Workers at 10, the Agentic Internet, & Slop Forks: Live from Cloudflare Connect 2026 with Rita Kozlov

Share via Twitter Share via Facebook Share via Linkedin Share via Reddit

Get more video from Redmonk, Subscribe!

Recorded live at Cloudflare Connect 2026 in San Francisco, RedMonk’s Kate Holterhoff and James Governor sit down with Rita Kozlov, Cloudflare’s VP of Developers and AI, as Workers turns ten. Rita traces how the platform began, from custom edge code written for enterprise customers to a sandboxed runtime built on isolates with no cold starts. She also explains how Cloudflare decides which primitives to build without simply rebuilding AWS. The conversation then turns to the agentic internet. Agents now generate more than half of web traffic, so what changes for identity, content compensation, payments, and discoverability through AEO? Rita also covers the acquisitions of Astro, Deno, and VoidZero, and Cloudflare’s responsibility as an open source steward. Other topics include Cloudflare’s first models, the vinext “slop fork” of Next.js, the future of Durable Objects and self-hosting, why Git and CI/CD are straining under agent-driven velocity, and why Cloudflare is the real neocloud.

This RedMonk video is sponsored by Cloudflare.

Links

Transcript

Announcer (00:07)
Ladies and gentlemen, please welcome to the stage Rita Kozlov, James Governor, and Kate Holterhoff.

Kate Holterhoff (00:14)
Well, hello, hello, hello. Wonderful to be here at Cloudflare Connect. My name is Kate Holterhoff. I’m a senior analyst at RedMonk. And with me today I have my, I guess, co-host, James Governor, co-founder of RedMonk, and of course our guest of honor, Rita Kozlov. She’s the VP of Developers and AI. Rita, thanks so much for coming here to chat with us about all the exciting stuff that’s going on today.

Rita Kozlov (00:36)
It’s funny for you to say thank you for coming here ’cause I’d like to thank you for coming here. Region Earth is what we like to say.

Kate Holterhoff (00:40)
Well, where is here, really?

You’re so good at segues. It

James Governor (00:50)
That was amazing.

Kate Holterhoff (00:51)
really was. so here being a state of mind, that’s that’s how we’re gonna kind of frame this conversation. Okay, amazing. Well, Rita, before we dive into all the exciting stuff that we heard about at the keynote today, I would love for you to just tell us a little bit about your background because I think it actually contextualizes a lot of the questions that we’re gonna ask. So tell us why you are

James Governor (01:12)
I’m gonna I’m I’m gonna I’m jump I’m already jumping in. I’m jumping in because you made me feel very old in the keynote this morning. And I mean I I you know I have I have kids who make me feel old every day. But you said when you told me that Workers was ten years old, I was like, my god, Rita, did you have to say that?

Rita Kozlov (01:31)
Imagine how I feel. Okay.

James Governor (01:34)
It’s like ten years. my goodness.

Rita Kozlov (01:37)
I thought that the comment that I made that made you feel old was when I asked you if you remembered three years ago when we wrote code by hand, which I feel like at some point our kids are gonna be like, Okay, grandma, let’s put you to bed

James Governor (01:50)
hundred percent. Yeah, yeah, yeah. No, I’m ready for that.

Rita Kozlov (01:52)
but to answer your question. Yes. Yes. so I started my career in software engineering. I feel like unsurprisingly now I get to build products for developers, which that that’s why it’s so near and dear to my heart is I’ve experienced I feel like every struggle out there that a developer gets to experience. and I joined Cloudflare initially actually as a solutions engineer. So helping our enterprise customers

deploy Cloudflare for their different needs and purposes. And Cloudflare ten years ago was a very different company.

James Governor (02:29)
It was not a developer company ten years ago.

Rita Kozlov (02:32)
You know, in in some ways it always was, both in spirit. we are just I think extremely nerdy at our very core. but also, you know, if if you think about Cloudflare’s original products, developers did use them, right? Like you need DNS, you need CDN, any website that you have, you need an SSL certificate. So I think we were always for developers, but

maybe at a different point in your journey, right? So you weren’t you were building elsewhere and then you were putting Cloudflare in front to help you scale and help you with all of the other, you know, help you secure your website, help you get a cert, all of help help you put it online because that’s what DNS helps with. but the building itself happened elsewhere, whereas now you can do the building itself on Cloudflare.

James Governor (03:20)
Yeah, I think that’s the big difference. And you know, that’s a transition that you’ve been responsible for. So tell us a bit about that journey. because as I say, at the time, it wasn’t it it wasn’t yet clear perhaps that that Cloudflare was going to be meeting developers at that point of their journey. So yeah, this this this 10-year overnight success story, tell us a bit about that.

Rita Kozlov (03:44)
ten year or overnight success story is a great way to put it. when we first started building the platform, I guess we we we didn’t know that we’re gonna build a platform, first of all. the there are two ways in which the way that it came about is actually very Cloudflare. One is we often the the products that we build come out of a need where we build something for ourselves first and foremost. So for example, Sam was up there today demoing Cloudflare OS.

That was something that came out of a need that we had where, you know, we need to give AI to like 5,000 employees. Not all of them are engineers. Nothing existed on the market. So we built it. Workers was kind of similar where we wanted our engineers to be able to go faster. And everything up until that point was kind of deployed as its own microservice. So we thought, you know, how do we build a developer platform that can enable us to go faster?

but secondly, and and in a in a different way, we also built it for our customers because at the time, and as I said, I joined Cloudflare as a solutions engineer. And we’re closing some of our first, you know, really large customers. And if you work at a startup, this is gonna sound pretty familiar to you. Your first enterprise customer comes along and they’re like, Do you do X? And you’re like, Yeah. And in the meantime, you’re you know DMing all of your engineers and you’re like, Can we build that? How quickly can we build that?

and so the the way that we would solve that problem is we had something called edge side code. And so basically if you needed to configure your CDN to a certain degree, where it was like, I want you to cache something, but only if this header exists and if it’s coming from this country, then do a redirect. we would have to write this custom code behind the scenes, and that wasn’t really, you know, it wasn’t a particularly scalable approach. So then we started thinking about, okay, well, how do we enable

our customers to self service and write their own code. But how do we do that without a creating cold start, which is what would happen if we put a container there? And how do we do that B in a way that doesn’t create a massive security incident every day? So it has to be highly sandboxed, right? And so out of like these three problems, and the third one being kind of the internal developer platforms, Workers kind of emerged and it was you know,

Let’s put it out there and see what happens and see how it goes. And the initial use cases were kind of more these edge use cases. But one of the things that quickly became apparent I thought was really interesting. So I was the first PM for it. And the interesting thing about building a product like this, but for you know, people that are looking to maybe modify their WAF a bit or their CDM configuration is they were like, well, I don’t write code.

