The Agentic Internet: Building and securing what comes next (Part 1)

I want to start out though with you know a rumor which I heard was going around yesterday here at Connect which was how did I break my arm and and I am and I'm here to clear up once and for all that it was not in a fight with Andy and Matt at AWS I am much like Cloudflare I am significantly younger I'm more nimble, I'm more agile, I'm friendlier and I think I could take both of them with one arm No problem. So so that wasn't the cause and instead it was something much more plebeian I was in a bike accident in Corsica about two weeks ago. I had surgery about a week ago I now have two plates 17 screws in my arm So, you know, it's a little painful, but there was nothing that was gonna keep me from being here with you being here at Cloudflare Connect. Thank you so much. It's amazing to do this. I also get to try out my sort of David Byrne look. So, you know, if I break into song halfway through, you're all too young to get the reference. Andy and Matt would have gotten it. But anyway, what I was thinking, though, as I was getting ready for surgery was, wow, what an incredible time we all are living in and what an amazing future we all get to build together. And it's actually kind of... a contrast for quite some time if you actually look at what the web has looked like over the last period of time from 2012 to 2025 it was pretty flat in fact for 13 years we really didn't see much growth across the web and by some measures the web was actually shrinking and from our perspective at cloudflare that was kind of terrifying because we thought web is one of the most things that humanity has ever created. It's brought so much value, so much incredible things to so many people around the world and if it's shrinking at some level it felt like maybe we were failing and we were failing it and we were failing society. And then something changed. In mid 2025 a revolution happened and what we see right now is growth across the web that rivals what we saw in the early 2000s. The web is expanding at an astronomical rate and more and more of it is sitting behind Cloudflare and we're helping power it and I know that some of you out there are thinking oh it's just AI slop and it's a bunch of garbage and and you know there's not anything good and yeah there's some of that but it's not most of it in fact what we see more than anything else is what's really driving the growth of the web is new creators that for the first time are actually building things without having to have coding skills well how to know all of those things that you used to have to be an expert in you can get up and tell your stories you can get up and create your businesses you can get up and actually bring things to the world in a way that was never possible before and i think that's amazing with no coding skills today you can create incredible things you can build things in record time with some of the vibe coding platforms and what i'm incredibly proud of is that as you use any of these tools from, you know, Lovable to Base44 to, you know, some of the things that they're doing at OpenAI or Anthropic, the number one place, the destination for that code to live is on Cloudflare. And it's incredible to see that growth. And so today, I'm excited that we have crossed a milestone. Last quarter, we announced that we had 7 million developers using Cloudflare, and that was incredible because that was up from just 5 million before. And today, today we've crossed to hitting over 10 million developers on the Cloudflare developer platform. And it's amazing the diversity of that. There are the experienced coders who are saying, I can build faster and better than I've ever done it before. But there are also people who have never touched coding who are saying, I'm able to deploy something and know that it's gonna be fast, reliable know that it's going to be secure because it is built behind Cloudflare and I'm so proud of the fact that we are bringing that to more people. Technology when it's at its best brings creativity to a larger group and I think that's what we have the opportunity with the agentic web of delivering and so most AI builders that are out there deploy directly to Cloudflare and we make it and we make it as powerful as we can. And again, technology is at its best. It allows people to express their creativity and do that in a way that they could do without that. What's amazing also is that who is using the web or what is using the web has changed pretty dramatically. When I was here last year, I was asked, when do I think that automated traffic would pass human traffic online? And I forecasted it was going to be in the second half of 2027. And that was based on really good data. We looked at everything. everything that was there. Fast forward a few months later, I was at South by Southwest in March of this year. We looked at the same data again, and we thought actually it moved up. We thought it was going to be in the first half of 2027. And so I was blown away when in May, my team came to me and they said, you won't believe this, but automated traffic has passed human traffic online. Agents are more and more what is consuming whatever. it is that you put online and i don't think most of us have appreciated how much that's going to change if we extrapolate going forward and and and this is with the caveat that i've been wrong with every prediction way too conservatively so far but we extrapolate it in five years we think that agentic traffic will be a thousand times what human traffic looks like online to see that visually this is what the agents will look like that's the blue line orange line is the human line in other words humans are going to be a rounding error online not because people are going to stop using the web we think it's going to hold pretty flat but because more and more it's going to be consumed by your agents they're going to do incredible things i mean we already see this how can you find me a flight how can you find me a contractor how can you find me a cheaper phone plan i mean it's incredible some of the things that folks like meadow are doing with muse and we've partnered with them in order to power lot of what they have done we are making this more powerful and it's giving us all superpowers so that you can do more we can take an afternoon's reading and summarize it in a minute and that's incredible and I think most of us here are experiencing exactly what you can do with these new tools and I'm proud that the primitives that Cloudflare has built are enabling more of these tools to run incredibly successfully it does the legwork for you so you don't have to but and again the cloudflare like we're excited for this but it's not free that legwork has real cost and the question is like who should pay for it like let me give you an example if you typed into your chat agent right now where should I go for lunch right you know what what's what's the impossible answer it's gonna go out and scour thousands of different menus from thousands of