Cloudflare Connect 2025 Highlights: Common & Company

Yo, I came down to Cloudflare, Cloudflare connects or what? To let you know Hip Hop, Hip Hop, Hip Hop, Hip Hop To everybody out there, we building Cloudflare with Cloudflare This is how I style, yeah Sometimes I used to have wild hair, but now it's bald Yes, yes, y'all Constance and Artemis, I still won't phone I made the call and now I'm here From the Cloudflare, now I appear Yeah, peace y'all Hello everyone, we're back to episodes after a week in Las Vegas This episode will focus pretty much on our global event Our first global event that took place last week called Connect But before we go into that, let's prepare a bit of what's coming So next week, we're launching a special series on the Cloudflare blog Let's focus not only on building the better Internet But also all about Internet measurement, resilience and transparency This week, for example, there was the AWS outage on Monday But that's not it, we're not going to discuss that We're going to talk about the foundations of a faster, safer and more reliable web for everyone We'll share updates and announcements across five critical teams It will be about the web being faster with next generation network protocols That redefine performance, also safer through cutting edge cryptography And new security models And also more reliable by optimizing distributed systems to stand strong at global scale Of course, more private with technologies that protect privacy without compromise And more measurable Because to improve the Internet, we first have to understand it So we'll have many blog posts for you to read and understand Something that we've been doing and building and talking about Join us from October 24th to the 31st As we explore how we're advancing the Internet together And while we're at it, today, Friday, October 24th In the blog, we announced that Cloudflare is partnering not only with Visa That we announced already during Connect last week But also Mastercard to secure automated commerce with AI agents The partnership introduces trusted agent protocol and agent pay And it helps merchants to distinguish legitimate agents from malicious bots Web Bot Auth is used as the foundation for agent authentication Allowing networks like Cloudflare to verify traffic from AI shopping agents So Cloudflare will validate agent interactions using HTTP message signatures And provide tools for merchants to manage agent transactions Of course, this is really important in this agents era Where we want to trust agents with payments and things like that But trust is needed So this is all about that You can read the blog And of course, stay tuned, geek out for next week Hello everyone and welcome to This Week in Net We're in Las Vegas where the first global event from Cloudflare called Connect took place With customers, partners, developers coming from many countries in the world Over 60 all joined together here for a week of conversations, sessions Of even education, a lot of education and of course networking But also some announcements For example, partnerships Cloudflare did a partnership with payment companies For example, Visa Protocols are needed to have security That's all about this partnership We have in the show a few things For example, we have Common, the rapper, actor, artist really He talks a bit about creators, about AI, humans and bots in this situation He participated in our event The same with Kerry Washington The same with representatives from Shopify But also OpenAI for example This episode is mostly focused on simple conversations These are actually small clips from longer conversations For example, we have Kenton Varda, our principal systems engineer The creator of workers really That has a few things really cool to say about agents, MCP, but also performance We also have David Tuber, our network expert talking a bit And also James Hallward talking about the new workers website Workers are developers platform So that's it from me Let's start the insanity And viva Las Vegas And you and me You and me together Every day together Always I really fear That I'm losing my best friend I can't believe this could be the end And I won't want to know And I won't let it go If it's real, well, I don't want to know Yo, I came down to Cloudflare, Cloudflare connects So what? To let you know HIPHOP's hiphop, hiphop, hiphop So, how does it feel being in a tour show situation The vibrations coming out How... the adrenaline How is that? Oh man, it's like a joy, like, you know, I began rapping, the first time I arrived, I felt the joy because of the joy that it made others feel too. And tonight I was feeling that, I felt like, oh man, this is a joy to be able to perform here at Cloudflare and have an audience that, you know, may not particularly always come to this show, a hip-hop show, or my show, and like, and just enjoy the music. And it was fun, I felt like we were free as artists, and I just was so happy with my band and my team and the audience, like, it felt good seeing people coming and coming and like moving, like, to get into the music and get into the performance. So it was a very great joy, an incredible joy. You've been on stage today during Connect, many questions were asked by Stephanie. One of the things that I noticed is like the, that although AI is coming, humans will always like to see other humans perform, create, for example, if there's a symphony, no one will look if it's robots playing a symphony, but they will look if it's humans. So how do you feel about creation, even in the AI age? Well, I mean, I think creation in the AI age, we will always need love. We will always need inspiration. We will always need compassion. Those are some of the things that computers just don't do. And I think when you experience art, when you experience dance, music, photography, any of the creative outlets, you, you feel loved, you feel a person's experiences, you feel the compassion. But I do see that AI can be used for some great things. So