In this episode of Leaders of Code, Jody Bailey, Chief Product and Technology Officer at Stack Overflow, sits down with Dane Knecht, the newly appointed Chief Technology Officer at Cloudflare.
Dane shares his excitement about the Model Context Protocol (MCP), exploring its potential impact on the future of technology. The discussion turns to the growing need for sustainable content monetization and fair compensation for creators in an AI-driven world, and how this connects to Cloudflare’s mission to build a better internet.
The conversation also:
Episode notes:
[Intro music]
EIRA MAY: Welcome to Leaders of Code. This is a segment of the Stack Overflow Podcast. On this show, we chat with leaders in business and engineering about the work they're doing, their challenges, their customers, how they go about building the best teams possible, how they're leveraging AI and much more. My name is Eira May, and I'm the B2B editor at Stack Overflow, and I will be your host today.
I am here chatting with Jody Bailey, who is the Chief Product and Technology Officer at Stack, and also with Dane Knecht, who is newly appointed Cloudflare CTO. Thank you both for making the time to come with us today. I wanted to just start off by asking Dane, you moved into this role as CTO of Cloudflare pretty recently after a long tenure at the company and other roles. So I just wanted to ask you about what that transition has been like.
DANE KNECHT: Yeah, no, I've been at Cloudflare for a little over 14 years. Kind of joined early on as first a product person when we were less than 30 people. And actually John Graham-Cumming joined just before that as our– he had the title of programmer, which later became a CTO.
But we were working hand in hand from the beginning, both building out all the early products together, building out the engineering teams, kind of growing with the company. And so, a lot of it is just things that we have been doing with the partner for a long time and now get to spend even more time with our customers, as well as help across the company with some broader technology standards.
JODY BAILEY: I'm curious, Dane, are you still responsible for product or are you solely focused on the technology side now?
DANE KNECHT: Yeah. So at Cloudflare, we have an interesting structure. At least for the last eight years, we’ve had multiple product engineering orgs. So we have a core product engineering org, which is today run by CJ Desai and it's where most of our products that generate most of our revenue come from. It's– and he's responsible for making sure kind of what we call innovation at scale.
I've been running what we call emerging tech incubation, which also was product engineering teams, about 15% of R&D. And our job was to really grow TAM, figure out how to build new products on top of our network that brought more value to our customers for every bit that flowed through us and kind of got to break the rules a little bit and kinda act as a small startup within a bigger org.
I still have that org today. So I still have that product– those product engineering teams. Then I also inherited the research team from John, which is kinda the third horizon. So if you kind of think of the first one is where the majority of the revenue comes from, the second one is where the new products are incubated, and then the research team, which is even kind of a farther time horizon, trying to figure out what new internet standards are coming, what new technology shifts are coming that we need to think about several years out.
JODY BAILEY: That’s super interesting. I'm curious how you evolved to that organizational structure because it’s a little bit different than others I've seen.
DANE KNECHT: Yeah. Matthew, our CEO, was a student of Clayton Christensen at HBS, The Innovator’s Dilemma, and wanted to make sure that we were always, you know, if anybody was gonna disrupt us, it was gonna be ourselves. And make sure that we had a team that didn't have the pressures of revenue each quarter so they could really build what we thought were the right things to grow the company for the long term.
And so he wanted that as a completely separate or that all reported up to– through him and didn't have to make those tradeoffs that you always wanna make when that deal comes in at the last minute and you want to make a few more customer commits.
JODY BAILEY: Makes sense. I love that way of thinking. At Stack, we're going through a bit of, I won't say it's “Innovator's Dilemma,” right, but AI has obviously been a disruptor in the core way that we've operated in the past in terms of question and answers and people coming, and it's created new opportunities for us. But it's one of those things that I think for most of us while we all know AI's been out there for decades, right, I mean, it's not like an overnight success, so to speak, but it kind of was. Right?
DANE KNECHT: Yeah.
JODY BAILEY: (laughs) And, in fact, Cloudflare has helped us a lot as we’ve looked at new ways of monetizing our business. You know, we've essentially got three different avenues, you know, a SaaS product, advertising, and then most recently we've launched a knowledge solutions product. So basically that's creating, offering access to our content for large language models, for agents, things like that. And one of the things that Cloudflare helped us with was preventing all the scraping, you know, that was happening in people taking advantage of content without going through the appropriate licensing processes.
DANE KNECHT: And that's something actually we're spending quite a bit of time on is, you know, Cloudflare's mission is to build a better internet. And part of build a better internet also means that we need to have a free and open internet where great content is created. We want to make sure that, you know, as this kind of transition is happening, content creators get to reap some of the reward because without them the open internet dies.
JODY BAILEY: Right.