different restaurants that are out there and one of those restaurants gets your business right and that's amazing except the other 99 are carrying the load of delivering that content for that and so if you're one of those 999 that isn't getting people coming to you but you're still having the agents crawl your site, consume your resources, use your bandwidth, consume the servers that you have, then the problem is that we face this massive tragedy of the commons problem where over time it's going to be in the interest of agents to check as many things as possible, but the number of people who are actually going to benefit from that is going to be smaller and smaller. And so last year when I was up here, I talked to you a lot about publishers, how we were seeing that the traffic to publisher sites was... dropping off of a cliff. And we've worked tirelessly to actually fix that problem and make sure that publishers who are doing incredible work, reporters aren't going anywhere anytime soon. They're really important for our society to exist. We need to make sure that they're compensated. And I'm proud of the fact that at Cloudflutter, we've partnered with the largest publishers in the world in order to get them better deals from the AI companies to compensate them for the hard work they're doing. We need a sustainable ecosystem that supports and anyone who is a creator that is out there, and again, we think our role is helping figure out what that sustainable ecosystem looks like. What I'm worried about this year though is the restaurant and the deli, because I don't quite know what to do about that. AI has made it easier than ever to build something, and AI has made it easier than ever to build something that is amazing, but agents might make it harder. for anyone to find that new thing right think about you today why you patronize the small businesses that are in your life like you might go to the deli because the guy behind the counter remembers your name right that's an emotional connection that you have to that or you might stop by the dry cleaner because it's on the path for your drive home it's convenient for you humans care about emotions humans care about convenience because we only have a limited amount of time. The problem is your agent doesn't care about any of those things. Your agent has almost infinite time. It's not swayed by emotion. It's doing what is best for you. But the problem is how is that going to drive business to small businesses that are out there? The agent isn't driving past the local store. It can get delivery from anywhere in the world. It picks what it knows the most about. which means that it's going to pick those things where there is the most information that's out there, which means that the winners, if we don't do something, might only be the largest companies that are out there. In fact, if you don't have a market where new entrants can enter easily, what over time happens is consolidation, where more and more goes behind the big and the powerful, and it's harder and harder for the developers, the startups, the small businesses that are here in the office. to be able to succeed and I worry about that because I know the story of Cloudflare we not many blocks from here back in 2010 stood on stage and launched this was the entire team at the time in the audience right before we stepped on stage we knew that there were still 27 bugs left to be crushed as Michelle and I started to present we had gotten it down to eight and when we said it was live we thought it was zero and went live with a new service. No one had heard of us before that moment. We had a network that was we described as massive. It had five data centers across three continents. The one in Japan only worked about half the day, so we had to turn it on and off to not absorb a bunch of traffic from the West Coast of the United States. And yet, like, people took a chance on us. People bought into it, and they bought into it because they wanted to root for us as being successful. The problem is that I worry nobody's agent would have ever recommended us. And if that's the case, how do we make sure that the next Cloudflare is able to be built and launched and succeed over time? People took a chance on us anyway. And what we have to do is make sure that, like the people who presented at Cold Start, I mean, amazing companies that presented, showing what they were building, we have to make sure. that they have a chance as well. That's what we're playing for at Cloudflare. We're playing for a world where we can make sure that there aren't five AI companies that are dominating everything. There are 500,000. How do we make it as easy as possible for anyone to build a unique model that's there? We're trying to play for a world where artists and creators, folks like Rashid, Common, are able to not wither away because... AI is just stealing their information, but actually get compensated and paid for their work. We're playing for a world where the biggest don't win by default, but where whoever has the best product, even if it's new, can get discovered. And I think that the thing that I'm spending the time thinking about is what replaces the Internet's old business model, which has done miracles but also had real challenges, new business model that can play for that world where there are many AI companies, many media and creator companies, many businesses large and small, all competing in a way that there is fair. And I think that's the most interesting question of the next five years. And so, again, we get another swing at this. We get another swing at figuring out and reinventing what the Internet has done. And we need to think about how. we can do things better so google who pioneered scraping and searching basically just scrapes and search and crawls no matter what and what you'll see is that when they go to a page most of the pages that they look at haven't changed and so as the ai companies built this they're like well that's how google does it let's do it the same way but that's incredibly inefficient and so what we said is how can we actually make this better how can we partner with great companies like open ai to say let's tell you when the page has changed so you can get the freshest data without consuming more resources because if we can decrease the resources on everyone, maybe the tragedy of common problem isn't as bad. So fetch only what's new and then get paid when agents use what you've built. And what's incredible is that we've delivered this and we've got even Google to buy in to thinking that this has to be the future of what is going to have a sustainable internet. And I'm proud of the fact that the Cloudflare team is leading the charge in making sure that we can build that business model of the future for the internet. We cut the load and price what's left and we are very much in the mission of helping build a better internet. And help is a really important word there. I don't think Cloudflare can do it alone. We need all of you. We need great companies. We need partners that are