a bunch of data foundations, which we're gonna come back to, and context, which is like a thing that's really difficult to answer and is, is like one of the, uh, unsolved problems. Um, but if you have a proper team doing this and spending their time on it, uh, it is my argument that you'll go much, much faster. Um, and so in this centralized model, the rep is no longer spending a bunch of time building tools and, and running their own agents. Like you see vendors be like, oh, and then the rep can kick off these agents and do this agent stuff. Um, I actually I actually don't really want the reps running agents per se. I want to run that all centrally so the quality is extremely high, and then I wanna share the outputs of those workflows and those agents in the surfaces that they're already using, in Salesforce, in Salesloft, in whatever your engagement tool of choice is, because again, it can be a distraction if you're running and managing agents as opposed to saying, here's what we've produced for your pre-call research, for your follow-up email, for your tiered and scored account list. Just give it to them in the surfaces that they're already using, because we have so much surface sprawl. All right, decision number 2, build versus buy. My opinion is that you should be buying your infrastructure and building your Intelligence. And so these are 5 questions that you can ask yourself if you're thinking about if you should build versus buy this thing, and we'll talk about some examples. So how important is stability and uptime? Is this a critical piece of business infrastructure, and if it goes down, things are bad, like people are panicking? Okay, that, that might be a candidate that you're buying that piece of, of tooling. How customized you need it is, is what you could get off the shelf, like 90%, 95% of the way there. Um, what is the ROI like? How technically difficult is this thing to build? And what is the ROI on those resources? Because it's easy to get excited about something, build it, and then the gain is incremental and you've just spent 2 super expensive engineers working on something for a matter of weeks. Again, that's like a, a good direc— a, a good framework for which direction you should go, build versus buy. Number 4 is, in my opinion, you have to own your intelligence. If you're using tools that lock away critical business context and critical intelligence behind a black box, I'm concerned about that. And we'll talk about some examples, but like your proprietary lead scoring model, or the way that you want to run a specific deal and manage an account, like those are important things important pieces of intelligence for you as a business, and I generally wanna build those things myself. And lastly, does it give you a real competitive advantage? 'Cause if it doesn't, like, why are we spending our calories there when there are so many interesting things to go build? And so I'm gonna give you two examples that I think highlight this pretty well. So we've thought about building lots of stuff, but it's just like an assumption that like, yeah, we're not gonna go build a dialer, like, why would we try to go rebuild Twilio? Like, they have spent an immense amount of dollars on this critical piece of infrastructure. And so, let's go through this framework. So, is uptime critical? Yeah, if your dialer breaks for a course of hours in a day, your team can't do their job. Okay, that leans me to buy. How customized does this experience need to be? Does the ability to, like, click a button and have a call go out. Is that a highly bespoke experience for your team uniquely? Probably not. Lean to buy. What's their ROI? Like, Twilio's a monster company with a lot of engineers. You're probably not gonna wanna go reinvent the wheel. There's a big engineering cost. Same thing, we use tools like OneMind, which is like an AI avatar that lives on a bunch of surfaces, and the technical problem of solving for latency agency, and a bunch of other things. It's just not a great endeavor for us. Is it core intelligence? No, a dialer isn't core intelligence. It's just a tool. And does this give you a competitive advantage? Is having a dialer that is slightly better than your competitor's gonna help you move win rates, move productivity? No. Okay, so that's a pretty clear buy. And so if you ask anybody without this framework, would you buy? Would you build your own dialer from scratch? You'd probably say no, and this is sort of what I think is going on behind people's heads. Let's look at a different example, which is something that we have built internally, which is our AI PCR, AI Precall Research. And so how important is stability for our PCR solution? It's not that important. If it breaks in the day, it doesn't matter. All the leads are batch enriched the night before anyways. You can call some other leads. So build makes sense there. How custom does it need to be? For us, very custom. Like our unique positioning in the market, our unique type of customer. Nobody's building like independent restaurant marketing specific research products. And so again, lean to build. What's the ROI in engineering? This wasn't that hard to build. It was the evals and iteration that took a bunch of time. And then we saw like really big returns on the volume of calls people could make, the quality of those calls, the book rates. And so 2 weeks of 1 person leading to a group of 50 BDRs now being able to book 85% more ops. Tremendous. That's how we got to that $100,000 ARR, close 1 per BDR number is like these types of investments. Is it critical intelligence for you and you alone? Yes. For sure. Is it a competitive advantage? 100% it is, because when we show up and make a cold call, we are delivering a very different level of experience, a level of personalization that our competitors don't. So this is like a pretty clear build. And you can use this framework, and it's fun to like take a bunch of things that you're thinking about and do these side by side, and I think it'll be really informative. You can see some of the things that we have chosen to buy versus what we have built internally. This is why I am like a pretty big bull on Salesforce. I think like if you take that framework and you apply it to Salesforce, this is why I don't think that they are at a disruption risk. You've got a bunch of AI tools in there like Momentum, which again is like a hard technical problem. It's something that I, it cannot go down for us 'cause so many workflows flow off of it. DataLane, hyper-specific, difficult technical problem. Avara is our sim platform. Again, like solving for AI sims is not worth the engineering, the engineering input because I can customize on top of that platform. So, So these are a few examples, and I wanted to just share some of our stack. Okay, where to start? All right, this is one of the other big questions, one of the things I get asked all the time. Like, where the heck do I begin this whole journey? I'm at level 1, I'm trying to figure it all out. It's like cliché at this point, but like, just start with the data. You have to start with the data. Unless you have really good third-party data, data of your entire market, who are your top accounts, do you have them scored, who are in those accounts, who is the right fit, what is their contact information, and then enriching them with why is your product a good fit for that person, or what is your hypothesis of need. Having great third-party data and this map of your market is a critical starting point, because if you slap a bunch of agents that are doing cold outbound or inbound SDRing, and your data is mediocre, they're gonna have mediocre conversations. So third-party data is that entire market map, and we need to spend way more time on this than we ever have before. It's always been important, but it's never been more important than today. The other thing that you wanna do is capture first-party data. So what is happening across your customer journey so I can have a really firm understanding of what's happening and how do I optimize it. I can't run interesting AI experiments if my customer journey isn't well instrumented, if I don't know what has happened across each of these stages and why. And so for us, we use Momentum. It basically takes all of our— it's our call recorder, and it takes all of our transcripts and fills out as many Salesforce fields as we want. Anything my brain wants to know about our sales and customer experience, I can write a prompt for, build a Salesforce field for it, have Momentum injected into that, into that call. So when we change pricing, when we evolve positioning, when we're dealing with new competitors, I can have this perfect understanding. 