I can you give me a very nice GUI where I can select what headers I want to add or block countries? and so I was like, okay, well then who is my audience for this thing? And it turned out to be very, very useful for developers. But then quickly developers were like, Well, I if you want this is very useful, but I also need some storage. So then we added KV. And then you know they they kept coming back with more demands.

And I kept saying, All right, let’s build it.

James Governor (07:03)
More and more primitives. From that perspective, I’d I’d love to ask you a bit about those design decisions. because you know, when I when I talk about the the the tenure overnight success story, you’re like, wait, we’ve succeeded from day one. What are you talking about, James? But I think the point about being appropriate for agents and some of the decision making that you’d made early on have been very appropriate. As you say, cold starts and security sandboxed. I mean, sandboxes is a is just such an explosive market right now. Everyone

Everyone is putting forward a sandbox platform and you’re like, well, we’ve been doing that for a little while now. So tell us a bit about those decisions that that, you know, like isolates at the time we were like, okay. And now it’s like, well, of course. So yeah, what what were when it came to the primitives, the design decisions that you because as you say, customers always want a bunch of things, developers always want a bunch a bunch of things. You have to say no as well. So what sort of leads you from a design perspective?

on the the decisions around primitives.

Rita Kozlov (08:05)
So two things. One is it’s innovation that comes out of constraints, right? So I was just saying if you’re a C DN and you’re looking to modify your C DN, but then if we put a cold start in the middle of it because we put a container there, that’d be completely unacceptable, right? Like you can’t be faster if to be faster you have to be thirty seconds slower. so so that was part of what was driving that decision. But on a continuous basis, so interestingly, and we

Do now have a container product. It’s great. It’s actually sandboxes are a big use case for it. And the way that we think about actually a container’s product is still a bit unique in its own way. But we we had this debate internally for a while, right? Of like, do we release this? Do we not? And we exercised restraint for a very long time. And I think that that comes from, first of all, Matthew and Michelle are both students of Clay Christensen.

the innovators dilemma kind of runs deep through the veins of Cloudflare. And a question that we found ourselves asking ourselves? Is that a can I say that?

James Governor (09:14)
I think he just did, yep. Always say that.

Rita Kozlov (09:16)
Great. so a question that we asked ourselves, I suppose, is w what’s our unique value proposition or how do we not just rebuild AWS? Because you can’t

You can’t confront a giant head on, right? That that’s not quite how it works. we knew that we didn’t want to compete, you know, directly on their own turf. And so that’s where we thought, okay, ISOLETS gives us a very unique capability and a u unique architecture where, you know, some of that comes from running on the edge and close to users, but I think a thing that we internalized pretty quickly is that even if Workers always ran in a single location.

It was very differentiated in that way that it ran. And again, obviously we couldn’t have predicted agents, but that idea that we could do something a hundred times more efficiently, that’s a reason to enter a market. If we were just going to be, you know, playing for, we’re gonna get ten percent of the container market, that that just didn’t make sense.

Kate Holterhoff (10:16)
If I can offer an observation here because you brought up AWS, so I’m just gonna go at it. so last year Twitter was blowing up because there was a bit of a feud between Vercel and the folks at Cloudflare. This year, Matthew Prince steps out on stage, immediately starts swinging punches at AWS. You know, it’s a lot of aggression here. I just, you know, I it’s an observation.

Rita Kozlov (10:37)
I I’d like to think that we don’t come from a place of aggression necessarily. Sure. Vercel for a long time was actually a customer of ours, so we generally have been pretty friendly with them. I

James Governor (10:53)
Not recently you haven’t.

Rita Kozlov (10:54)
Well, I y you know what the irony of it is, is in in some ways all press is good press, right?

James Governor (11:00)
There go. Yeah, no, of course. Having that having that conversation. Yeah. that very excitable conversation. Certainly, you know, drive drives clicks. yeah,

Kate Holterhoff (11:10)
It’s an era of heated rivalry.

James Governor (11:11)
there you go. Yeah. Yeah. Outrage, outrage and confirm confirmation bias are the most the the the key forces in the universe, I believe. And certainly outrage is yeah. it’s been quite funny watching people go at it like that.

Rita Kozlov (11:27)
Ultimately, we’ve always from the beginning viewed AWS as the ultimate competitor, right? we’re we’re both infrastructure companies. and yeah, it wa it was interesting to us that, you know, I think a big portion of our team thought that they were headed to re:Invent after Thanksgiving. And well, a large portion of them still is, just maybe to Vegas instead of re:Invent.

James Governor (11:52)
Too funny. Well, hopefully some of them will come to RedMonk Beers. Absolutely at on Wednesday night at re:Invent. Always a good time. And hopefully some some of our some of our our lovely listeners and viewers will also join us there. Yeah, I mean I think you’re gonna be doing some that does sound quite funny. Maybe yeah, it’s the problem is there isn’t much space there that isn’t already taken. Like it’s hard to do things like they literally have taken all of the space. So it’ll be interesting to see how you find

Those off grid activities.

Rita Kozlov (12:22)
Mm-hmm. Yeah. I mean V Vegas is pretty big though. Yeah. The there is gotta I think I I think we were planning on doing things like happy hours and stuff anyway. So I imagine those are booked and then yeah, honestly I think those are the more fun parts of conferences and re re:Invent itself.