crazy. And so we have partnered with some of the best companies in the world in order to say, let's build something. it is sustainable. This is too important to just leave up to chance. Let's articulate what that future should look like. A future where we're not, again, going to have just five AI companies, we're going to have 500,000. Where we're not going to have a tight, small number of creators, but all of you can be creators. And when we're not just going to have a consolidated few number of large businesses, but where anyone can start a business and thrive. That's what I'm spending my time on. That's what a huge amount of the Cloudflare. team is spending our time on, and I can't imagine any work which is more important right now. Whoever is building what's next, they're building it on Cloudflare, and I'm incredibly proud of that. The agentic internet is here. It's amazing for what it is going to be. It also has some challenges that we need to think about and get ahead of and work on. I'm proud of all of the work that we're doing, and so with that, I'm going to hand it off to our chief strategist. who is going to talk to you about everything that we're doing in this space. Thank you so much. Thank you for being in Connect. Stephanie, break it. Thank you, Matthew. Rashid said it quite beautifully. And Matthew showed you the numbers. The agentic Internet really is here. Models may have been the foundation of AI, but today's world, it is dominated by agents, and agents are different. A model decides, while an agent acts. Deciding is uniquely a GPU problem. Acting is a CPU and a network problem. In some ways, agent behavior mirrors human. behavior, they get things done, and they make mistakes, or at least this human does. But an agent uses the network differently from a person. Matthew told you that an agent reads more pages in a minute than you could in an afternoon. A person will pass out eventually. An agent doesn't. People will trickle in, but a million agents are going to go after this. It's the same thing in exactly the same second. A person will log in as themselves, but an agent arrives on someone else's behalf. Humans have always demanded a lot from the internet, but those agents, they need even more. They need a network. It's fast on every call. It has room when demand spikes. keeps working when something breaks and is protected from attack. GPU clouds were built for deciding, but acting happens on a network. Matthew told you that in 2010, we launched our network with those just five data centers on, I guess, three continents, but we've been building on it ever since. I'm going to start with how close we are to every single thing on... on the internet, but why does that matter? When an agent acts, it's acting on the internet, on a website, an API, a store, a database. More than one in five websites on the internet uses Cloudflare. As you move through the internet every day, hopefully not during this presentation, you touch Cloudflare continuously. Agents, they're doing exactly the same thing. They're just doing it a lot more often. So for a large share of what agents do, the thing they're interacting with is already on our network. It's often on exactly the same machines. Our network today spans more than 340 cities across more than 130 countries. But those dots on a map tell you where a company has equipment. They don't tell you whether the same services run everywhere. In a traditional cloud, different regions offer different products, and traffic travels back to a regional hub for security or for compute or for data. At Cloudflare, we took a different approach. Every single service runs on every machine across our network. So an agent acting for someone in Tokyo actually runs near Tokyo. It doesn't make a round trip to Virginia for a security check halfway through its task. and nobody building the agent had to assemble a global footprint to make that true. They inherit it directly from our platform. Cloudflare connects directly with 14,000 networks, and direct connections make things fast and make them reliable. Our global, highly interconnected network means that Cloudflare is within 50 milliseconds of 95% of the world's population. So how fast is that? A blink of an eye takes approximately 400 milliseconds. We're eight times faster. For a person, that's going to make a website or application feel immediate. For an agent, it compounds. At 50 milliseconds a call, an agent makes 30 calls in a second and a half. That's the time it takes another agent, a region away, at 300 milliseconds. milliseconds to make only five speed isn't a region our customers have to choose it's a consequence of our platform being everywhere that's fast by default our network has nearly 600 terabits per second of capacity that number is almost impossible to fathom so I'm going to try to put in a little bit of perspective when cloud for launched at TechCrunch in 2010, we had less than one terabit per second of capacity. But just two years ago, we were only at 250 terabits per second of capacity. So in two years, we've more than doubled to a level of network capacity that is greater than most countries. But why does that matter? When millions or even billions of agents go after the same thing, it could be a product launch, a ticket drop, a break. the demand will land in exactly the same second. So does an attack. Nobody building an agent or interacting with one should have to predict that spike and provision for it. Our network just absorbs it. That's scalable by default. The Internet, it feels virtual, but it's actually physical. Cables get cut, routers will fail, data centers are going to lose power, weather, construction software and simple human failure all can disrupt a connection. The question isn't whether something will fail, it will. The question is, who notices? A person notices, they'll hit refresh, they might come back tomorrow. An agent acting for someone who isn't watching may be your agent. agents, they don't have anyone to hit refresh. The failure gets routed around or that task just fails. When part of our network becomes unavailable, traffic can take another path. The application will stay online and the agent will just keep on working. Go agent. That's resilient by default. Security is built into our network. It isn't a separate stop traffic makes on the way. to an application. As I said before, it runs on the same machines. And our scale makes our network smarter. The more it sees, the faster it recognizes what's dangerous. When we spot a new attack on one customer, what we learn protects every other customer on our network. That's secure, by default. For the last 20 years, building globally meant making choices. I'm going to show you what that can feel like. We're going to imagine for a minute that this room is the world and each area is a different region. A traditional cloud would ask me to choose where I should run this presentation. I'm going to choose over here. You can hear me. For you the experience is is working as designed, but most of this room, they're having a