if a human being is behind the, the direction of the AI, and then we combine like what humanity is with, with these ideas, then they can come out with something beautiful. And, you know, it can turn out to be something great, but we will always lead with, with the human, with the human being and humanity and connection. If you have a thought, a wish for the Internet for future years, the Internet in the AI age, what would that wish be? Really? I would love for us to feel like the things that we think of as human beings that matter to us, that, that we value, that we, the way we want to be treated, I would love for us to express that on the Internet. Like it's not always necessary to say something bad about something going on or to take somebody's likeness if you haven't had any consent from them or, you know, any, like you haven't given them like compensation, you haven't given them the credit, you haven't given them what they need, like even, even checked in with them. So I would love to see like more consideration and caring when it comes to the way we, the way we socialize on the Internet, the way we communicate on the Internet. In terms of creation, just to wrap things up, in terms of creation, what still excites you the most, even using technology sometimes, in what way do you use technology to create sometimes, to make it easier for you? Well, I mean, the thing that excites me first is to continue to grow as a person and as an artist and to spread that art to humanity, like in the best ways possible. So what excites me about what technology can do is to help me spread that, that message, help me spread that energy and that, that art that I love. And I've used it in ways where it's been an advantage for me to, to like be able to do studio sessions with, with, with other producers in other cities at times when I need that. But I also value how important it is to be in a room with them and to start the projects all together in the connection we have. But to see, and I'm excited about us using those advancements in what technology and the ways that align with our, with our integrity, align with our values and can enhance our creativity. Building Cloudflare with Cloudflare. Yeah, building Cloudflare with Cloudflare. That's it. Yeah. Okay. To everybody out there, we're building Cloudflare with Cloudflare. This is how I style, yeah. Sometimes I used to have wild hair, but now it's bald. Yes, yes, y'all. Constance and Artemis, I still won't fall. I made the call and now I'm here. From the Cloudflare, now I appear. Yeah. Peace out. Yo, I came down to Cloudflare, Cloudflare Connect. So, hello, James. Welcome to DC Connect. Hi, Joel. Thanks for having me. So, this is a special edition of This Week in Ed because we're at Connect Las Vegas. You're usually in Singapore. Yes. And now in the US again. Back, yeah. I mean, when I started at Cloudflare, I started in San Francisco, but moved to Singapore partway through in 2022. But it's always good to be back in the States. For sure. Actually, why not start with Connect? So, we're in Las Vegas. It's the middle of October, our main conference. Actually, the first of its kind because it's the global conference. Yeah. You've been here for a few days now. Yes. What is the feel of the global conference here? I mean, the energy is great. Like, it's great to see so many customers. It's also, I've got to be honest, there's a little bit of this where it's great to see everyone from Cloudflare, too. We're all distributed. We don't get to see each other. And like, beyond getting to see all the cool stuff that's going on, all the customer stuff that's fantastic and partners who are here as well. Get to see the whole team. And like, normally we're all spread out across the world. Like, I'm in Singapore. A lot of folks in the US. It's just really good to see all the familiar faces. But the energy has been really great. It's really big. There's many panels. In terms of these type of events, other than talking with colleagues, even customers, you get like some feedback in some of the new tools, new things that is being. I mean, always like getting to see stuff that people are working on. Like, that they show you and customers as well. Like, we've been doing this or had one. We had a hackathon last night. Really cool to see all the things people were coming up with there. And then it's just the conversations where things are coming up. Like, someone's like, oh, wow, I didn't know you guys did this. We should really consider doing this. Or we've been using one of your competitors. We don't like them. That was a conversation. Like, always a conversation I'm happy to have. So, yeah, it's just great. And like, this is the joy of doing things in person that you otherwise just don't get. Makes sense. One of the reasons I wanted to talk with you, there's many to be honest, but one of the reasons is actually the workers page. Workers are developers platform. It has been increasing in tools, workers AI, the new AI age with MCP servers, agents. There's a lot there. And this workers .Cloudflare.com new page is, in a way, a matter of putting everything together in a more ready to the AI age, ready to all the possibility that we've been actually heading. Yeah, yeah. How the challenge came about? Well, my aspiration for the developer platform is we are thought about as the fourth cloud. And we have a network that's capable of doing that. We have a team that's capable of doing that. The underlying architecture that's been come up, it's capable of doing that. One of the things that developers like they care a lot about. I mean, it's funny to say it that I'm sure many would reject it, but fashion like style design is really important to them. And when I joined way back seven years ago, there was a small group of us that pulled together a page for the developer platform. And for seven years ago, it was a pretty good effort, but it was very, very stale. We wanted to get something out there that was beautiful, that spoke to the