DANE KNECHT: And so we've actually spent a lot of time with small and large media companies trying to think through that. Is there a different way to monetize the content? Is there a way to find an economic solution that works between the large language models and the publishers? Obviously ,a lot of some of the larger publishers have cut deals with large language models.
JODY BAILEY: Mm-hmm.
DANE KNECHT: But there's a huge long tail of amazing content out there and how should they be compensated? We are spending a lot of time working through that and trying to put together some interesting things that you'll be seeing from us over the summer.
JODY BAILEY: Excellent. Yeah. I know Ellora from my team spends quite a bit of time talking with folks at Cloudflare looking at ways that we can continue to evolve. I'm curious, I'm sure you've adopted AI internally in terms of building out new features and such, and I'm curious if there's anything in particular that you're especially excited about or proud of or think it's pretty cool.
DANE KNECHT: Yeah. Most recently, we did a demo day with Anthropic on– with MCP servers. You know, I've just been hugely fascinated with the MCP, the concept, model contact protocol, of the ability for your LMs and your clients to be able to connect essentially over a programmatic interface to different resources internally and control them and have the LLM actually do more than just summarize something, but actually take actions…
JODY BAILEY: Right.
DANE KNECHT: …And kind of really be your assistant. And we just saw a phenomenal demo built on top of us with MCP and I think that's really just kind of beginning of where it kind of goes from AI being this tool that everyone kind of sees the possibility, but really, you know, are we actually gaining the productivity gains? And actually– is it really changing our life? Which I think was a lot of kind of the last couple years.
JODY BAILEY: Right.
DANE KNECHT: So I think over the next couple years, I really think that the way I do my job is even gonna change in a much more dramatic way.
JODY BAILEY: Totally agree. I mean, the easiest thing, you know, that's changed is just my writing has improved a lot. Right? And it's gotten a lot faster, but we're also looking at MCP, right, in terms of making it easier for LLMs to integrate with content. So our SaaS product, you know, it's a Stack Overflow for teams so that you can have all your proprietary information. And while we've partnered with LLMs and provided access to the core content, and they do a great job with that, when it comes to the proprietary information, how do you incorporate that information?
So MCP protocol is one of the things we're looking at for that product. But also on the public platform because you see companies like ModeWorks and others building agents that are helping people code and then making it easier to interface with our data, you know, is kind of the flip side of that, enabling the LLMs. So I'm curious, you know, how the MCP standard has stood up for you so far? Has it met the needs or, you know, I know it's new, so…
DANE KNECHT: We love how open, Anthropic’s been to working with us on evolving the standard. Obviously it was originally built as something that you kind of run on your desktop and, you know, no authentication and no concept of remote. They were quick to take our feedback and work with us on evolving the standard to have both the remote MCP servers as well as adding authentication on top of that. And I think the standard is still very early and I think we'll continue to see it evolve quite a bit.
But for content, the whole idea that people are scraping a website that’s meant for a human and trying to decipher it, it's kind of wasteful. So the idea that, yeah, there's an MCP server that has access to the content, that has some contextual knowledge of how your content's structured and how the data and, and even what the data been used for in the past, and to be able to give even back better information, I think is a much better way than just kind of scraping the websites.
JODY BAILEY: Yeah. One of the things, and I don't know that it applies as directly to websites, but one of the things that we're looking forward to is the ability to push data back, right? So if you're interacting with an LLM and content writing code, as it were, maybe you have a discovery that you want to share with other folks– that's more or less what our platform is about. You know, being able to push that content back in a kind of a standard way, I think is gonna be super valuable for folks.
DANE KNECHT: I mean, it almost gives a new interface to Stack Overflow
JODY BAILEY: Exactly. Exactly. So one of the things that we’re seeing in the market is all these really small startups that are, you know, creating code editors or code writing agents, et cetera, and lots of small companies that are having really fast success, with AI and such, and I think it's, it's forcing people to think about how they operate internally. I don't know how old Cloudflare is– I mean, you've been there, you said 14 years?
DANE KNECHT: Mhmm.
JODY BAILEY: Yeah, so I mean, it's at least 14 years old, right? And Stack is like 16. So both of our companies have been around for a while. And I’m curious with AI with the different tooling, whether it's Windsurf or Copilot or Rep or whatever, how has that changed the way your team is thinking about or actually building code and how have you played a role in kind of advancing or controlling it?
DANE KNECHT: Yeah. There's kind of three ways I think about how Cloudflare think about AI. One is, what tools can we bring to make our employees more productive? The second one being how can we add AI and OMs into just the product itself into the configuration and showing customers the value of our products. And then, workers' AI, you know, giving people kind of the tools to AI enable all their applications, develop a platform for that.