'Cause you can't ask a rep to fill out too many fields. They don't do it well, they don't do it at all. But these two things are, are the foundation for all of the interesting stuff that you're going to build on top of it. Okay, let's just assume we've got good data. So this is how I, I like to help people think about the use cases that they want to start with. And so it's this 5P framework. So first I want to map the possibilities, like what are the, what are the different areas of opportunity that I have in my business to solve for, not just solve with AI, but solve for as, as, as business critical. Um, then I'm gonna map the payoff of these possibilities. If I solve, uh, if I I solve X problem, it's a $10 million ARR hit improvement, or is it like a modest improvement of reducing the cost of customer support? This, this will mapping these, uh, payoffs is, is really important. Then the probability, what is the chance that this thing actually works? Some of our, the AI projects we've taken on are like extremely speculative with like, with a pretty low likelihood of working out, but if they have a really big payoff and they're extremely innovative, like, we might consider them. And so those two things, payoff times probability, gives you the expected value of that investment. And so this is a riff on a bunch of Annie Duke's work, who wrote Thinking in Bets and How to Decide. And so now I get the expected value of these projects, and I want to, I want to understand, like, how much effort is there? That's perspiration. What is the effort by how many people to to get this thing to work. And I want to map not just the effort to build the tool, what is the effort to get this thing adopted? How big of a workflow change is it? Am I asking people to fundamentally invert their behavior 180? Okay, that's a lot of perspiration. Human change is difficult. And if you take your payoff times your probability, which is your, the expected value from this project, and you divide it by your perspiration, you don't actually have to write out all these numbers. This is a mental model. You can sketch it out into a table. It is helpful, but the actual math here isn't the important part, but you're gonna get a priority score. You're gonna understand, all right, these are our high expected value bets, high probability, high impact, and they're not that much effort. Bingo, let's start with those. The other thing to think about in terms of where to start, especially for companies that are earlier on in this journey, or if you're a big company trying to go through an AI transformation, the change management is, is the number one challenge that you're going to go— that you're going to run into. And so no matter what you do and what you build, if your reps don't use it and if your reps don't lean in, it's going to be really hard. And so I— if you're newer to this journey, what I encourage you to do is to think about about building things in your prioritization framework that are just like clearly net positive for the rep. Build momentum with your organization, get them excited about what this applied AI team is doing. When I tell my team, hey, YG's gonna come, he's in New York, so I'm like, okay, he's gonna come into Toronto and work on some stuff with you. All those reps are fired up. The managers are excited 'cause they know that he's gonna spend 4 days ripping with them, solving some problems, problem, and then they're going to make more money, their job's going to get easier. And so when I ask them to change a workflow or, you know, prioritize a different list, they're much more open and amenable to it because they, they have this understanding, and, and they've gotten over the fear that, like, we're trying to replace them. Because a bunch of companies have not been helpful in the market with this, like, everybody's getting replaced messaging. And so, so there is some fear And so I think being able to assuage that fear is really important to get the momentum that you want. I knew I had too many slides, so we're going to skip this one. All right, decision 4. This is the people stack, like who does this thing and, and what are they going to do? So we talked about centralization. I believe that you need a centralized team of super experts that you probably gotta pay a bunch of money for, and they need to drive this transformation. They can bubble up ideas, you want your reps talking to them and bringing them their wish list, 100%, but this team needs to own it, and you need to build it and then cascade those things down. We do, you know, small pilots as we work out the kinks, and then we roll things out more generally. Where this lives, I'm open-minded. To it. I've seen it work a bunch of different ways. A lot of people have it in their RevOps team. Our Applied AI org lives inside of our data function. Sometimes it reports directly to the CEO or the CRO. That depends on how your own org is designed, but it has to sit with somebody who is cracked themselves. Like, they are really excited excited and AI-pilled and willing to push hard because you're gonna have to push through a bunch of things, getting budget, forcing behavior change. Somebody has to be there that is gonna shepherd this thing along. They should have some technical acumen. When we originally put the applied AI leader in the budget, it was in my org. We started the process and I just noticed my VP of data. I'm like, you're gonna be way more helpful for this person to to solve problems. So like, let's, let's put that role over there. Um, they're still embedded with us and do like work exclusively on our projects. But, um, you, you have to be thoughtful of where, where this thing lives. The, the most important thing here is it's gotta live with somebody who really, really cares. That's the key takeaway. Um, this is, so in a, in identifying the AI talent, It is definitely a different job. Like, this is a different job, and there's not many of these people out here. You'll see that top right is Clay's information about how many GTM engineering jobs are posted, which is just, like, insane, and it's probably way more than that today. And so this person, this type of person can come from a few different places. They could be technical from a CS and engineering background. I've seen RevOps people convert themselves into, into this applied AI function if they're just like deep in it every single day. But the big difference is it's not about building the old stuff and doing the old stuff better. It's about trying to do things in a very different and, and novel way and like reinventing, uh, the wheel. And so generally we like to hire these people from an engineering and data background. Um, our, our RevOps team uses AI all over, like all the time. Time, but this person who's building like technical foundations, I think should have a more technical, uh, a more technical background. Um, the other thing that's important on, uh, thinking about the people stack here is AI should make us rethink the job function. And so when I recorded the podcast with Jordan Crawford, who I learn a lot of this stuff from, he had this great framework. He was like, well, job's just a big bundle of tasks. And in this AI world, we need to unbundle all those tasks and think thoughtfully about like, okay, where, where do those tasks actually need to live? Should it live with a machine or should it live with a person? And get people doing the things that people are good at and get the machines doing things that the machines are good at. And so you could just take the BDR job, for example, For example, most BDR job descriptions have like building prospect lists, and they're doing research, and a bunch of companies spend 60% of their BDR hours just like finding the right people to call. That is a bad job for a sales rep to be doing, and so we're unbundling that, and we say like BDRs don't do that anymore. That is done centrally through like a core data team, and you can think about this in a few different ways, and so I, encourage everybody to be thoughtful about the jobs to be done and mapping that across the customer journey, because we're seeing things that used to be done by sales are now done by onboarding, onboarding teams, or things, or roles collapsing, onboarding functions and CS functions, AM and CS. You want to, you want to unpack all of these jobs and think through where should these things live and how much of these tasks can get done by AI and And then of the remaining tasks, what is the right grouping? And, and it, this is a big thing that a lot of companies are, are, um, struggling to think through and we don't have the answer, uh, either. Um, all right, decision 5 is like what version of AI needs to be done in what function? And so assistive is Copilots. So like a rep has a tool that they are invoking themselves and it's, it's a toolkit that they have, but they're doing all of the decision-making. They're coming up with the next actions that should be taken versus all the way on the agentic side is just AI workflows and loops that happen totally autonomously. There's no human in the loop, things just happen and then that output either happens for the customer and to the customer, it gets done done for the rep. And then there's this in-between where, and I'm gonna talk about lossiness in a second, there's this in-between where we have a deterministic framework where decisions get made, sometimes be human, by humans, with generative steps. Because what we see happening a lot is this, is lossiness. So the accuracy of the model in, each of these steps. There's always some level of degradation or hallucination between, between steps. And I'll give an example. So if you let an AI crawl your website and come up with your value prop, and then figure out who your ICP is, and then figure out what your positioning should be, and go decide who your competitors are, and then write you an email based on, based on all of this stuff, you have like 5 different different generative steps. If you slap them all on top of one another, the email that's gonna come out at the end of it is gonna be AI slop. And we all know it, we all get it like hundreds of times a day at this point. And so, uh, what I encourage people to do is be thoughtful about how many generative steps you are chaining together before there is a human in the loop to make a decision, inspect an output. And you could do this, uh, in that workflow. Workflow where things are delivered to reps. Okay, you know, say you're an enterprise rep who wants to be more selective about their book. An AI can give them an output and say, hey, I think these are the right accounts for these reasons. But you probably don't want to just like send go on the prospecting engine. You probably want an enterprise rep who might have some other context and nuance to be thoughtful about that list before that thing gets sent. And so we don't need too much of the time We think about all AI or all human, and we wanna be thoughtful about, about how these things overlap between one another. Um, and this framework of having deterministic, uh, deterministic components in this chain, I think is really helpful. Um, the other, the other thing that's really important here is evals. And, and so where AI slop comes from, and these like, and having too many generative steps on top of one another, is you building an app or a program or a workflow and then just sending it out into the world. The unfortunate reality, going back to my, uh, brutal experience trying to create this deck, uh, is you are going to build