really bad experience. Welcome back. So for part of this room, nothing went wrong. The presentation was just working, but it was simply working in the region I selected. No hard feelings, sorry about that. But this is the trade-off. we've been asking our customers to manage. They choose where an application runs, who gets the fastest experience, and what happens when a region fails. At Cloudflare, we start with an entirely different assumption. You get to deploy once. Your application and your agent working with it runs across our entire global network, close to whoever is serving. Compute and storage and inference, networking and security all run. on exactly that same network. Routing and scaling are handled by our platform. And global, though, doesn't mean giving up control. You get to decide where your data is processed, where your encryption keys are stored, and where your logs are kept. Global availability with local control. Other clouds ask you to choose a region. At Cloudflare, we give you region Earth. The Internet was built for humans. Cloudflare was built to make the Internet faster, safer, and more reliable for everyone. It turns out that same architecture is uniquely suited for agents. Agents are more numerous, they act continuously, and they can create enormous demand at machine speed. They need infrastructure that is close, elastic, resilient, and secure, all by default. Cloudflare's network was built before the agentic era, but it was built for this moment. One network, one platform, available everywhere. Built for humans, ready for agents. This is where the agentic internet gets built. So many of you know us as that company that acts as a filter that sits in front of your website. It makes it fast and it keeps it safe. That's still true. But our same network, we now run code, store data, run inference and host agents. We're a platform. And the rest of this morning, it's about what you can do with it. Today, we'll show you. how you can build, secure, and scale with Cloudflare. I just talked about scale. We'll take the rest in just two parts. First, we'll cover build. Rita is going to show you how to build and run agents on Cloudflare. Evan will show you how to make what you build fast for everyone. and for everything that uses it. And Matt Conroy will show you how you can stay in control and get paid on an internet where more and more of your visitors are agents. We're then going to move on to secure. Wesley and Grant will show you how to keep your data, your network, and your people safe. Then Sam will come out and show you how we put it all together inside of Cloudflare. Along the way, you'll get to hear from our customers. partners who are already building on us. Rita? Thank you, Stephanie. All right. Let's do a quick show of hands. Who here has used workers before? Woo! lot of hands up. It actually makes me emotional to see it grow every year. And it's, believe it or not, been 10 years since we originally announced Workerz, which marked the beginning of our developer platform. Now, there's no way we could have known it at the time, but Workerz has turned out to be the perfect foundation to build on, especially in the agent era. Let me explain to you why. Workers wasn't the first ever functions as a service service, but it was the first ever cloud primitive that was built not on VMs, not on containers, but on isolates. The reason that distinction is very important is because isolates can uniquely, unlike VMs or containers, only bring in the application code without having to spin up the entire process overhead every single time. This results in orders of magnitude added efficiency, 100 times more efficiency, in fact, which in a time of abundance of code, like we're living through with agents right now, becomes really, really important. But, of course, workers was just the beginning. Today, our developer platform spans millions of years. many products and services that enable you, developers, to truly bring your ideas to life. Anything that you need, storage, compute, databases, if you need email for your application, agents, images, videos, you can find it all in one place on Cloudflare. And what's truly remarkable is that today, as Matthew said, there are over 10 million developers that are building Cloudflare, and that doesn't even include the agents. For the first time ever in history, the way that we're building software, though, is changing. It used to be pretty simple. As a developer, what, you write some code, you ship it, then a user logs on and uses it. That's it, right? Okay, maybe it was never quite that simple. I don't know if you guys remember three years ago, but we used to write all of that code by hand, every last semicolon, and then you would have to go in and test it locally to make sure that it worked. You would then push it up. A coworker of yours would review it. Eventually, it would make it to production. And then a customer would log on, start their day, and they would start clicking on buttons, right, and looking at dashboards. things and filling out forms and eventually they would run into a bug in that code that you wrote or they would find a feature that didn't exist quite yet and ask you for it and you as the developer would go back to the drawing board and the cycle would start all over again but this is now all changing because it's no longer just a developer or a team of developers it's a team of developers and a whole army of agents and on the other side of it two, it's no longer just the customer, but it's a customer and all of the agents that represent them. And as Matthew and Stephanie already talked about, agents and humans are fundamentally different. But the problem with that is that the cloud as we know it, both on the software development lifecycle side and the web itself, have only taken humans into consideration. So... It's time for a different cloud. Let's talk about what that means, and we'll start with the software development lifecycle. The SDLC was built for humans, but the thing about agents is they are incredibly efficient at things that we can be quite slow at. In fact, the part of the process that used to take the longest, writing out all of that code, that's actually now the shortest part of the journey. And agents are much faster at other things, too, reviewing code, deploying it. But there's one thing that's getting in the way of you being able to take full advantage of that. And it's that it's kind of scary to give your agents access to your production environment. Now, we're in San Francisco, the city of self-driving cars. But for us to get in that car for the very first time, we had to be really... confident that that car was going to be a better driver than we were, because inherently we trust other humans more than we trust machines, right? And now for you to be able to take advantage of all of the gains that are to be had from letting agents run loose in your code base, you need to be able to build up that same trust. And the tools to allow you to do so, that's exactly what we're building at Cloudflare