On the productivity side, from our roots as a security company, we are very conscious about making sure that all of our customer information, our data, our code is very secure. And which has led us to being a little slow on the early-to-adopt tools until we had a good understanding of where our data was going and how it was being secured and making sure that it was never gonna be trained on.
And so, we kind of sat over the past most of the year watching other companies do that and having small experiments. And then, you know, in the past six months, we've kind of decided, okay, now is the time to really become an AI-first company. And that goes everything from how does sales and BDR operate down to how to developers code to how to PMs build specs ,and docs and figuring out which of the tools that, uh, you just want a good hammer and one of the large language models and their primary interfaces, which ones you want to go build your own tools, and hopefully build them on Cloudflare.
Cloudflare is our own best customer for our workers’ platform. I think we're now getting the place where we've been rolling those pilots out and, you know, really saying that just, it's not just the productivity that goes up, the quality goes up too in a lot of cases.
Everybody gets an editor and an intern in one, right? Someone that can give instant feedback, even challenge them as well as get them doing work in parallel, while you're focused on another task. One of the things I love the most is we've been using Claude Code, a lot of developers, and Claude Code, it's the command line. It's not, not like a gooey ID, you actually kind of just talk to it as you're building the code. And what one thing we ask the developers to do is we ask them to keep the transcript and save the transcript into the PR. And it's so cool to be able to just kinda open up a PR and actually to see how the developer was thinking and how they were problem solving.
Especially for some of our younger engineers to be able to go in there and just, you know, not just see the PR and see how it was solved, but actually see the intermediate steps that led the engineer down to that solution. I almost think that, you know, on an interview, if you do coding exercises, you almost should say AI is required and, and say, because everyone's gonna have to be using these tools. And it's, can you think in a way that is, uh, you know, kinda show, show your work is almost the answer there.
JODY BAILEY: It's funny you mentioned the interview 'cause that's exactly where my head went to. Right? It is like, how many times, you know, have you sat down in an interview and asked somebody to walk you through how they would solve a problem?
DANE KNECHT: Yeah.
JODY BAILEY: And it, it's kind of the transcript to the interview that's actually being used to build the code, which is super cool. Right?
DANE KNECHT: Yeah.
JODY BAILEY: Which also, you know, kind of goes to that– how do you help people…like if you don't have experience writing code, if you don’t have experience putting all the pieces together to think through that process, I’m sure you’ve heard this question too, like if AI becomes the junior developer, you know, how does somebody go from not a developer to a senior developer? Like how do you cross that gap in order to guide the AI?
DANE KNECHT: Yeah, and actually the way we've rolled out the pilot so far is only the most senior developers on some of our hardest code bases we're given the tools first.
We really want to understand what their productivity would look like. And actually I find that actually it makes the best developers just, you know, really, even better. It doesn't take a mediocre developer and make them great and it doesn't take someone that doesn't sweat the details and all of a sudden make them sweat the details.
Those things come through with the code and we actually have a rule that there actually is no vibe coding allowed at Cloudflare because we expect everyone to basically attest when they check in code that yes, AI might have ruined it or assisted, but I'm owning the code that I'm checking in. I fully understand it and am taking responsibility for it.
JODY BAILEY: What does vibe coding mean to you, right? Like, because on one hand you say no vibe coding, but on the other hand I'm thinking, well, isn't vibe coding just prompting it and getting the code and everything, But it sounds like maybe vibe coding is doing that, but not understanding it.
DANE KNECHT: Yeah. I think that, I mean, at least maybe that's just my definition of it is basically more of just telling AI, oh, build this for me, and then you can deploy it and start using it without actually digging into the details and being part of the process more.
JODY BAILEY: I like that. I mean, frankly, I just feel uncomfortable saying vibe coding in, in general (laughs). Having it be more engineered sounds better to me too. So the other thing I'm seeing and, and something I'm puzzling through myself, right, in terms of like becoming AI-first, I’ve seen a lot of announcements lately, you know, whether it's Box or Duolingo, or I think Zapier, or whatever, where CEOs are coming out and basically saying, hey, you're not gonna hire any people unless you've tried it with AI first, or some version of that. Did you all have any sort of internal declaration of you must do this or, you know, try that or has it been more organic and just kind of encouraging and such?
DANE KNECHT: So Cloudflare culturally from day one, we’ve always run hot, so we always kinda run very close to capacity. And so, I think anyone here, if you put a tool in front of them that you say that you're gonna get more capacity, everyone here wants to, wants to take advantage of it. You can't hire your way out of the problems. You really have to engineer your way out of the problems. This has just kind of just historically been the culture at Cloudflare.
JODY BAILEY: Yeah.
DANE KNECHT: You know, using AI is just another way of engineering your way out of the problem. And so, you know, a lot of it is still kind of the responsibility of the teams and the managers to make sure that they're kind of optimizing their teams and helping, you know, Cloudflare’s overall mission of building a better internet.