James Governor (12:39)
So let me ask you a bit about that. So I asked about you know the primitives and y you’re talking about that we don’t want to just replicate what AWS has done. It needs to be differentiated. And certainly from Matthew’s perspective, when you talk to him from a from a product management perspective, one of the points he really is at pains to put across is that you’re you’re not trying to be a new AWS, that you are in fact about the network.

rather than all of the storage, you know, and it’s it’s the y you if you think about your whole approach to egress. that said, you’re looking more and more like a center of gravity now. I mean, it’s pretty hard to avoid that, isn’t it? So how does that affect the thinking of the company? Not just a platform, but a center of gravity as well.

Rita Kozlov (13:29)
when you say gravity, how do you mean that?

James Governor (13:34)
I think that that and you know, I remember talking to Matthew and he’s like, There there are certain things that we we would we wouldn’t we feel we wouldn’t need to do. Yep. And I think you’re beginning to do things that perhaps you said you weren’t gonna do.

Rita Kozlov (13:46)
I would be surprised if we verbatim said we weren’t going to do no, there are things that are sacred cows, right? I can tell you with I would say pretty absolute certainty that Cloudflare will never not have a free tier. certainly he literally tweeted the other day, not as long as Matthew is CEO. That’s a that’s a never. but as far as building things, even internally, we always say, you know, never for now.

that there are things that are less suitable for the way that we’re architected. So for example, one of the products that we offer is Workers AI, right? It’s our inference product. And when we entered the space, we said, training on us probably doesn’t make sense. Yes. right? We have a global network, and so we’re always thinking about what’s the best way to leverage it.

And what’s going to make the most sense. and whenever we do announce new products, even if it’s something that maybe you didn’t expect that we would venture in that direction, we still try to think about what’s something differentiated that we can do there because we have leverage in some ways. So for example, maybe something that I don’t know. funnily enough, I think our our two might have been around for five years ago for for five for five years now.

James Governor (15:05)
You make me feel old again there.

Rita Kozlov (15:07)
I know. I’m making myself y’all. but something that let’s say when you had a conversation a while ago with Matthew, he might have said we’re not gonna do is object storage, right? And then we launched R2. But R2 has always come with free egress, right? So that’s the unique thing that we’re able to offer because of the leverage that we have across our network and capability to uniquely offer that. So even those things, again, we still

try to find an answer for like why should we enter this market. Now increasingly with the developer platform, the more pieces you have, the more there are gonna be things where it’s like, okay, developers are asking for this. It’s a missing building block. We just need to give it to people and it will help fill that gap. So there’s not maybe always going to be that unique differentiation. There are going to be things that are just kind of that glue product. But for the most part, part of our

process for launching new products is thinking about, okay, like, why why not partner here? And we do partner a lot, right? why, why should we specifically offer this?

James Governor (16:18)
Mm-hmm.

Kate Holterhoff (16:19)
Okay, I want to ask you about the agentic internet. So there are posters saying this. It felt like it was a big part of your keynote. talk to me about why this is so important. And maybe for folks who didn’t get a chance to catch it, like what were some of the highlights about why we’re currently in the era of the agentic web, the agenc internet.

James Governor (16:37)
Great job getting us back on track. That was that was a Absolutely.

Kate Holterhoff (16:40)
I do what I can. Thank you for that. Professional.

Rita Kozlov (16:46)
so I I think w so much of we were thinking about, you know, how do we how do what what’s our what’s our thesis for this whole conference, right? And one of the things that kept coming up is humans and agents are very different. And we have this vantage point of we get to see the web. we get to see how it’s evolving. if you’re ever curious, we have our radar.

website which gives you that level of visibility too where you can kind of go explore data about like what’s the internet doing today and you can see an outage in a country maybe or you can see, yeah, what is the percentage of bot traffic?

James Governor (17:25)
were you were misinformation recently but because people put forward this the when Zelensky was saying that they had taken out one of the Yandex data centers and then suddenly there was no bot traffic, everyone was talking about Cloudflare and the radar and the data you had. And in fact it was some of it was quite spurious. People were yeah there’s no bot traffic it yeah so that was you you now hold a a a position as a you know

I guess a system of record and people are you know, creating misinformation based on that. I thought that was kind of

Rita Kozlov (18:00)
I thought that was interesting too. but yeah, it’s a it’s a point from which we can observe what’s going on in the web. And I think Matthew kind of opened the day with saying if you’d asked him a year ago when agents were going to overtake humans, he would have picked a date that’s still in the future. And it turned out that we’re already past that point. More than half the traffic on the web now is agents.

So we started thinking about well, what what does that mean? Because agents and humans, yeah, we we do things very differently. we humans need to sleep. I don’t know about you. I need my eight hours a night.

James Governor (18:39)
Hundred percent.

Rita Kozlov (18:40)
eight agents less so. humans, we we vote with our hearts often, right? I think that his point about, yeah, I I like going to the coffee shop that I go to because I get to

banter with a barista in the morning and she knows my order.

James Governor (18:57)
Agents, by the way, do kinda need to sleep, unless you have an infinite token but token budget. seem to spend a bit of time awaiting, going, that agent sell I can’t can’t use the agent. There you go.

Rita Kozlov (19:12)
but yeah, the so so if if we’re different in all these ways, and another one, right, is the the web, I I think one of the beautiful things about it is because it’s built for humans, it’s very visual, for example, right? agents are less visual, they’re kind of more information-based. And so with with all of these differences, you started thinking about well, what what does that mean and what needs to change.

And that that is kind of how we’re thinking about the agentic web. And one of the things is we anticipate that there will be a lot more just traffic periods. So how do we enable everyone to serve that traffic and do so in a way that’s easy to scale and affordable and all these things. That’s what I spend most of my time thinking about or I spend a lot of my time thinking about, okay, what does it mean to help developers build with agents and how do we make that smoother?

And then there are whole parts of the company. Matt Conroy talked about this quite a bit, right? That think about what does it mean for agents to be the ones that are doing the shopping now? Or what does it mean for content creators and how do we give agents access to content but in a way that in a way that still allows people to get compensated in the same way.