something, you're going to try it, and the output's output's gonna be like, okay, most of the time. And then you have to go back to the prompt that, that was used to generate it, the context you fed it, the steps of, of that workflow, and go back over and over and over and over until you get, uh, an output that you feel good about. And a lot of the time you can automate the evals if you have like a dataset that you can compare your outputs to and give the model what the answer should have Then you can build, like, traditionally what people think about as evals when building AI tools, but like, we are the eval as well. And so especially for this group, as we're thinking about building GTM, I think the human eval is probably the more important step here because we need to spend the time and inspect these outputs and be, and like, bash your head against the wall and be like, it's not quite good enough. It's not quite good enough. I'm going to tweak this. I'm going to tweak that. It's getting worse now. What's going on? And you have to go through this process. And I think this is the biggest place that people fall down is they give up. They build a thing and then they just launch it, or they build a thing, they try it a couple of times, like, ah, it's not good enough. And then they give up. Building the thing is really easy. Like, building the MVP is really simple today. But iterating on it and spending the time and grinding through it is the thing that I think helps break through to like real, to real value. I know that is for me with all of my personal AI productivity stuff is just you're grinding through it. And it was funny because last night I was at an event and somebody was like, oh, well, didn't you try Claude Design with it? It's like, no, I didn't try Claude Design. And then it one-shotted the deck. Earlier today, which I didn't use that version, but, uh, like you, you, you learn as you go and you have these conversations. Um, and, and you like, now I will be better anytime I need to generate something. Like that's where I'm gonna go. But it, it is that like learning loop that's really critical. All right, so, um, these are the takeaways. These are the decisions that I would go encourage everybody to think through. Centralized versus decentralized. My personal opinion is a centralized model is, is the way to go. You could have decentralized innovation and ideas, but making sure that there is a team bringing these things to production grade is really important. What do you build versus buy? Build your intelligence, buy your infrastructure, buy the stuff that needs uptime that is boring and undifferentiated. Where do we start? Start with the data, then your prioritization. Validation framework, then go give your team some wins, earn trust, build momentum. The people stack, who owns it? Who runs it? It's that centralized team and it's gotta be technical. You, these have to be like legit technical, technical folks with engineering chops if you want to get to level 3 and eventually level 4. And then agentic versus assistive. We've talked about the importance of human in the loop evals and spending the time and grinding. All right. I wanna talk about, um, one, so we're talking a lot about AI for our companies and how, how to build this, but like what about us? Um, I think this is probably underdiscussed and we really need at, to think about as leaders, our personal stack, our compounding AI system that we are going to build and is basically our intellectual property. It's our GitHub repos, it's our skill sets. It's our personal context that we're using and bringing with us wherever we go. And so Gary Tan has awesome stuff on it. Go follow him on Twitter and just go through all of his articles. I've taken a lot of inspiration. And I really like this quote where he talks about, you know, a lot of people want productivity. What are your productivity hacks with AI? And he doesn't think about productivity. He thinks about how do I, how do I build things that compound over time, that learn over time, that get better over time, and skills on one on top of another give you a competitive advantage. And so whether it's your meeting notes getting ingested and then adding to your personal context file about how you make decisions or the action items that you need, your, your areas of opportunity, these things get better over time. And then, you know, you read something, you're like, oh, this I I really like this principle from this book. You can add this to your personal second brain. He calls it G-brain, but your, your personal productivity and, and personal compounding AI stack. But you gotta build, you, you just have to go do it. There's no way to learn other than getting in the, the tools and like doing this yourselves. And so these are the 4 places that I would suggest people start. So one, you gotta learn the fundamentals, like what is it about these models that that make them act the way they do. Like, what is pre-training and post-training? What's the difference and why is that important? You wanna go learn the primitives so that things downstream make sense. There's the two Andrej Karpathy videos. They're like 3 hours each, and they give you the best baseline I think exists. And then I would direct people to the Gary Tan stuff and Peter Steinberger, who was The OpenClaw founder, like start there, start learning about these things. It'll be overwhelming at first, but pick some harness that you want to commit to. It could be OpenClaw, it could be Hermes, it could be just using the Claude stack or Codex, but pick something and be like, I'm going to go all in and I'm going to build my infrastructure here and cram in my context, my skills, all of this stuff, all my connectors. Like your system gets gets better the more you invest in it with connections to other things, with compounding assets like skills. And then just like pick a project. Just pick one thing that might seem silly at first, but just like drive it to ground and do not give up until the output coming out of that you're really proud of. And like you're gonna spend hours, like 2, 3, 6, 8 hours just grinding through. That didn't really work. I'm gonna change this. I'm gonna do some I'm gonna ask Claude, hey, the output's this, why is that? And just go over and over and over. Give me that again, give me that again, do it again, run it again. And then you're manually, you are the human eval until that thing can get to a good place. And then you can turn that into a skill and then that's replicable for the end of time. The way that you want to, you know, coach one of the people on your team or the way that you wanna, make introductions to your network. You can build skills for any of these things, you just have to do it a whole bunch and test and iterate. These are some of the things that I've built in my, like, personal system, I think are just, like, useful examples. So, like, a workflow I built was a guest agrees to come on the podcast, my agent sees a positive response to our outreach, and then just automatically sends sends the intake email. So it's like a pre-written email. It doesn't need to be generative every time. It sends this like one structured email, and then it's got a Tally web form. And I picked Tally because it's got webhooks. And so the, the guest fills out like, what are some of the topics they have interesting non-consensus beliefs about? What do they want to, what do they want to spend time discussing? And then this gets pulled into my OpenClaw instance. It kicks off a research skill. So ingest everything about this person that you can find online. And then you find these little things like, well, it can't scrape in their LinkedIn stuff. And you go like, all right, what are the tools to scrape LinkedIn? And you can add that to your toolkit and your skill, and then you run it again and run it again. And so this workflow took me forever to build. I probably put like 6 to 10 hours into it, and I iterated on it every once in a while, and now it saves me It saves me like hours a week. I don't do any podcast prep. It does all the research. I basically have it give me like 5 ideas of key topics. I pick the 1 or 2 that I like, and then it one-shots the docket. But like, it wrote a lot of dockets that were really bad and obvious and boring until I got it to a place where the thing came out of it and it was useful. The generation of the newsletter, even like creating assets that I can pair with publishing the podcast, like LinkedIn carousels. Surprisingly annoying to build, but it was, it was worth it. And these are compounding systems. And then every time I do a podcast, it ingests the whole thing and puts it into my, into my knowledge graph. It breaks apart the transcript into what we call atomic ideas and stores them. So if I want to reference those things later, I can go and it's all properly stored. And that's again, like a Gary Tan thing based on the LLM Wiki, uh, work that Karpathy did. So that, that's a good example. Um, there's a bunch of unanswered questions, uh, which I'm not gonna spend much time on cuz I want to have some time for questions. But, um, the last thing I will leave us with is, uh, this call to arms from our great host Jason M. Lemkin. Uh, he said this on the last, one of the last 20 VCs. He's like, you gotta lead from the front. You have to lead from the front in this. You have to go build your own tools. You have to, like, really make the commitment. You can't have your team, like, be the source of all the innovation. Like, you gotta do some stuff on your own. And there's a bunch of super smart people in here. I love the Jeff Charles quote from Ramp. He's like, "Your job, if you wanna keep your job, your job is to automate your job, and then do a new job that's on top of that job." And so I would just say, like, if you haven't already, now is your call to action. Go out and like build some stuff. It's gonna suck for a long time, but this is how you'll learn and you'll make yourself irreplaceable in the end. All right, thank you. We have time for like one or two questions if you wanna use the mics. Or not. Okay, all right.