capabilities and the potential of the developer platform. More than that, when people look at these sites, when developers look at these sites, it's almost like a tour de force of what the site is capable of. If it's a really cool, like if you've got a really good team, if your platform is really cool, the site tends to reflect that. And there's this zeitgeist that's associated with it. We really wanted to capture that in the site. So a small team of us pulled it together. We also got some help from an outside agency. I spent a lot of time going through those best of websites to find someone who really felt like they understood, they got Cloudflare. And we managed to find someone, a very small team on both sides, just a couple of Cloudflare people, a couple of people in the agency. And we worked for a fair few months, really hard to pull this together. And I'm not too ashamed of the way it turned out. You shouldn't, because it's really amazing. It's really, actually, it's geeky in the Cloudflare way, but done with real style, done with real details. And the amount of little gadgets that are included there is quite amazing to see. Your role at Cloudflare, what do you do, really? Yeah, so I work on our infrastructure engineering team. And so basically, we have this huge network, over 300 terabits of capacity, 350 cities, so much network, so many computers, so many data centers. And we have to deploy code into all of that stuff. And my team oversees all of that and basically makes it so that our network runs as reliably and as fast as possible. And so all of the stuff, all of the amazing products built on Cloudflare leverage our tools and systems. They kind of run under the hood to just keep Cloudflare humming. So we are kind of the little people in the dungeon who kind of keep the lights on, like shovel the coal into the furnace that keeps all of the water warm and all those movies. And every so often we get to pop out and say hi to remind people that we exist and are doing our best to keep Cloudflare reliable. Makes sense. Actually, because we're at Connect, do you have some feedback from customers, partners? Have you been around? Yeah, I mean, like, obviously, we talk to customers all the time. Generally, when we talk to customers, it's not because we're announcing some new product or anything like that. It's usually because, you know, something's broken. But that's OK. I mean, like, still, that feedback from customers is really good. And you hear like some pretty consistent things all around, which is like, hey, don't break, please, which is pretty obvious. But the second one is like, hey, if you're going to break, tell us when you break. And then the third one is, you know, make it so that a break doesn't impact us. And those two, the last two are actually very interesting problems to go solve. And, you know, we've kind of helped drive a lot of innovation behind products to actually make things better. So a really good example of that is kind of an innovation of Argo that we launched a couple of years ago called Orpheus, which is an initiative that kind of came from our team. And it happened basically because Argo obviously finds the fastest path back to a customer's origin. So if you are hosting a website that has cat pictures and Cloudflare can't reach your website that's hosted in OCI or whatever, then you can't serve the cat pictures. But if with Argo, in theory, you should be able to detect that the path that you're going over is slow or out and it'll route you around. And so the idea that we had was, well, what if we did that for everybody all the time? And instead of optimizing to find, you know, the fastest path, we optimized to find the most reliable, basically a path that we knew would get you to your origin. And so that was born out of a, hey, if something breaks, because stuff's going to break all the time, make it so that we don't feel the pain. And we spent a really long time building that and rolling that out. And it's called Orpheus. And it's great. It saves requests all the time. Basically, what it does is like whenever we try and reach an origin, if we get some network failure for any reason, we've got backup and we've got backups on backups and backups that we can just route traffic around and send it. And so like we build stuff like that to kind of help keep our products online. And so, you know, I'm sure that we've definitely spoken to a bunch of customers about, you know, the magic of Orpheus and how it's saving our traffic and stuff like that. It's kind of how we work to make Cloudflare as reliable and as performant as possible. Makes sense. There's a lot of machine learning in some of those processors, right? Not sure about the Arpys one, but in one way there's automation. There is a process there. How does it work, really? Yeah, I mean, like it's really just, you know, signal and actually it's not anything like, you know, you'd love to believe that there is like some complex series of equations. But it's actually really, really like a lot of automation for networking and stuff is is at its core very simple. There's a signal that goes that shows when you when something is bad and when that thing gets bad, then you do a thing to make it good again. And the hard part is not on the action to make it good. The hard part is actually making and finding that signal and getting that signal to be clear and unambiguous as to whether or not there's a problem. Because like the Internet is all gray area. Like, it's very rare that there's like it's very rare that like something is black and white out, like maybe a fiber cut happens and like you can't reach this path anymore. But like even then, it's actually not as black and white as you think. There's a lot of gray because there's a whole bunch of paths to an origin and or a whole bunch of paths from a source to a destination. And sometimes one fiber cut isn't going to just take everything down. Sometimes it's just going to make some requests that would try to go over that path