James Governor (20:25)
I think from the keynote this morning, that was one of the things that like I just don’t think the world is ready. you know, we’ve seen software software development. you only need to look at the travails of GitHub recently, when you put this much extra traffic, and traffic in slightly in in sort of patterns that you didn’t fully expect. as you say, like agents, they love documentation, right? They’ll just snarf all of that up. Whereas humans, slightly different approach to documentation. we take

Rita Kozlov (20:53)
Yeah, we tend to think that it’s optional.

James Governor (20:55)
Yeah, exactly, right? So I I think that when we look at the world of of transaction-based systems, I mean it really reminds me I’m old enough to sort of remember like what the web kicking off and what this meant in terms of scale. But I don’t think organizations that are out there that have transaction-based platforms are anywhere near ready for the levels of traffic that agents are going to generate. And I I thought that was useful that Matthew was perhaps helping people to to understand that, because we’re gonna need

We are gonna need an architecture that protects those back ends. And obviously I guess that’s a a a big part of what you’re building now.

Rita Kozlov (21:32)
Yeah, I mean I think it’s a combination of protects and supports, right? And I think that that’s where some of the tension is because I I think some of the initial thinking was, and I think a lot of the organizations that I talk to, some of their instinct is around the way that we were thinking about bots a couple of years ago before agents, right? Which is to say not human bad, human good.

but now the thing about agents is they do represent humans a lot of the time. So it’s not as black and white. And your business in some ways does depend on allowing agents to act on behalf of humans. So I I think yeah, what we need to provide is the infrastructure and the systems that will then enable people to kind of more granularly decide like how do I want to manage this for my

James Governor (22:23)
I got this is Jason Langsdorf’s question. We were talking to Jason just before this. He was really interested in this question about identity. I think identity in this agent era is a genetic era is very interesting because they need to be ephemeral. sometimes it’s delegated. we want it to be delegated perhaps for a certain amount of time. at the moment, you know, we’ve got a a lot of you know, you you it doesn’t really make sense to give agents all the bishop permissions that we have. that way madness lies.

So tell us a bit about where you are in terms of the agent identity and what you as somebody that is is building products for developers, how can you help them manage those identities? So what are you announcing? What’s what’s happening there?

Rita Kozlov (23:05)
Be honest, that’s less my personal area of expertise. there are people that are smarter than me that are thinking about this. I think a lot of what we’re trying to just kind of lay the foundation around for now is you know, standards like web bot auth and making sure that agents just first of all can identify themselves transparently and in a way that people know how to handle. But I think for solving something like agent identity, it’s going to require participation across many more.

Organizations so that, yeah, however an agent represents itself is consistent across several platforms. But personally, I would love nothing more than to be able to, you know, in some way delegate my identity, right? Like my number one use case that I would love an agent to solve for me is I would love to never have to manually make an appointment at the DMV.

Kate Holterhoff (23:55)
Yeah.

James Governor (23:56)
Right.

Rita Kozlov (23:57)
But I would also like it to happen in such a way that only my agents can create my appointments at the DMV, you know.

James Governor (24:04)
Well, I think that’s part part of the point I was making about transactions is that certainly th from from a government perspective, yeah, humans are gonna just use agents to ask a lot more questions. and we’re gonna have to find ways to, you know, scale to meet that demand. and and yeah, as I say, I don’t I don’t think the world is fully ready for that. So that was that was that was a question that you you weren’t Kate, go ahead.

You’re like, James, shut up. You said you were just gonna you I you were like I said I was just gonna stay in the corner and shit post. Yeah. And I’m doing all the I did so.

Kate Holterhoff (24:40)
He said this

is how we framed it. absolutely not. No. I mean, I do I have so many questions about the identity as it relates to payments. And so there were kind of a lot of, you know, good examples there. But the one that I still have so many questions about is like, who ultimately is gonna be paying for these bots to have access? Like, and you know, there’s god, it’s the incentive structure is just so complex and so multifaceted. And

And I just I feel like it needs to be built out so much. So I I feel like there was such a rosy picture drawn during the the keynotes that we heard. and I’m there. I’m with you. You know, I’m very bullish. I I love using Claude. I I’m I’m all in. But at the same time I do worry. I mean, you at RedMonk we we create a lot of content. So we certainly want to be paid and compensated. and, you know, we have historically thought bots are a bet, bots bad, you know? but now bots are good? I I don’t know.

I’m I’m struggling with this one. Yeah, the the the who who is left paying for this confuses me. So I don’t know. Why maybe a a way of framing this as a question is like why is Cloudflare the best positioned right now to address these huge issues that are going to restructure a lot of folks in the way that they’ve historically done business?

Rita Kozlov (25:57)
Yeah, I mean I think what puts Cloudflare in in a good and an interesting position here is because so much of the web sits behind us, we can kind of act as a broker on behalf of it, right? so and provide like step one was just providing content creators with the tools to even start to think about this and

decide, you know, what do I want to allow or disallow today? And again, it it’s not just content creators, it varies so much business by business, right? If you’re a content creator and the content is the thing that you’re producing, then of course it’s important that you get compensated for it. And we can we’re thinking about how do we put the right mechanisms in place for for that, right? Whether it’s micropayments or something else.

or just allow you to block it so that only humans can access it. Or yeah, in the future, how do we allow you to propagate maybe your subscription to your agents so that they can pull from resources that you already have access to? but it actually, if you’re not in the content business, then inversely you want to allow agents to like scrape as much of you as possible or

the thing that we’re trying to do is rather than have all of that load come at once, one of the things that Matthew was talking talking about was at least provide signals to the labs, right? So that if the agent already has that content, it doesn’t need to, you know, generate more traffic for the sake of it. It has an idea of of staleness. But again, there there is is this new concept of like AEO in place of SEO.

And having agents be aware of what you’re doing. I mean, I feel like I’m learning all about that now and that lesson now in real time, building out the developer platform because the way the developers discover products is now by asking Claude or Codex, like, hey, build me this and what what do you recommend? And if you’re not one of the top recommendations, then

James Governor (28:08)
That’s been interesting from Red Monk’s perspective because you know w we the original sin of the company was giving our content away. So we would give away our analysis so that that would drive consulting for us. And it was partly because we were about open source. but then certainly when when blogging happened, it made perfect sense. This you know, this was a model that would make sense. over time, as as traffic patterns changed, we went through social, so that was great.