Really appreciate the talk, thank you. My question is pertaining to some of the AI slop that you referenced before. Looking at your tech stack, you very much have a best of breed, like when you look at Momentum and Salesforce and Salesloft. Yeah. But you also had a graphic earlier pertaining to the simplicity of consolidating your tools, like the guy was sitting there with his hands up meditating. So can you speak speak to, like, I find that maybe I'm missing the mark.

They're at tension.

Yeah. Like, you see some of these AI-native go-to-market platforms out here, CRM tools that kind of do signal detection, call coaching, transcripts, and everything all in one, which kind of resonates to the simplicity aspect. But then you also reference that you can have generative AI slop when maybe you're generalizing how you go about your tech stack. So what What's your take on some of these AI-native CRMs you see out here?

I wanna simplify for the reps. I can have a bunch of tools and stuff happening in the background that we are managing that complexity, but I want the rep's life to be simple. I don't want the rep to go from Salesforce to Salesloft to AI, like, pre-call research app to, like, write the email after app and, like, have to go through this chain. I, I want, like, fewer surfaces. And, like, if we were a small smaller company, maybe we'd use one of these companies that are off the shelf, more of these tools in one. We're like almost, you know, we're 500 people, 200 in go-to-market, and so what we do might not be like the same decisions that everybody makes, but you can use those frameworks to try to make those right calls. But what I do strongly encourage you to do is just like keep it simple for the reps. Don't fracture their experience across a bunch of different places. All right, I'm getting the please wrap it up sign, so I'll hang out over there if you guys want to come by and ask questions, but I really appreciate the time and attention, and go get 'em.

Let's welcome to the stage the CEO and co-founder of Higgsfield AI, Alex Mashrebov.

Hey everybody, special guest moderator me. I think Alex is going to join. One of— how many of you folks have been on the SaaStr AI agents page? Anybody been on it and check out what we use? Few of you. More should. It gets a lot of traffic. So, so We're gonna have some fun. So Alex is co-founder of one of the most explosive video marketing and other apps, Higgsfield. You guys should use it. There's a free version, right? Can you use a free credit for free?

Can you get going for free? Yeah, you definitely can make maybe 10, 20 images for free.

Good. So Higgsfield AI, while you're on site, use it. And, um, it's changed a lot, but if you look at any of the Saastr videos that I create. If you watch the agents, the intro of me and Amelia, it's all built on Higgsville. That's all little videos that I make. And I've been using Higgsville basically, I think, since it launched or something like that, right? For all of our stuff. Close, right?

Yeah. I mean, and that means a lot to us. I think it was August, September. It was the first time when we started to evolve just from like kind of social, pure social media play to professional creation tooling. Yeah. And you were the first to spot it.

So I love it. I use it all the time. If we have time and I get the internet using, we can even log into my account. I can show you all the stuff we're building. But it's for all of us just trying to learn. I mean, and I'm going to ask a lot of questions. Alex is going to present. I'm going to pepper a few questions. Now they're 300 million in a year, 15 months, or how quickly?

Yes, it's It took us maybe 11 months to get to 300, now it's bigger than that.

Pretty good, right? So Alex is gonna explain to it and where they're going, but I'm gonna pepper, like we hear these stories, you know, how did Higgsfield and Lovable and Replicant get to hundreds of millions in months? And now actually someone's gonna be honest and tell us the real story behind the story. I'll pepper with some questions, so I'll hand it off to Alex, but I'm a super fan and if you guys watch any of our little shows, short videos.

It's all built on HiggsField. Absolutely. This is a great honor for me to present here. So, um, I understand you may be busy, but you all should come here in 20 minutes because we are also gonna give a lot of, uh, free credits to everyone. We are gonna have a QR code. So even if you have to go somewhere, you have to be back in like 15-20 minutes because the last slide is gonna have the QR codes. In the meantime, if you can stay and listen about the Hik-Story, I would deeply appreciate that. So let me maybe briefly overview our journey in numbers. We are excited about a new wave of AI-native video creation for social media. First and foremost, social media is relatively new. It has been around for 15 years, and last 5 years it has become the largest media in the world. We are seeing numerous direct-to-consumer companies who actually make hundreds of millions or billions of dollars of sales by leveraging social media as a distribution channel and using platforms like Higgsville for content production. And this is exactly the reason why we are growing so quickly, as we are addressing this massive demand in the markets to make content for social media every day and build a brand and sell products on social media. So we were the first to maybe introduce camera controls, and camera controls are are extremely important for professional creative directors. We were really surprised that it took off so quickly, and within just, in like maybe 5-6 weeks, we got to $10 million ARR. And then in August, September, and October, we realized that adoption goes beyond just individual creators. And since October, the priority is to make sure that creative teams can work together to make videos very efficiently. So let me give you a few numbers. So for example, this— in a week, we are going to present full feature movie in Cannes, and We are doing that to showcase the possibility and opportunities unlocked with video AI. So roughly it took a team of 10 and maybe 3 weeks to accomplish that. So the efficiency gains are unparalleled and that applies both to long-form and the short-form content creation.