along every single step of the We're reimagining the SDLC or the agent development life cycle. Let's walk through a few examples. With agents writing code in a brand new level of velocity, the tools that were built that assumed that humans were going to be writing all of that code, they can't keep up. They're not scaling, right? We're building more websites, more applications. There are more PRs being created. And, well, the source control that we've had so far, it just doesn't scale. So that's why earlier this year we announced Artifacts. Artifacts is version storage that speaks Git that was built from the beginning for agent-level velocity, and it just went into open beta two weeks ago. But uploading the code is just the first part of deploying it, right? Next, you need CICD. And that's... where, again, we're reimagining CICD for the era of agents because agents don't actually have to go through statically defined steps every single time. Agents can dynamically think about what's going to happen next, what features your customers are going to click on, and go and test those out, find bugs and fix them before they even make it into production. Lastly, for you to be able to have that trust, the thing you need most is visibility. And that's what we're providing you with Cloudflare traces. We want to give you the observability that you need along every step of the way into what your agent and what your code is doing in production, such that even when your humans are asleep, your agents can go in, find something that's going wrong, like an exception, fix it, and roll it out before the humans even notice. So that's the SDLC part of it. Now let's talk about the web or the applications, because the requirements are changing there rapidly, too. As Stephanie was saying, agents are capable of generating much, much more traffic than humans. So what was considered to be scale yesterday is going to be nothing tomorrow. And if you remember the bet that we took with isolates, well, this is the moment that they were built for. This is exactly what workers do. is really, really great at. It's being able to scale from zero to 100 instantaneously without you having to do any pre-provisioning and without you having to intervene in any way. Anything that you build on Cloudflare will just scale. And it's not about just executing the same code over and over again. The thing that's unique about agents is they require their own execution paths. And for them to be able to have that, what you need... is hundreds, thousands, millions, or even billions of little environments, little sandboxes. And again, that's exactly what we provide you with with dynamic workers, is that ability to spin up a sandbox, just run that code, and move on. But if we're talking about applications changing, well, agents are just another type of application. The thing that makes... it unique though is the thing that sits at the heart or I should say the brain of every single agent and that's the model which is why with workers AI and AI Gateway we're committed to always providing you and your agents with the latest and best available models on the market but that's not it just yesterday we announced a brand new model called cleric And it's not just a new model. It's a new type of model, an extraction model. It can take unstructured data. Imagine something like a receipt. You don't know how many items it's going to have. You don't know what all it will contain. But it can take that data, extract it, and give it to the agent. We think this is very complementary to something like decision models, like Clef, which we announced two weeks ago, which was Cloudflare's first ever owned model. It's performing at the top of the benchmarks today. It's incredible. incredibly fast, and available for anyone here to use. All right, I'll be honest. I could stand up here and talk about all of our products and features and ways to build agents for hours. It's a good thing that they didn't let me do that, but instead we have a whole two-day track that's dedicated to agents that has really incredible speakers. I encourage... every single one of you to check it out. And on that note of hearing less from me, I'm excited to welcome on stage Fabian Hadeen, the co-founder of Lovable, to talk about how and why an agent-native company is building on Cloudflare. Fabian, coming up. So good to have you here. Thank you for joining us. Thanks for having me. Very exciting to be here. Flew over from Stockholm, Sweden, where we're headquartered here just yesterday. I want to talk to you all about what we're doing. And I think Matthew did a great job kind of laying out the lay of the land of this authentic era and how creators are being empowered to create. more than ever. And that is what we're all about at Lovable as well, helping people go from their idea to actually making it real out in the world. I'll skip the slides and I'll just talk about Lovable. I heard you like to show rather than tell. Exactly, exactly. We're going to do a demo here as well. But I just want to talk a bit about the scale of what has happened here. So you've talked about how your kind of platform has grown so fast because of all of these new people being able to come and create. And we started in 2023 exploring this, and we set out on this mission to create something that anyone could use. And so we've really focused on simplicity. And so what is lovable then, actually? So on screen here, we have a prompt box. And you can ask Lovable to create things for you. And so I have a quick demo I thought I would do where I'm doing a session later today around some of more the technical deep dive around our partnership and how we're using Cloudflare to host all of these 70 million websites that we've created on our platform. And so I thought I would create a landing page for my session. So I'll ask Lovable here, hey, Lovable, can we create a quick landing page for my session I'm doing at Cloudflare Connect? I'll give you some quick context. And then I'll just paste in some info around my session. And so it's always fun doing these demos with AI, because as we know, it's non -deterministic. So we see what it actually puts out. No one knows. But... I'm hopeful, I'm confident, and so right now, as are you, yes, great, and so currently now I sent this prompt away, and so Lavable is beginning to understand my prompt, it's looking through some of this information I pasted from your website, and it's starting to write the actual code to build a landing page for us, and this would take me as a developer, maybe not this quick, right? So this took around 40 seconds or so for me, an hour maybe. But here we got some initial version of our site. And so I'll be talking about Tanslack Start and workers and all of this stuff. And if you're interested around this, come to my talk later today. I got a preview. But I'll paste another prompt here as well. And so we've seen that writing software has gotten really easy. And so we're all about removing the bottlenecks to creation. And so what we've started focusing a