But when agents hit, I was a little bit worried because A, we’d seen the the traffic patterns certainly through X had reduced the the the traffic to our website. then there was this question about okay, models, people are gonna ask those maybe before they come to us. And I was a little bit concerned, is this existential? But it turns out from an AEO perspective, RedMonk isn’t a great place because we don’t have a paywall. So unlike some of our competitors that are like, here is information that the

the models are not being trained on. We are weighted and we are part of the discussion and increasingly as the models use citation, it turns out that we are okay. So I’m quite happy about that.

Rita Kozlov (29:20)
Yeah, I mean not only that, I I think that so many things as we’re navigating this in-between world, I think we’re all finding surprises in all of it, right? Like I feel like two years ago there was so much panic about software engineering jobs going away or other roles, right? And I feel like what we’ve seen so far is the market for engineering jobs has only grown.

And similarly, you know, I I think we’re seeing a different flavor of what you’re describing, but especially customer-facing roles have never been more valuable than now, right? The thing like agents lead to you, but then having those relationships, that’s something that the agents just can’t do, right? Is you know, spending that in-person time or having having that discernment that comes from experience, right?

people come to you because they want your opinion, not the aggregate necessarily, or like maybe they’ll take a first stab at the aggregate opinion of like, okay, what all is out there? but I I think as as humans that are still making decisions, we we do value social proof. We need that social validation of like, okay, well the the person whose opinion I really respect, what what do they think?

James Governor (30:35)
Hundred percent. So I have a question from Stephen O’Grady, my business partner, and he said, Can you remember where we put all of our Workers apps? ‘Cause he he’s he can’t even remember which ones he’s using the platform so much he’s like, wait, what where is this app running? So I guess that’s that’s a a a plus for you. So do you know where those apps are?

Rita Kozlov (30:52)
Well, would say ask your agent to be honest. You know what’s funny is i if I can say so myself, I feel like our dashboard has undergone a massive glow up. we’ve really invested heavily in improving our UI and UX. We have a really incredible team. And at the same time, I feel like especially on Twitter, my feed is full of people being like, you know, Cloudflare offers such a remarkable experience.

I’ve never had to log into the dashboard once. Everything just happens through my agents. So that that is something that so we are very much investing obviously in both. And yeah, if you if you ask your agent, you know, where are all my Workers and what do they all do? The and you know, can you go deploy these two to do these things? They are very, very good at that now.

James Governor (31:40)
To be fair, Claude did recommend you. So we there was an app built by a guy called Brian LeRoux some years ago. he it was it was basically a proof point for a framework that he was building on on Lambdas. And he built this platform called Begin. And it turns out that’s the only to-do app as an organization that we’ve ever found that we liked. And so when he deprecated the app, we were always curse you, Brian. We were angry for many years. And then this year we stopped being angry.

And so Steve used Claude to basically reverse engineer and build us a new begin. so now we have a to do app that hopefully we’re using. It still has a a few rough ed edges. But when he was talking to Claude, the first thing it said was, You don’t want to run off lambda. that is like way more complicated than it needs to be, and you should probably be using Workers. So there

Rita Kozlov (32:33)
Aw, Claude, what a homie.

James Governor (32:34)
you go, right?

Kate Holterhoff (32:37)
All right, so I come from the front end world and Cloudflare’s done some amazing acquisitions this year, Evan You was on stage today at the keynote, creator of Vite, and then Ryan Dahl’s Deno. So I mean, talk to me about the decisions that went behind some of these acquisitions. I mean, these are super important parts of the internet, and in the front end sphere specifically, so you know, JavaScript developers.

Rita Kozlov (33:00)
Yeah, I mean, JavaScript developers were kind of some of the first developers, first of all, to utilize Workers because Workers was always a JavaScript first platform. So we kind of thought that it was a good marriage between, okay, if you’re a front-end developer but you have to do back end stuff, Cloudflare should be the place that you do it. So that’s always been a big area of investment for us. I will say, you know, honestly, I first of all, very excited about these acquisitions.

All three of the teams. So Astro, Deno and VoidZero, they’re all just fantastic, fantastic human beings, fantastic technology. but but with that comes a lot of responsibility, right? When you acquire an open source project. and so we we our number one priority, especially kind of given how much of the front and dev world depends on, especially something like Vite.

Yes. Where it’s it’s not even a framework, right? It’s the tooling that’s behind every single framework out there. And so we we want to make sure that we’re good stewards of it. And so we’ve the the number one thing that our number one priority was like, okay, do you guys have the resources to continue to do what you’ve been doing? Because obviously development here has to continue.

But yeah, I I think it’s that that’s been kind of the primary goal is to make sure that those projects find a home where they can continue to keep going.

Kate Holterhoff (34:33)
Yeah. I can say a lot of Vue developers were extremely concerned when they heard about the acquisition. but the press release was very clear about, the distinction and how you are trying to make sure that you’re, good stewards of that community. I feel like it it shows that there is a lot of momentum and attention to to some of these projects that are just extremely important. And I mean gosh, from an aquahire

Perspective, it’s some great, great team members that you’re bringing on.

Rita Kozlov (34:57)
I couldn’t agree more.

Kate Holterhoff (34:57)
Yeah. Okay. well, I got another one. Okay. Well, so on on stage, you showed a slide with all these different things that were announced, I’m assuming during birthday week. so talk to me. What is the most exciting one? If you had to pick one of those, I mean it was very small font.

Rita Kozlov (35:16)
I have my first child on the way, so I feel like one day I will thank you, thank you. I didn’t mean that as like that’s my favorite announcement, but I feel like, you know, one day people will ask me who my favorite child is and I will experience the same dilemma.

James Governor (35:29)
You’re not gonna say Workers and get in trouble some future time.