fail. And that doesn't exactly show up in the millions and millions and millions of requests that we get every day. So, like, even in the things that you think are black and white, they actually end up being way more gray than you think they are. Big sense. Yeah, well, so what happened was there was one, I guess, someone on YouTube posted a video where they claimed they claimed that Cloudflare was faster than a competitor of ours called Bracel. And to prove it, they showed this benchmark in which they were doing a lot of math.sign, you know, trigonometry operations. And for some reason in this test, it ran three times faster on Cloudflare than on Bracel. And they said, ha ha. And then another YouTuber that's better known named Theo, the independent developer and YouTube personality, took issue with that, rightly so, and came up with his own set of benchmarks, which were designed to simulate CPU intensive workloads, which is actually not what most people are doing on either of these platforms. Usually, you're spending most of your time waiting for network communications, talking to your database, and so on, and not spending a lot of CPU time rendering a result. But for the purpose of this test, point was to measure raw JavaScript execution time. And he came up with a bunch of benchmarks. And in those benchmarks, it appeared to be that Cloudflare was slower by as much as like 3, 3.5x. And he published those. And we were very surprised because, you know, we... There's a big gap there. Well, there's a big gap. But also, like, we use the V8 JavaScript engine. Bracel uses Node .js, which uses the V8 JavaScript engine. It's the same code, essentially. And since these are CPU intensive benchmarks, they're not about network latency. They're not about talking to external services at all. It should just be JavaScript execution time. So why was there this disparity? So we all on the workers runtime team got together and said, let's figure out what happened here. And found actually a lot of things, a lot of little things, fixed a lot of things. During birthday week, outside of this benchmark experience, during birthday week, you had a few blog posts, actually, about agents, AI agents, about MCP servers. Some are actually trying to give a different perspective to these areas. Can you explain to us a bit of those blog posts, starting with the MCP servers changes you mentioned? Yeah, so Cloudflare Workers is a popular place to host MCP servers. For those that don't know, MCP is a way to connect AI agents to external services so that they can manipulate the world. But the way MCP usually works is it presents, an MCP server presents a set of tools, like functions that an LLM can call. But the way it does that, the LLM has a way of outputting a special sequence of characters that basically says, I want to call this tool. And we thought it would be interesting to take a different approach there. Instead of directly giving the tools to the LLM, we see that LLMs are surprisingly good at writing code. And we thought, what if we turned this set of tools into a TypeScript interface, and then presented that to the LLM and said, please write code against this interface to do a task. And we found this was pretty effective for a couple of reasons. Like one is, LLMs have read a lot of code in their training set, they've read all of GitHub and so on. And so they are very familiar with that language. Whereas the tool calling mechanism that LLMs usually use to call tools is not something they learn about from GitHub. It's something that they're sort of trained at the end of the training cycle, especially on like with a bunch of artificial data saying, here's how you do this. And it's actually harder for them. It's like when you're asked to present in a different language than your native language, right? So by giving them the exact same tools, except exposed as code in a programming language, they can actually be smarter about it. But then where it's really powerful is when they want to use multiple tools in sequence, and like feed the output of one into another, like you can write code that does that. Or if you need to run a bunch of tools in a loop, like loop over like the first one gets 50 results, then for each one of them, you need to do something. That's obviously super easy to do in a programming language. But if you have an LLM doing that through tool calls, it's going to have to read back all the output of every call into its neural network, process it back out into the next tool call. And that takes a long time, uses a lot of tokens, it's really expensive. Whereas code is fast. So I kind of think that the future of AI agents is going to be, they're not making tool calls, they are writing code all the time, behind the scenes, you may not see it most of the time, but it will be there. But the interesting part about that is, if you're going to have AI running code, you better hope that the code is run in a secure environment in a what we call a sandbox prevents it from doing anything bad. And if the AI is constantly writing little bits of code that need to be run all the time really fast, you need a lot of sandboxes, you need to spin them up really quick, they need to be disposable, you know, short lived. And it turns out that is a great fit for the workers platform, because in workers, we don't use containers, we use isolates and isolates as our unit of compute are way cheaper at the lower end, they're far less, you know, you start one up in a couple of milliseconds, it takes a couple of megabytes of memory, you can start it up, run one snippet of code, throw it away, that's efficient, whereas with containers, so most people are trying to do a lot of things at the same time. And to do this with takes a little while to spin up a container, and then you have to like, try to reuse it for things. But once you're reusing a sandbox, then you have to worry about is any data leaking between the different uses. And so it's a lot more expensive that way. And that's a wrap.