lot on is, okay, what happens actually after you create a software, after you write this code? And so the first thing, of course, here is that you want to host this. And so we use Cloudflare workers, dynamic workers. We use KV, R2. And then you also need a database. We help you set that up if you need payments, emails, and 200 plus connectors. We have the Cloudflare API as a connector as an example. But basically just helping you focus on the idea itself and not have to care about anything around really the execution about implementing this. And so you can just work on this high level of abstraction. And so I asked it to add a QA form here. at the bottom, and we see, oh, there it appears, and so now if you want to ask some questions for my session, you can do that here, and I'm quite happy with my site here, spending a few minutes on it, so I'll go here and I click publish, and Lovable generates a subdomain for us, I could add a real domain if I want, I'm happy with this, so I'll press publish, and hopefully, we'll have. this site live in a second or two. And so this is all hosted on Cloudflare with the workers. And it's something where it's become so easy to do this, right? And now I showed you a very simple kind of example of creating this landing page. It's spun up a database for us and other things in the background, but we didn't really have to care about it. And so we see people. everywhere from solo builders to two-thirds of the Fortune 500 use this platform to do everything from simple landing pages like this up to actually very complex applications such as SaaS platforms. And so this is what we're all about, empowering the 99 % people closest to the problem and enable them to take their ideas to reality. Thank you so much. Thank you for joining us. It really is the best time to be a builder. Thank you. Thank you, guys. I don't know if you guys noticed how quickly the application itself was updating after there were code changes being made. And actually, you have our next speaker to thank for that. He and his team are building the tools that enable applications to be built fast and be fast. So please help me welcome on stage Evan Yu. I've spent more than a decade building open source tools for the web, and for most of that time, the web stack was much harder than it needed to be. There were way too many moving pieces. It took too much effort to put together, and it was very easy to end up with slow feedback loops if you're not an expert. That was the problem I wanted to fix. I started at the framework layer with Vue and went deeper into the tool chain with VEAT. VEAT gives you instant feedback when you change your code, so you can stay in the flow. Today, VEAT is downloaded more than 200 million times every week. It powers applications built with mainstream frameworks like React, Vue, Astro. It also powers applications built by AI coding agents. That last part matters. AI can generate code very quickly, but if every build, test, and preview is slow, the agent is still slow. Faster creation requires a faster loop, particularly so for agents. At void zero, we set out to build a fast, unified toolchain for JavaScript and TypeScript. We shipped V plus as a result. Then, we started building Void, the native deployment platform on top of Cloudflare. That made something obvious to me. We didn't need to recreate a cloud. We needed to connect the tools we knew how to build with a platform that was already That's why joining Cloudflare made sense. My team can stay focused on compilers, bundlers, linters, test runners, and caching. We can keep that work open source and keep building with the community. And together with Cloudflare, we can carry the same fast experience all the way into production. The goal is one development model from the first file all the way to production. You should be able to use Cloudflare services locally, iterate quickly, and trust that what worked on your machine will work when you deploy. Less translation, less configuration, fewer surprises. Because speed isn't one fast build. Speed is preserving momentum across the entire process. But getting to production quickly is half the story. Once the application is in the world, your users need it to be fast, too. That includes people, and increasingly so, agents. Up to the end. person might wait on one page. An agent may make 20 calls in sequence. Stephanie showed you what distance does to an agent's calls. Every round trip compounds. Latency becomes part of whether the task finishes at all. This is why we automate optimizations in the tool chain and why the Cloudflare network matters after deployment. The application is built efficiently, delivered close to the user, and fast by default. Deploying globally is important, but global doesn't automatically mean fast. Over the past year, the number of requests handled by Cloudflare's proxy doubled. Normally, you'd expect that additional load to make the system slower. We saw the opposite. While serving twice as many requests, we cut processing time roughly in half. Both average performance and P90 performance improved substantially. More traffic entered the network, but each request spent less time inside it. That result didn't come from one feature. It came from improvements across the platform working together. A more efficient proxy, faster and more stable connections, and optimizations across routing, caching, and application performance. This is the other half of creating best-in-class experiences. We help you build and ship faster, then the network... keeps your app fast as demand grows. If you want to go deeper on the network, performance, and framework behind all of this, we have two full tracks over the next few days. Network and infrastructure, and web performance and frameworks. I'll be there, too, talking about developer tooling in the agentic era. I encourage you to join me. Now, we're going to show you what network performance means for a business. Without a CDN, Every page has to come from one server, far away. With Cloudflare, it comes from a city near the user. Less distance, less waiting, the page is just there. So rather than have me describe that from the outside, I'd like you to hear from the people who work on it every day. Please welcome Stephanie back to the stage, joined by our customers from Shopify and Tekovas. Start a new store. Let's do it. Well, I'm super excited for this session. I even... Kevin's team sold them to me, but Shopify, Tecovas, and Cloudflare all had a hand in how. So, Aparna, though, I'm going to start with you. When a shopper taps on a product, every second counts. Shopify storefronts are run on Oxygen, which is the world's best product name, but that runs on Cloudflare. So I want to talk about how you're making your sites faster for your merchants. Yeah, absolutely. So at Shopify, we are obsessed about performance and minimizing our latency. And like Stephanie mentioned, TechOvers runs on custom storefronts, which is our Oxygen's platform, and Oxygen runs on Cloudflare workers. So the first thing is the storefront