Alex, how big is the team today and how big is engineering?

So this is a good question. Uh, we have roughly roughly half and half split. So 80 people engineering, and engineering is now really broad. I need to acknowledge that software engineering, machine learning engineering are always kind of traditional professions, but, but the amounts of people who do automation, prompt engineering also increasingly grows. And the rest of the team are— we have maybe around like 70 people creative team who are not native prompt engineers. And this actually is very important for us. We learn by external signals and internal signals how to make those models actually usable, how, how to empower creatives to actually use them. Because I mean, now, and what we are seeing is that one creative director can now make ad end-to-end, like let's say within a day. Like before that, it would take weeks. It would require to assemble like physical production crew, rent a lot of equipment, book a space, and so on. And now it gets— now it's completely different. So for us, we believe that's like one of the core unlocks for Higgsville. Is to having creative team and engineering team working together.

But the core, just for folks, so at $300 million AR, you have 60 people in the PRD org, in product and development, doing $300 million in revenue.

Sorry, say this again.

You have 60 developers, engineers, product folks to support $300 million in revenue as a ratio.

No, no, we have roughly, so just in terms of like traditional engineering and product, it's probably 60, 60. So it is still a substantial effort. We were moving very quickly when we were smaller. Now we have to invest a lot in stability, safety. We have to invest a lot in like things like anti-fraud and so on. So the engineering team grows because I mean, vibe coding is not really good for this. Really deep infrastructure work yet.

But it's, by traditional standards, it's a pretty, it's still a pretty tight engineering team even today as you scale, right?

Yeah, I would agree with you because let's say typically a ratio of engineers to ARR would be maybe around 2 2. For us, this ratio is probably around 5 today.

Yeah, maybe 2 to 3 times more efficient or something like that.

Yeah. Cool. And going back to engineering, it's also very important. The very important trend which I see is that the engineering jobs are getting very nuanced. So there are For a lot of new product developments, the team now can consist just, just of one person and they can maintain fast pace of development. So for us, what's important is we help, we deliver time to value faster than any other platform and we have this feedback loop as we see what's trending, we see what users posts on social media and using this to improve HiggsField. And I think that's very important for us to make sure that we have this unique blend of researchers and filmmakers on the team. Most of the team is still— is really centered around kind of those who made commercials, and they— and all the materials published by HiggsField, HiggsField YouTube All of that is generated with AI. It's always generated by our platform. That's how we establish trust. So the reason why we're doing full feature movie in Cannes is another way to show what's actually possible with technology today. And we have over thousands various tutorials on the, on YouTube to help with this transition from traditional production AI-native production for creative professionals. Jason M. Lemkin, this is exactly the point what I wanted to bring up. Like, we are not building in a vacuum. We listen to feedback on social media, on Discord channel, Reddit, but also we have a team of 80 creative professionals who help our engineers to build. So, we We also learned that, also we made a very controversial bet early on. We made a bet that number of people who need social media video, number of people who wanna sell is probably tens of millions of users, but number of people who wanna learn prompt engineering is probably hundreds of thousands. And there is this immersive gap which someone has to close. Close. That's why we started with social-first user experience. What's also very interesting is we see people buying up more and more credits on top of their subscription. So, and that's very natural for all the marketing technology. Jason M. Lemkin, I'm sure, like I personally like, like, I know like 7 years ago learned a lot about ads from reading SaaS stories. That's like one of the positive thinks about marketing technology, one of the positive trends is that customers come and buy more, more, and more of the products if they use it more and if they can convert and drive more sales. And we build the trust that all new video AI models are available on Hixfield so that it's one AI video studio for Canon production. Let me please show how it works.

Let me show you simply.

I'm creating a fashion campaign with 3 locations and 1 model. First, I build my character in AI Cast. The platform generates a full character sheet I can reference across every scene. And same with locations: a rooftop at dawn, a minimalist studio, and Tokyo at night. Now I can start generating. And here's my model in the studio holding the products. But I want to explore other options, so Relight gives me the different lighting directions and color. Angles generates new camera perspectives from the same image. I pick the best one and I move forward. Now I'm ready for the video. I have full control over the camera, the lens, the aperture, the focal length, the style settings for color, the lighting, and how the camera moves, as well as the genre that shapes the pacing. Or I can ask the AI Director. It knows my characters, my locations, and the style I'm going I tell it what I need, and it writes the prompts that pick the best settings for this project. I hit generate, and the character stays consistent, the location matches, and audio is synchronized. Now, when I'm ready, I share this project with my client to review and collaborate. And, and this is very important. I wanted to pay attention to, to the last piece. What's important is that video creation becomes digital and that's asynchronous. And that's very important. Let me please double down here. So the reason why software evolves so quickly is that there is a lot of, there are lots of open source projects and people naturally build on top of each other. Historically, video was extremely gated, meaning that it was, it's nearly impossible to reproduce, to kind of see how these popular movies, how how music clips are built. And we see a major democratization across the space as those prompts become available, projects become available. People can see like behind the scenes, they can see the prompts which were used to make a video and they can collaborate, teams can collaborate together digitally to really like, let's say, make videos together. Like historically, we have seen with incumbents like Adobe is that it used to be like desktop tooling, all the storage was on desktop, and it was just very cumbersome process of constantly exchanging files and so on. We bring it all in one platform. What we also learned, it's like these models evolve so quickly, so we work on behalf of our customers to build the trust with them that they will get the best models optimized for the, for specific workflows on Hixel platform. So we ship pretty much every day. Like last, last year on average we were shipping 6 times a week. This year, as we pay more attention to stability, it's probably 2-3 times a week. And earlier today there was another major launch where we introduced our marketing agent called Supercomputer. So, with, with all of that, we, we are excited that we, we are able to emerge as number one platform in multimedia AI. And there are many direct-to-consumer consumer brands who actually can achieve product and brand consistency, and they are scaling their social media budgets to tens and hundreds of millions of dollars with the AI-generated videos. It's very important to say that in some categories, especially when it comes to mobile apps, AI-generated ad creatives are absolutely necessary. This is the— AI-generated creatives is the only way to get people's attention because there is a strong novelty effect to that, and we pay a lot of attention to that. So those use cases around people building content machines are very important for us when we— for high-scale video production. But also there is a strong adoption by creator economy with people like Madonna Snoop Dogg, Will Smith using Higgsville just to create engagement across their audience.