Rita Kozlov (35:33)
I suppose in some ways that was my first. what was my favorite announcement? I mean the the one that I I guess I dedicated some time to during the keynote just because it’s also so different that I’m excited about was Clef and Clairic, which are our first models.

so that that was that was something that we did that was actually funnily enough, going back to what you said in the never category, we asked ourselves a few times, like, are we gonna put a model out there? And I remember telling my team, like, that’s not our thing, like guys, what? but then this this is actually a great instance of me being wrong. the team came back a week later, it was after Jev was launched, and they were like,

So we post trained this and it’s I was like, what are what are the benchmarks saying? And they were like, We’re they’re really good. Yeah. And we’re like, okay, let’s launch it. but but it’s actually well, what’s interesting about it is that it’s a really good fit for Cloudflare running this type of model because even though we’ve been running Workers AI for some time and it runs actually some really, really large models now.

against something but at a point in time we said, we’re not gonna do that. And then we said, actually, you know, one so if I think about capabilities that we have, one of the capabilities is being able to leverage capacity very intelligently, right? And build software layers that allow us to do so. And so even with large models, that that’s something that we get to apply a strength of ours. I will say this is very, very different with GPUs than it is with CPUs and a lot more difficult.

but the smaller models are are actually a really good fit. And I I think that what TypeSafe did with Jev is very clever in introducing this new type of interface. I feel like we overloaded LLMs in such a way where actually, yeah, you don’t need all these input or output tokens. Actually, what you want is something much simpler that’s kind of more

binary decision or classification, but you want it at that level of intelligence, right? That you get from the LM at that level of precision. and so I I’m excited to see what other system one models we can come up with from here.

Kate Holterhoff (38:00)
a good choice. and you’re reminding me that yeah, another thing that a lot of the developers I speak with ha have mentioned about Cloudflare this year is I guess you would you would call it a fork of Next.js and then WordPress, so there’s there’s these sort of interesting choices that you’ve made to I don’t know compete.

Rita Kozlov (38:20)
Slop fork. We embrace it.

Kate Holterhoff (38:22)
Okay, okay. You heard it here. you know what what motivated that? I you know, I a lot of front end developers were excited, they tried it out. I I know th there were some complaints about the Next.js fork did not work in the week after, but I’m sure it’s it’s just completely polished now. so yeah, I mean what you know, what’s the thinking Is there is there like a a strategic plan there?

Rita Kozlov (38:46)
it kind of ties to something that I I think I was saying before. We’re pretty bottoms up as a company, right? So not not everything is strategic in that way. I mean, actually we’ve we’ve had different initiatives to fork Next.js over the years, right? And if you read the original blog post, kind of the really interesting thing about it was we tried it in I think like October.

Right around the time of year, this time of year last year, right? And it we weren’t getting far enough and it was taking up a lot of time, even though you know th there were coding agents available at the time. Yeah. And then Steve Faulkner, who’s on our team.

James Governor (39:30)
He’s amazing. Love that guy.

Rita Kozlov (39:32)
I love that guy. He’s incredible. And funnily enough, he’s he’s not even an IC anymore, right? Like he’s he’s an engineering director who spends most of his time in meetings.

And then gets those like half hour bursts in between meetings to try to pursue his passions. So he was like, you know, let me try and see how far I can get. And I think that it’s really a testament to how far there was that step function improvement in models that happened in like November, December last year. And it I mean, this is a perfect example of how unbelievable it was, right? Was where we tried something a few months prior and then we tried it again. And like

Basically, you know, not to say that it was a one shot, but how far we’re able to get. It it was such a stark difference.

James Governor (40:19)
people out there that still say like LLM software doesn’t work. Like there are still there are people out there that are like, no, you can’t generate and I’m like, this is you’re just ignoring facts. I mean, yeah, the the Nov November transition was unbelievable in terms of capability.

Rita Kozlov (40:37)
And if you use actually Chat GPT sites, if you d deploy any sites using that, that all uses vinext under the hood. we have someone from the team that I I think is speaking at our agent track tomorrow to to talk about that. So yeah, it it’s grown quite a bit in maturity since we announced it as well and has you know, we we did a 1.0 release, I think either during Agents Week or Birthday Week, the two of which I get conflated at this point.

Kate Holterhoff (41:05)
I mean we we are here, we’re gathered here at Connect. I’m I guess I’m interested you you had your you were in the limelight up on the stage. What sessions are you most excited about? What should James and I be attending here?

Rita Kozlov (41:17)
it’s another favorite child type.

Kate Holterhoff (41:19)
Don’t pick yeah, I know.

James Governor (41:22)
By the way, the Lovable demo, you never clicked on the to go to the site. You built the site. We didn’t actually you know I was like, maybe you didn’t actually build the site. It was like it’s done. It was like you didn’t click the button.

Rita Kozlov (41:34)
that’s so funny. you know, you’re up there, the lights are so bright.

Kate Holterhoff (41:40)
I

didn’t notice.

James Governor (41:41)
Maybe

you just maybe you were just maybe you were just like shitting us. Maybe he didn’t actually do it.

Rita Kozlov (41:45)
That’s true. we’ll have to follow up and ask him to put the link up. Well he did press the button.

James Governor (41:50)
He’s gotta press the button.

Rita Kozlov (41:54)
No, we just did we clicked on publish site because I time takes longer on stage than it does when you’re holding a plank. Yeah, basically. So th that was, you know, taking a couple of seconds. And so I and then we watched it deploy and we’re like, Great, but you’re right, we never clicked on it to confirm that the results were you know, well

James Governor (42:13)
I was waiting. I wanted to see that out.

Rita Kozlov (42:16)
Wow, I love that you called that out. but you should attend his session and then you can huckle him. But

James Governor (42:21)
Okay, there you go.

Rita Kozlov (42:23)
I think he will actually dive into all of the details of how all of that works under the hood and how every single website under the hood uses TanStack and runs on. They were among the first customers for dynamic Workers, which was really cool. so I’m very much looking forward to that.

we have also a lot of, you know, you were talking about before about the front end acquisitions and Clowflare’s investments there. we have practically, I think, every a representative from every major framework, right? So Tanner Linsley is here. Yes. Fred is here from Astro. He’s gonna be giving a talk. obviously Evan is here. so yeah, I I’m

Very excited to to hear from all of them.

Kate Holterhoff (43:15)
Okay. I mean, for a while there was a new one coming out every week. So to say that they’re all here is quite ambitious. I mean that you know that would that would probably double the the size of the the attendance here. which how many folks are here?