is rendered in a location that's as close to the buyer as possible. So that's one. And then we cache as much as possible, like all the images. the scripts, the CSS. So it's all cached in the CDN layer. So it's also running as close to the buyer as possible. And then the third thing is we really, for the requests that have to go back to the origin, it uses Argo smart routing, which is the most optimized, fastest path to go back to the origin as possible. So those are the three things that we do. And at Shopping . we are also, like, paying a lot of attention to tail latencies because at Shopify scale, P99, when you're serving half a billion requests per minute, is millions of buyers. So we are continuously profiling the request paths and making sure that the request is served in the fastest possible manner. Okay, so, Kevin, I want to talk to you. Your boots are nice. I think you have 40 pairs. Those are bigger than mine. Close, yeah. Those are much bigger than mine. So you're online, and you started there, but you actually have stores across the country, and you decided to go all in on Shopify four years ago, and you run on Oxygen, which runs on Cloudflare. But I want to talk about what you've built on top of that foundation and how it goes all the way into the stores. Yeah, speed is important to us. It keeps customers moving and browsing through the entire Tecovis .com journey all the way through checkout. And it matters especially during our... product launches, and Q4 in the holidays, which is just a couple of weeks away. Shopify and Cloudflare take care of all of that infrastructure for us, and that enables my team to be able to focus on creating incredible experiences for our customers. And on top of that foundation and in this world of agentic development, we've been able to do even more on the site. We've built our own custom product recommendations engine. We've built an A-B testing platform system, and we've even deployed. our own techovus .com chat bot all built internally and hosted on cloudflare. And at techovus, we strive to serve our customers with radical hospitality. In stores, when someone comes in to try on boots, we want our associates focused on them. We don't want them disappearing into the stockroom or trying to coordinate something over a radio. And that's why we built Bootrunner to help deliver on this hospitality. the associate can request a size directly from Shopify point of sale there's a runner in the stock room who gets an alert immediately on their handheld and we also have a TV dashboard that shows the entire operations of the retail store in real time our HQ team can even check in on the entire fleet and see it from our entire macro perspective it's incredible and runs completely on cloudflare and delivers real time updates across all of the services it's built on top of Shopify POS UI extensions, 100% built with AI agents and maintained by AI agents. We aren't building a point-of-sale system. We're trying to build a Tecovus experience on top of it. So, bootrunner's one thing. Super cool, but you've built a lot more on top of Shopify and on top of Cloudflare. Can you talk a little bit more about what you've been building? Yeah, we are truly building on the shoulders of giants. Shopify provides us all of the commerce building blocks. And Cloudflare provides all of us the infrastructure and security. And when you add AI into that, it's incredible what teams can do. We're taking on business problems across the company that were way too big for us just a year ago. And now we're doing things like building our own homegrown BI platform with agents that serve data to all teams across the company. We've also built a supply chain command center that's built by the actual team members of the supply chain team. So they're able to help build the software. they need to get their jobs done my team's development has output has increased over 400 percent just this year alone so to go as a plus student but talk to us a part about what you're seeing across the entire Shopify merchant base yeah definitely AI has given all of us superpowers we are seeing this across the board and what I'd say is a the benefits of AI is threefold one is you know it has accelerated the innovation you know for people like Kevin all of our merchants. The other thing is it helps buyers find exactly what they're looking for. So if Stephanie goes to her favorite agent and says, like, I want black color boots, you know, no laces for my Cloudflare K -note, she gets taken straight to TechOvers, right? And it helps Kevin find the exact customer who's looking for exactly what he has. So, like, you know, what we call this as, like, merit-based discovery. So helping match buyers to merchants, and we don't want all of the merchants to take on the task of having to integrate with every single agent out there, so that's why Shopify has launched UCP, which is Universal Commerce Protocol. We co -developed it with Google, so what it does is it enables agents to buy directly with the merchants, so it's an open standard that specifies how an agent interacts with the merchant, and using their check. using their rules so yeah so like you know the shopping experience starts with the agent but ends with the merchant and every merchant on Shopify has UCP enabled by default so any agent can go shop on their website amazing well thank you Kevin thank you Aparna thank you both for being here thank you for being our partners I strongly recommend the boots. They're awesome. Tecovus is moving so fast, but they're doing it without giving up control of their brand, and they're getting ready for agents without having to rebuild their business. I'm now going to welcome on Matt Conroy. He's going to go through in greater detail how every business, not just Shopify and Tecovus, can retain control and optimize and monetize on the agentic Internet. I'm sad I didn't make the boot cut for the new boots. I must say before I get started, for a day that we're going to spend a lot of time talking about agents, it is so nice to see so many humans in one room. So, indulge me for a second and picture the internet. So, if you're like me, you probably still see people reading, shopping, booking a flight. But, as you've heard many times already today, that is just the tip of the iceberg. And at Cloudflare, we see what's underneath it all. And what we see are agents in astounding numbers. Last quarter, AI agents made more than 600 billion records. requests on our network, almost four times as many as at the end of last year. And in the time it takes me to finish this one sentence, we will see one million more. Now you just heard how Tecovus and Shopify think about every millisecond of a shopper's A person like me might check five stores, if you're lucky. But as agents do more of the shopping on our behalf, this number is set to explode. An agent may check hundreds of stores and ask every one of them for prices, sizes, and what's in stock. Now every one of those requests costs the store something, and most will never turn into a sale. But