I've been dropped on the floor, I've been squeezed way too tight.

Seen the tears, seen the laughs, same, same look every night.

Every morning, every day, different mess, same old way. But you always come back to me. Life gets messy, I don't mind. I've seen it all a thousand times.

Spit, splash, love and play.

I'm right here every day. Sorry.

And Alex, I want to show my account in a minute if they give it to come. Oh, it's here. Well, we just saw that video. I haven't done this yet. I'm gonna show my account for fun, if you don't mind. But, um, that was a prompt, or how long did it take to build that?

This is exactly the reason why I bring it up, because I mean, as you can see, there is a decent level of character consistency. Yeah. And historically it just requires a lot of prompting to achieve that. That's why we have proprietary products like Cinema Studio and Marketing Studio, which help to make those ads through visual interface without complex prompting. And please pull up your—

So let me show you if it's helpful to you guys. When I started using Higgscald right after it launched, I did— Higgsville does a lot. If you look at the top, image, video, audio, supercomputer, which we just demonstrated all of this. I'm a basic guy, okay? But here's— in the beginning it had, it had a couple cool things, but this is a great use case for B2B. If you see, you go into it and it has start frame and end frame, and you can update a photo from this event of like the stage and me, and it will just connect them with different models and create a movie from it. There's other ways today you can do it. I'm gonna ask Alex some questions, but it was just magical that I could take a picture of the outside at Saster and the stage and get a 6-second or 10-second video. I'd never seen this before, and what I want to show you, and I still make these videos, here's one for, we have this new podcast called The Agents. I just wanted to animate it with me and Amelia here. It seems very simple compared to what Alex saw, but I'll just, this is my whole account of all the assets we've built, fun promos for this event, this one, all different ones. I think it's pretty fun, right? I'm not yet running an ad campaign. It's pretty slick, right? And we can go back in time, see if I can find any other ones. Here's our— I mean, it just makes these movies fun. Hold on, let me see if I can find one of these. These are pretty goofy. I don't know if we use them, and you can just try different ones, and I want to ask Alex Alex some questions.

The new model architecture is deployed.

Well, that one's not good. Let's see which one— if I bookmarked one of these. Pretty funny. Let's see this one, then I'll show you a good one.

Welcome to the event, everyone! We're so glad you're here! Welcome to the event!

What model is this one? Uh, uh, VL3i. I actually think Sling is like my favorite. I don't know why I'm going to ask Alex about it.

Lizard!

All right, let me show you some other ones you might have seen. Animated in our vibe coding classes. And then I want to ask Alex some questions. Here's all the collateral for the event. Well, let me show maybe one or two other cool ones. You might have seen this in our videos. So like I wanted to make an intro. I don't know if I picked it right. Make a little techy to introduce the event. Sometimes there's artifacts, you rerun it. Hold on, let me show you one more, then I'm gonna ask Alex some questions. We go back in time. This is like to promote sponsorships, we built this. Any of you guys ever played Founderscape.ai, our game? It's pretty fun. If you go to Founderscape, I added it free. It simulates being a founder. Alex is living it from the beginning through hiring your first team, through IPO, through everything, through your stock crashing after your IPO to the singularity. And so let me see if I can show you one for this one. Hold on if I got the right one. Let's see this one. This is for the game.

The network's hum, streams of power, of information, of influence. Every line— oh, that's not the best one.

Hold on, I know I'm jumping around, guys. It's in the afternoon, have some fun with me. Hold on, I got to find the best one. I know I'm probably driving Alex nuts. You You can see I use Higgs field a lot. Hold on, let me find one more fun one for you. Okay, while I find it, I'm going to ask— hold on, this one's pretty cool. One second. These ones are fine.

Hold on, let's see. Yeah, but the thing, this is all these visuals showcase that— 75% fit, that's top of the list.

And 5 solid VCs, this could be our round.

Let's line them up.

Great to finally meet you both. Welcome aboard. Couldn't be more excited.

Faster AI for the win. Get funded.

Yeah, we have all these tools if you haven't used it. We have this, uh, Alex has used it to— a startup evaluator that's used over a million times. All the promos. I don't have time, I don't have a team, I built them all here. And so I know we're jumping around because the end of the day, but when I think when Higgsville launched, maybe you used one model, right? Is that possible?

You use One model. So yes, we started with our own model.

Your own model, which was Open Source Plus Plus Plus.

Is that what it was? Yeah, Open Source Plus Plus Plus. Exactly. So, and then we realized very quickly that, um, the number of models just increases, basically new model coming every week. And it's very difficult to just follow everything what's happening in the space. And this is actually an opportunity for one platform. To consolidate all the models and let's say pick the best model based on each use case.

Yeah. And so now it's much more powerful. You look at it now, you can pick from all the models so you can make a video or an asset, a short video. And, and what's to me interesting, then I'll let Alex talk. There's a couple of interesting things. First of all, you're embracing in a sense your competition, right? In a sense, you have Google Veo here, right? Yes. You could use Sora before it got deprecated, I think, right? Yes. You have your own model. I think you've got Cling and Cdance and WAN, which you know what they are and I don't even know what they are. On the one hand, the product is more complicated. On the other hand, I can run all of these things in seconds in parallel and use different models to see which makes the better video for me, right? Yeah, that's super powerful. So talk about complexity. And then talk about the business model and why it's not a thin wrapper.

This is a great question. And I think fundamentally there is a question for every business in the world if they are a thin wrapper or not. Because I mean, most of the software is going to rely on AI anyway. Right? So, the way how we think is there are two major unlocks for us. So, first is to empower teams to create together. Like, I think the network effect and collaboration is very strong. That's what made Figma successful. That's what made Canva successful. And we do believe there are many opportunities in the video space. Space. That's probably first. Second major unlock for Hixfield is to help brands to sell more products. So we launched our MCP 2 weeks ago. Today we launch our own agent called Supercomputer, and that's really designed to make bulk of creatives to sell more.

And so how— and sorry, maybe you explained it in Jason, I missed it, but just a basic— because we're all dealing with this, right? We all— the LMs are getting so good, right? We're all a wrapper at some level, thin, thick, left, right, on top of it, right? So when I pay for Higgsville, and for me, I know there's two different segments in the market. I'm not, I'm not counting every penny. It's cheap to me. It's cheap, right? But do you, do you mark up the underlying models? Is that a good deal for your customers? How do you think about it? Because like, Sometimes I actually don't love video in Hixfield, but video I could get in Gemini or I could pay for directly, right? So how do you think about the economics? And it's just a mystery to me.