Rita Kozlov (43:28)
n I think near five thousand. Five thousand It’s

Kate Holterhoff (43:30)
Five thousand it. That’s a lot more than last year.

Rita Kozlov (43:32)
a lot more than last year. I I’ll be honest with you, I’m glad that we’re in San Francisco, not in Vegas. It’s a lot more

James Governor (43:37)
Is it really a developer conference though?

Rita Kozlov (43:41)
How do you mean, James?

James Governor (43:43)
Well,

I mean, you know, especially look, I mean if if you’re gonna say we have ten million developers now and there’s only five thousand in the room and then you know, I I’m

Rita Kozlov (43:51)
Saying that four thousand are agents?

James Governor (43:53)
I mean they might be. I I’m I’m just I I’m I’m just saying. I you know, it could be maybe I I’m just asking questions.

Kate Holterhoff (43:58)
Just ask questions. Gadfly.

Rita Kozlov (44:01)
I

Kate Holterhoff (44:01)
my god.

Rita Kozlov (44:03)
if if those are deep fakes walking around the expo hall, they’re they’re pretty convincing. Yeah. no, but I mean honestly the the interesting thing about these conferences is Cloudflare does serve a very broad audience, right? And you kind of see that reflected in the keynote as well. Where we talked about so many different things that we do. for me, it’s definitely an exciting part of it is how much of our audience is developers. So that’s

who I spend my time talking to. but it’s yeah, it’s certainly a mix of, you know, security professionals and creators and the full spectrum.

James Governor (44:41)
Yeah.

Kate Holterhoff (44:41)
So one of the things that I didn’t hear addressed on stage, which shocked me, I’m kidding. But is Durable Objects. And the reason is because, RedMonk, we think a fair bit about moats. And it seems to me that Cloudflare has a real moat with that. so Party Kit, another acquisition, I think last year. and Sunil Pai. Speaking of great guys.

James Governor (45:02)
L

lovely lovely mm.

Rita Kozlov (45:04)
Did you guys go to the cold start last night?

James Governor (45:07)
We didn’t know.

Kate Holterhoff (45:08)
Jason’s literally in the audience. He can hear us . Awkward

James Governor (45:10)
We we we say we look we did not know about that.

Rita Kozlov (45:13)
I think Sunil has a future as a talk show host.

James Governor (45:18)
He’s amazing. There you

Kate Holterhoff (45:19)
yeah, absolutely.

James Governor (45:20)
go. Is it are you are you hiring him? Is it yeah, yeah, yeah, a hundred percent.

Rita Kozlov (45:24)
Yeah.

Kate Holterhoff (45:25)
offers in the email.

James Governor (45:27)
Jason

Lengstorf, amazing. Yeah, definitely if you want like TV quality shows for devs, he is the guy building them. Amazing, amazing organizations. So yeah, anyone that wants, you know, that that sort of I I mean I’m waiting to see them on on on like Amazon Prime. I think that’s that’s a a future thing that’s gonna happen. Or I I don’t know, maybe maybe maybe Cloudflare will be doing some streaming.

Rita Kozlov (45:53)
Maybe. What is he doing after Silo? We don’t know, you know?

Kate Holterhoff (45:59)
This is a good point. So okay, well so to frame this as a question, talk to me about the future. my god. The future of Durable Objects here. Like what you know, is there anything exciting going on around that?

Rita Kozlov (46:11)
yes. Yes. Yes. If there’s one thing that I actually regret not sneaking into my keynote and honestly, like just putting it together and deciding what I was going to talk about was in and of itself incredibly challenging. And so I opted for what are the things that you’re not naturally thinking about when you’re thinking about us strategy? which I feel like Durable Objects has had like many moments this year. So people are very much thinking about it.

I I think that actually in general this year has been a year of durable execution is something that’s being talked about more and more. And actually also going back to you know our conversation about acquisitions, celld was a project that we got very excited to see from Ryan Dahl. So part of the reason for the acquisition

And one of the things that we’re looking forward to is actually working with him and the rest of the team to improve the self-hosting story for Worker D and Durable Objects. So workerd, the Workers runtime, has been open source for some time now. but if we’re being honest here, it today, if you’re trying to self-host an environment outside of Cloudflare, there’s still a bit of elbow grease that you have to put into it. And so

one of the things that especially if you’re specifically for the Durable Objects component of it. So one of the things that we’re investing in is, you know, how do we allow anyone to be able to to deploy it in any environment. But also Durable Objects is such a core primitive to our platform. that yeah, that there are heavy investments that we’re making there in general. actually, so many of the announcements that we made during birthday week.

Are basically different implementations on top of Durable Objects. actually, if there is one talk that I will make a plug for, it’s Dylan’s. he’s one of the engineers on Artifacts. Artifacts, don’t tell anyone, is basically just you know, pretty syntax and stuff on top of Durable Objects. What in some ways?

James Governor (48:22)
So let’s let’s on that note, let’s talk about that. Because again, I think I think you know, spoke to Jason before this. obviously the just the whole world of of of CI has there well it now that’s something we have to ask questions.

Kate Holterhoff (48:35)
I

know you had a big X through CI/CD. I was like sh shots fired.

James Governor (48:39)
So so yeah, like what how do you see your role in helping developers with the challenges of I mean, you know, the questions being asked, will we be using Git? What does the future look like? w what what’s your view on that?

Rita Kozlov (48:55)
So we just held our we just held a competition actually for people to define what the next GitHub is going to look like for for agents. And I think the winners are getting announced tomorrow. So I’m not going to spoil that. I there were some really interesting submissions and there are really interesting takes on it. The the way that we’re thinking about it is maybe one level below that, which is again kind of at the more infrastructure level, right? So

yeah, giving agents access to storage that speaks Git protocol. And you can use that for, you know, more traditional kind of source control. Or one thing that I thought one use case that I thought was really interesting was someone was using it basically for like versioning for their customer configurations. there are so many infinite use cases. But then yeah, once we started and keep in mind also a lot of our customers are platforms so

think Lovable, think Base44, right? And so they they started having this challenge of they have all of their customers and all of their customers agents that are creating more code than ever before. And traditional Git solutions just don’t scale for that. So we gave them this tool first, but then the next thing that they asked for was like, well, do you have a GitHub actions equivalent? Right. And so

that that is kind of what we’re building. But then again, we started thinking about, okay, what are what does that mean in the era of agents? And do you want just yeah, do you want the same predefined YAML file? Do we want YAML at all? I I think the answer for a lot of developers and agents is actually maybe not, right? Like just let me write code. and so that’s that’s what we’re shipping next and what we’re thinking through.