it's not that people have stopped coming. Human traffic on our network is up more than half since early 2025. But automated traffic has doubled and That means for almost every business traffic is growing faster than revenue Now our job is to help you serve the whole internet not just the part you can see in Practice this means giving you tools to control it to optimize for it and of course to monetize it Okay Control starts with no who's there. And with agents, we're really talking about three questions. Who is it? Can we trust them? And on whose behalf are they acting? Now, many AI companies already signed their agents' requests with WebBotAuth, OpenAI, AWS among them. And we're working with everyone building agents to make sure getting verified is easier than ever. And we're working with the ecosystem at large to help shape the rules as we go. But you might ask, what about the ones who choose not to verify who they are? Well, that turns out to be most of them. Fewer than one in ten automated requests comes from a bot that identifies itself. But not to worry. Our network sees enough to understand what they're up to anyway, and we will publish what we find on Cloudflare Radar. Of course, once you know who's there, you get to decide what they get soon agents will be able to vouch for the humans they're assisting with an anonymous ID the operators own guarantee or a verified identity we'll provide the controls but you get to make the call you can let a vouch for shopper check out you can turn away any agent that can't vouch for anyone and if a single person abuses an agent you can block that person without blocking everyone else who's doing it right. Now, that builds on controls that every Cloudflare customer already has, search, training, and agents, set separately and enforced at the network level so your choices hold. And as Stephanie mentioned earlier, underneath it all, adaptive intelligence, which retrains continuously on traffic across our network. So a new fraud technique that shows up today can be recognized tomorrow with nothing new for you to configure. That's control. But protecting what you've built is only a piece of this puzzle. You also want to be discovered. Matthew asked an important question this morning. When agents are doing the browsing, who actually gets found? This matters because if you're not discoverable to the agents your customers are using, You're effectively invisible to the humans behind them. Our agent engine optimization tool asks leading AI assistants the questions your customers are already asking, revealing how you stack up against competitors, in what context, and which sites shape the answer. And because we also see which AI tools are reading your site and what content they're most interested in, you can stop guessing. what AI says about your brand and start shaping it instead. But visibility is just the start. By early next year, our goal is to help you win more customers with AI and give you a bigger say in how your content shows up in AI Answers. Now, earlier this year, for the first time ever, crawlers collect data to train AI models were fetching more from the web than Google's own search crawler. Incredible right? And for every visitor that some AI platforms send back to a site, they crawl it thousands of times, sometimes tens of thousands of times, often rereading pages that haven't meaningfully changed. But we don't think getting found should cost so much. So we're. working directly with the companies that run these crawlers to help them crawl smarter. Sure, less waste, but also better discovery. I'm excited to announce that in a recent experiment with OpenAI, leveraging just 1% of our global network, we helped them find and index hundreds of millions of new URLs they had never seen before. This should be especially exciting for all the smaller sites, local businesses, and unique stories out there that have always struggled to get noticed. All right. That's all well and good, Matt, you might say, but we still need to talk about money. Remember that . checking hundreds of stores, well right now every one of those requests is a cost. But each one is also representing an agent that wants something that you have and that means it's not just traffic anymore, it's also a customer. And where there's a customer, there's value to exchange. Our monetization gateway lets you put a price on anything that passes through Cloudflare, a page, a data set, an API. even an MCP tool. An agent asks, we respond with the price, and it pays you directly. That means a sports stat site can charge a fraction of a cent every time an agent asks, who leads the league in assists? And when AI uses your work in a summary, that's an exchange too. With pay-per-use, AI companies can make you an offer. You can count. offer you can say yes and get paid and critically you can also see exactly what was used and what it earned but it's not just about getting paid for information exchange once agents can pay something bigger opens up as well whole new ways to shop and we want every business out there and in here to be able to build those agents and transact with them too imagine a world where you can tell an agent what you want your goal your timeline and your budget and it does the rest this is my personal favorite example taking my toddler skiing for the first time but to do that for that to work the agent has to understand and find what you want the merchant needs protection from fraud you need to get the right agent in the door we can help with that and we've just talked about how and that agent needs the right permissions and It needs to be able to check out with a real payment behind it. And speaking of that payment, the money has to move at the speed of software, which is why we back open standards like X402. So, you put all these together and you can control what reaches you, optimize to be found and monetize what you're worth. Now both sides of this market already run through our network. More than 20% of the web, and nearly 80% of the web, and nearly 80% of the web, and nearly 80% of leading AI companies. We can see the internet changing in real time, and that's why we can help provide what it needs, a network that can carry all that traffic, as Stephanie talked about, payments that keep up with it, and open standards that anyone can build on. Since day one, more than 16 years ago as of last month, helping build a better internet meant making it work for the people using it. Now, it also means making it work for the agents acting on our behalf and for all the businesses who intend to serve both. The Internet that you can see isn't going anywhere, but the part underneath is growing faster than ever. Build here, and you can be ready for both. Now, I am incredibly excited to dig into all of these topics and more during our many breakups. sessions this week so please come and find us and with that I'm going to pass it back to Stephanie and one of our fabulous partners who's working alongside us to make the future of agentic commerce you just saw there a reality master