Absolutely. So today, roughly 40% of usage is not associated with just picking the model, but they use workflows like Cinema Studio, Marketing Studio. So that's a way higher level abstraction. That's probably first thing which is important to say. Second thing is we are still one of the most affordable platforms out there. So for some of the models, we are the most affordable. For some other models, we could be top 2, top 3. But I mean, we, we are moving with the market there and increasingly trying to implement agentic workflows. That's where That's where we evolved from kind of per token costs to charging by outcome, by video.

So the model, if you just do the seemingly basic stuff like I have, which is leverage the models more efficiently inside of VIX Field, that's a lower margin product. The upsell is to use it for more cinematography and more value workflows like you described. But you can still make money on the basic workflow of marking up the model, right? Yeah, yeah.

So let me just throw like some numbers so that it makes sense. On average, customers on Higgsville spend around $1,000 a year. So it is like, for example, Canva is $200 a year. And for us, we, every, like every quarter, we almost pretty much double our ACV as we move substantially to potentially up the markets. As we want to attract more Jasons on the platform, the reality is that social media, like marketing budgets, I mean many companies have like hundreds of thousands, millions, tens of millions of dollars. So like the, so when we think about experimentation, from experimentation budgets from even $1 million, it's $10,000, right? And I think eventually Everyone would want to try Video AI if that really yields positive ROI on social media, and we want to make sure that these customers, they will be successful on the platform when they try.

And so it is, it is, I hadn't fully thought about it, and effectively your job is to keep marching them up the value stack, but on average folks are paying you 5 times more than Canva. Yes. Now people are gonna start paying more for Canva as Canva gets more agentic with 2.0, but it is a reminder from the theme of this week. If you build, I mean, listen, I made these videos in seconds on Higgsfield, right? And, and I probably pay more than for Canva. I've been a Canva customer since the 1970s or something like that. I mean, I don't know, I pay $18 a month. I'm sure I pay Higgsfield more than $18 a month and I don't care. But if you provide that agentic value, but the value I get, out of Higgsville, even though I love Canva, is higher. I can build a whole— like, what would it cost for me to build a video, like, without skills and people? I mean, infinity. I'd have to hire an agency. It would come back 2 weeks later. It would be terrible. They would charge me thousands of dollars, and I would delete it because I would never use it. Instead, I can do this in 60 seconds. And so it's just— it's a visceral example that Alex is able to get 5 times the ACV off Canva, even though Canva is mass scale. Yes. By having an agentic workflow that replaces humans. It honestly, for me, it replaces contractor and agency that I don't have the patience to hire anyway, right? Rather than just helping me create an asset manually.

This is a good point. So there are several things which I want to say is that we're seeing that increasingly companies bring the capabilities in-house. As you are right, that's like back and forth back and forth with the agency is just very frustrating, and that's both budget-wise and timeline. Another huge issue is that it's very difficult to go vice versa, go back and say, oh, I wanted the actor in this ad to look differently. It's just impossible with physical production. But like very often what we are seeing on the platform, customers create variations of the same ad add, but just with different actors. Do you— would you return the presentation or we just go? I think the presentation dropped too.

Oh, yeah. Can we put the slides back up, guys? Oh, cool.

So, essentially, what happens is the— essentially, what we are seeing here is that Content that's like one social media marketer can make several minutes of commercially viable videos, social media videos a day, and cost-wise there is also a massive drop in the cost of production. And as promised, these, those are the promo codes.

All right, what do we get if we use the code? What's it we get? Oh, we get Hipsfield Plus free for lifetime?

No, I think it's, I think it's 1 or 3 months trial.

Oh really? It's a good deal, guys. Honestly, do— I mean, not that it's like you can afford it. If you're here, you can afford it. But if you go to saster.ai/ouragents, it's, you know, it's been one of my favorites since the beginning. And I could chat with Alex. It is interesting that, you know, you could do the most atomic, simplest parts of Hipsfield you could do in other places, but even now having it all in my asset list in my dashboard like I just did, it's so valuable. Even if I do it in Veo, I have to figure out how the interface works, I have to figure out where to store it, where to find it later. It's not worth my time versus having everything here in my Hixfield.

Yeah, absolutely. And look, I want to go backwards in kind of the history of innovation, and the reality is that for creators creative workflows and just the assets management, there are so many platforms who, which are basically wrapper on top of S3. We deliver value way more than just asset management, but asset creation itself. And that's why I do believe that like historically in this space, customers were picking products, not just underlying technology.

For sure. Yeah, maybe just ask, let me ask you 1 or 2 questions and then we can break 'cause we're at the end of the day. Agencies, it's a meta question for a lot of folks. Do you enable them? Do you have a Hixfield, I should know this, do you have a Hixfield offering for agencies? Because you're disrupting to agencies, but also if I have budget for agency and they can give me 10 times more assets faster, it enables agencies too. So how do you, what's your interaction with, creative agencies?

So this is, this is a good question. So first part, which is creative agencies, is this is the number one category on the platform. We love to bring up examples of brands building. Number one customer are agencies. Agencies for sure. Creative agencies is number one. We are excited to see more and more direct-to-consumer companies kind of using Hixel directly, but this is still an emerging trend. Like if we just talk specifically about $300 million ARR, most of that's like almost 70% are agencies.

Oh, I wouldn't have guessed that. I would have guessed, I mean, I would have guessed you were getting there, but your early adopters were webheads and AI nerds and people like that. But it's actually agencies found you early because it made them radically more efficient.

Because I think all agencies, a lot of especially creative agencies. Yeah. They have been mostly struggling, frankly, and they use AI as a new opportunity to sell, to sell. And I think that's, there is definitely a lot of demands. We have seen Super Bowl ads generated with AI. We have seen many Olympic ads, Christmas ads, like Coca-Cola Christmas ad was generated on Higgsville. So I think there is just a lot going on. There is clearly, there is clearly a lot of demand for innovation. And also all these companies, they have experimental budgets like millions of dollars. So it's natural that they are all trying AI.

When, when, and I'm just a solo user, so I don't see it. When agencies use you and I think how go-to-market channels change in AI is interesting. When agencies, because you're also disrupting some agencies, you're enabling agencies that use Higgsfield, 70, you know, $200 $100 million of revenue, but you're disrupting agencies that used to charge $10,000 for something that you can do in minutes. Do they hide it from their clients? Do they white label Higgsfield? Are they worried their clients will see it's on Higgsfield and feel like they're being overcharged? This is a good point.