Kate Holterhoff (50:52)
Okay. I mean that sounds good. However, I can imagine some developers are have a little anxiety when you start talking about ripping out their existing CI/CD.

Rita Kozlov (51:01)
Yeah, so we’re we’re not ripping it out. The the the the strike through was, you know, just get you p to perk up and pay attention.

James Governor (51:08)
You’re you’re gently augmenting it.

Kate Holterhoff (51:10)
Yeah. Yeah. Well.

James Governor (51:13)
Strike through.

Kate Holterhoff (51:15)
I don’t know. But y it was workflows. Is that that was what replaced it? Okay. Yes.

James Governor (51:20)
But we’ve certainly seen, I mean, look, the you know, if nothing else, you can’t

Rita Kozlov (51:26)
Still a pipeline that you want to run after you’ve hit you know.

Kate Holterhoff (51:31)
Get push.

James Governor (51:33)
And you know, pretty clearly at the moment there’s you know it agents are putting everything under strain. And downtime is something that developers don’t really like. And so I think it is it is kind of all up for grabs. I mean the the the tools that have been the foundation of SDLC for the past twenty years or something, you know, all of that is kind of up for grabs.

Kate Holterhoff (51:55)
Yeah.

Rita Kozlov (51:55)
Yeah, I mean something that I wasn’t expecting at all is so last year we had a held our first one and then this year we did our second one. we do these customer de developer customer advisory councils where we get a small group of our customers together. We did it in Park City, we went snowshoeing and skiing, it was very fun. but also it allows to have kind of a smaller forum for getting feedback from

you know, industry leaders, CTOs, heads of engineering at different industry leading companies. And the the goal is kind of to get feedback on our roadmap or understand what people are thinking about that’s top of mind. And we thought that it was gonna go in one direction. And then I just kind of asked this open ended question of like, okay, what what are the things that are slowing you down? And I I was kind of shocked by this was in February, how many people said like, yeah.

Git is just breaking down for us. Yep. We we can’t it’s not keeping up with the velocity at which we’re shipping to.

Kate Holterhoff (53:03)
That’s huge. So we are nearly out of time.

James Governor (53:07)
How did you know that? Did somebody speak in your ear? I’m like, are you are you watching the time? How do you how do you know we’re nearly out of time? I mean, just

Kate Holterhoff (53:14)
I know things.

James Governor (53:15)
end You’ve had

Rita Kozlov (53:15)
I saw Kate checking her Claude

James Governor (53:17)
enough of Rita You’re like, you know, let’s let’s get out of here. They want more. They want more. How can you what do you mean

Kate Holterhoff (53:22)
Okay. Absolutely not.

James Governor (53:24)
you’re reading the room? Read the room, Kate.

Kate Holterhoff (53:27)
Reading the room.

Calm down, James.

Rita. Give us a CTA here. What should folks that are listening to this, what what should be their big takeaway? what should they be thinking about when they the when they’re they’re reminiscing on on the wonder that was Cloudflare Connect 2026?

Rita Kozlov (53:48)
Hmm, it’s a good one. I have conversations with a lot of builders, right, at organizations. I I I think especially over the years, we would come in and we would pitch Cloudflare and as the developer platform has matured, I would kind of say, You can build anything on us now. And they were like, Well, I already have a cloud. I don’t really, I’m not looking for another cloud. and so what what I would urge people to think about

Is yeah, how is your cloud serving you in this new era? And encourage everyone to go go play around with Cloudflare’s products or go prompt your agent to, I suppose. we’ve actually, you know, when you were asking me about my favorite birthday week launch, the other one that I was thinking about is the CFCLI, which we got to again reinvent from the

very beginning and I think i invent in a very agent forward way. but yeah maybe the CTA is for your next weekend or work project prompt your agent to you know how would you build this with Cloudflare and tell us how it goes.

James Governor (55:03)
Yeah, there you go.

Kate Holterhoff (55:04)
That’s useful. All right. Well I I did the the introductions. James, would you go ahead and and sign us out here?

James Governor (55:10)
yeah, I could do that. I mean, I I was just gonna say on that note, by the way. So like one, yeah, I’m gonna talk some trying

Kate Holterhoff (55:15)
I know, I know. I can’t do acrobatic.

James Governor (55:16)
to wrap these These mics are amazing. I’m just you know, I can

Rita Kozlov (55:19)
Hear your voice.

James Governor (55:20)
my own voice is amazing. It’s amazing. Horrible face,

Kate Holterhoff (55:22)
Have you heard James’s voice?

James Governor (55:23)
great voice. the the I from my perspective at least, it it’s been quite interesting, this industry rubric around neo clouds. which basically just means companies with a bunch of GPUs. It’s kind of weird to me.

You it’s like we do one thing, inferencing, and that’s neo clouds. And I’m like, that’s not really a neocloud. I think if we look at cloud as was and cloud as it will be, RedMonk’s been kind of on the record that, you know, where you are, if there is neocloud, it’s probably Cloudflare, which seems like a good place to end the show, Kate.

Rita Kozlov (55:58)
Right. Wow, you really drove us home and like put the car in park. Yeah.

James Governor (56:02)
Yeah.

There you go. So, Rita, Kate, thanks so much. And that’s been another RedMonk Conversation, a podcast. We did it.

Kate Holterhoff (56:11)
Thanks for coming out. And

Rita Kozlov (56:12)
Thank guys.

Kate Holterhoff (56:13)
thanks, Rita.

James Governor (56:13)
That was great. Thank you so much.

Rita Kozlov (56:14)
That’s so much fun.

More in this series

Conversations (157)