So this is the second part of the markets, which are larger agencies. Larger agencies, they make money from kind of media consulting. That's kind of first, kind of McKinsey but for marketing, right? And the second part is paid, is paid marketing, like basically media buying. So, um, so his video AI and Higgs field did not change the way how media is bought. This hasn't happened yet. But with our supercomputer products, we have direct integration to Meta, MCP and some other MCPs as well, so that we can play broader role, not just from ideation to creation collaboration, but also we want to add distribution piece, basically distribute ads across different ad networks. Got it. And inform clients what works the best. Cool.

Maybe two last questions, then we can wrap unless you want to add anything. So I can't help some for just because I haven't really chatted with many founders about it. For these agencies, it's 70% of your model. I know the team is only 120 people at $300 million in revenue. Is that how you said? Yes. Do you have a dedicated partner team, enablement team, people that are making these agencies that maybe aren't at the cutting edge successful with these tools?

Yeah, we are very, very actively building for deploy engineering team. You are right. Your own team of that, yeah. Yes, absolutely. I think what we are seeing is that the demand, sort of like there is a lot of interest in customization and higher volume content creation. Yeah. We have seen, I don't know, like I think out of top 30 media organizations in the world, at least 10 build internal tooling because they feel that everything available out there is not good yet, but to me it feels like this is actually a signal to us that we should make our products more customizable and like our enterprise, and like we really double down on enterprise adoption.

Okay, and maybe one last high-level question. I really believe you guys are doing 6 releases a week because the product has changed so much since I've used it, right? Some of the stuff you did in the beginning didn't work out that well, right? Or you deprecated or stopped investing in it, like Just, we'll break on this. What sort of, I mean, maybe it's just usage in Mows and Wows and Dows, but how do you decide quickly what to abandon, what to put to the side, what to hide in the nav, where to go deep? Because this product radically evolves. You haven't just, maybe missing part of story, it's not like you went from 0 to 300 million with the exact same product.

This is absolutely true.

I still kind of am in the past using maybe Higgsville 2.0, but 95% of folks use a radically different product than when I started. So how did— how do you pick that? Look at the data, decide what to do, decide what to cut?

Absolutely. So we really saw maybe March last year that there is an interest from top creative directors to use AI, but camera control does not exist. So they just immediately reject it. So that was our first innovation. Then we, I kind, we kind of went to our roots. So I sold my previous company to Snap where we build the face filters basically. And face filters is just kind of one-click experience. So we did the same for visual effects. And this helped us to get to $10 million ARR within 8 weeks or so. Then the interesting part for us was to also I mean, we noticed maybe in July that some people start to make commercial projects end-to-end with AI. And this is where like kind of AI, video AI is no longer a toy. It's not just a tool kind of for some fun effects, kind of face masks or whatever. It actually brings a lot of commercial value. And we reoriented the whole company around that. And now we reorient the whole company again around Adgentik, because as I mentioned, like Adgentik workflow is like, it's like marketing workflow is extremely repetitive. Every day it's important to understand the trends which are relevant, which are irrelevant, what's worked well yesterday, then decide if which videos we need to make today, how it's different across different channels, what's the difference What's the audience's overall sentiments? And then make the videos, post, and rinse and repeat every day. And that's very important to stay relevant on social media. So, and I think kind of agentic, actually, I mean, like our agentic workflows is our first step to cover the whole marketing workflow end to end. Okay, that's amazing.

One last wrap question. I have a question just at a high level for founders over stuff, because this is, I think, an over-discussed topic, but what is ARR? You've got $300 million ARR, but this isn't all Salesforce subscriptions at $150 a month per seat. How do you define it? What does it mean? I think you're pretty honest and direct on this, so just give us a snapshot of what you think ARR today is in an AI and agentic world.

Absolutely. So I think there are two subcomponents. First, I think it's important to take annual subscription and divide by 12. In wherever you, how you measure ARR, it's just very important to do and kind of be intellectually honest. And I think a lot of kind of wishy-washy stuff starts with credits. Because if someone uses on-demand, Then there is a question of like where it all gets attributed, right? Yeah. So like for us, for 300 million, we see the on-demand usage for specific last 4 weeks. We take the monthly subscription revenue, annual subscription revenue divided by 12, sum it up altogether, multiplied by 12. And that's how we get the result. But I think on-demand usage is the most tricky one. Yeah, but you're using—

but you're still using the last 30 days of revenue in essence and multiplying it by 12, right?

Yes. So, but it's revenue. It's not just sales, it's revenue. What's important?

Booked revenue, GAAP revenue, or—

Yeah, yeah. Revenue, meaning that we take, we take like annual subscriptions or annual contracts and divide by 12. Sure. Which is important because some companies, they take whatever like cash they got. Yeah. And multiply it by 12. This is, I think— Well, that's—

but okay, and but of $300 million, how much is on annual versus monthly versus— I mean, you have a lot on annual?

Look, annual is very substantial. I mean, like, look, in terms of the number of subscriptions, it's not high. But I mean, it's definitely like 40% of total revenue for sure.

I just, yeah, so we could break on it. I just, I think some folks are stretching the definition of ARR so far it stretches credibility. But also, I also think sometimes people take this issue too seriously. If you're doing 300, what's 300 million times 12? I'm getting tired. 26 million or something like that, right? This month you're roughly doing $26 million honestly, then you have a $300 million ARR, revenue rate, right? It shouldn't, we're making this too comp, whether it's credits. Yeah. I think if you're giving 80% marketing discounts at zero and claiming it's revenue, it's pretty suspect, right? Yeah. If it's cash in the door or cash properly recognized for accounting, I think we're, what am I missing, Alex? I think we're overcomplicating this. Should be your AAAR, you know, close enough to what your real revenue is, whether it's one-off, 10-off, 20-off, annual, weekly. At some level, I don't care. What is your rep? Just show me their money, right?

Yeah, absolutely. So for us, I think what's— and I also spoke a lot about that with Stripe team, frankly. A lot of businesses are run on Stripe. Stripe also has the orchestration across the various payment systems, and I think, I mean, in the At the end of the day, I think very soon we will be sharing Stripe dashboards because, you know, like it's just kind of this, otherwise there is just a lot of discrepancy in definition. Stripe says $25 million this month.

I call it $300 million revenue.

Yeah, we will call it a day.

That's good enough for me, right? All right. Incredible success story. Let's give it up for Alex and thanks. Thank you. One we can be proud of.

Thank you. Thank you. Thanks for doing this. Thank you. Thank you for having me.

Alright guys, I think it's happy hour time or whatever else we want. So we'll break. I know everyone's gonna get tired, but if you come tomorrow and you have the energy, again, I think at 9 or 9:30, I gotta check the schedule, but Amelia and I are literally gonna go through every single agent in the SaaS Stack, show you the backend, show you how the agents work, show you how AgentForce works, show you how some of the products here work. We will show you the the real bones and how it all works. And so you can see it, ask questions. So if you have the energy and you really want to see how it all works, our agents, we will show it to you for real. So thanks, everybody. Let's get a drink.