JODY BAILEY: Makes sense. Whatcha most excited about, like, anything going forward that you're super excited?
DANE KNECHT: Well, I mean, we've been working on the workers’ developer platform for almost eight years now. And Matthew early on said that a platform takes about 10 years to really develop and kind of hit its stride. And so we're closing it on that point. And, the Cloudflare workers’platform is very opinionated.
You know, if you build the way that is designed for Cloudflare workers, you instantly get this application that can be deployed anywhere, is horizontally scalable, has data localization, all these kind of not maybe day one requirements, but kind of day two requirements for an application. But it's not easy to port a legacy application over to us and so, you know, we've always been great for greenfield applications.
And what's really exciting for me is we're kind of hitting the point where the platform has matured and we're hitting this kind of natural inflection point in the world where I think there's be more new code written, you know, in the next couple years than there was the past 10 years, which it means there's an opportunity for all these new greenfield applications to be written on Cloudflare and really have people kind of focusing on the business value of their applications and not having to worry about what infrastructure they're running on or how they're gonna scale it. And being able to deploy something that both kind of meets the needs day one, as well as the needs of a fortune one-hundred fully scaled application.
JODY BAILEY: I'm a little embarrassed to say I'm not very familiar with the workers’ platform. Can you maybe just describe it a little bit more?
DANE KNECHT: Yeah. I mean, so initially we released it was a– kinda a serverless JavaScript functions platform where all these requests go through our platform and we wanted to make it programmable both for our customers as well as just make it easier for us internally to develop applications.
And then over the past eight years, we've continued to add more compute and data primitives to that to make it easier to build full stack applications from rolling out our R2, which is an S3 compatible blob storage to D1 and D.O to different database primitives to our real time turn web RTC components.
And then most recently–and then obviously adding GPU storage and that and being able to do inference outer edge and then most recently announcing containers, which will expand the type of compute that you can deploy to us as well.
JODY BAILEY: That's awesome. Going back to the adoption of AI or even AI in general have there been disruptions for you– like anything to figure out how to work around or adapt to?
DANE KNECHT: Going back to Cloudflare’s original core products where we are infrastructure protection and provide security performance that's sitting in front of web servers, you know, all your internet endpoints. A lot of the inference companies had this kind of asymmetric problem where the cost of the inference was so high that it led to a lot of people wanting to find ways around, from a fraud basis. The expansion of how people are trying to attack and get around those defense security protections has evolved quite a bit because of that. And so, we've had to really change our tactics and ways that we protect those sites.
JODY BAILEY: I imagine that that's a constant battle, right? As you protect more things, the attacks get more clever, more sophisticated, and it seems like AI makes all those things a little easier to create, which has to have a downstream impact on y'all too.
DANE KNECHT: Yeah.
JODY BAILEY: For us, AI, I think I alluded to, while it created opportunities from a business perspective, you know, it also introduced risk in terms of traffic to the site– people can get those easy answers through the different coding tools, et cetera. You know, I would say we're in a bit of a transition period ourselves as we kind of evolve from this canonical Q&A to expanding to a more true community with different ways of interacting and communicating with folks. So it's been interesting for me. You started on the product side, it sounds like maybe you were an engineer before you were product. I've just recently taken on the product organization here, you know, so I've kind of gone the other way a little bit and it's an interesting and fun experience to go from just being responsible for building and delivering things to being accountable for what you're building and delivering. So that's been interesting for me. You know, I'm curious, you know, for you, like what is the most rewarding part of, of your role?
DANE KNECHT: When I hear from customers how we've helped them, whether it's protect them from attacks, let them scale their business with just a few people because of the tools we provide to doing things that really changed the shape of the internet. Like when we doubled the cript internet overnight when we announced universal SSL or we rolled out IPv6 for everyone and doubled the amount of IPv6 on the internet. I think we have the opportunity to both shape the internet as well as make sure that everyone has access to build and share their content with the world so it’s pretty special.
EIRA MAY: Thank you for listening to Leaders of Code. If you have any questions or feedback, suggestions for folks you'd like to hear from, or topics you'd like to show to cover, you can email me at Emay@stackoverflow.com or you can always find me on LinkedIn as Eira May. So I wanted to thank Jody Bailey, he's our Chief Product and Technology Officer at Stack Overflow. Thank you so much for joining. Can you let folks know where they can find you online?
JODY BAILEY: Sure. I'm not super active online. I'd say LinkedIn is probably the best place to find me. It's Jody Bailey, find me there, or of course, at Stack Overflow.
EIRA MAY: Awesome. And we've also spoken today with Dane Knecht. He's the newly appointed CTO at Cloudflare. Dane, if people wanna learn more about what you've been working on or connect with you, where can they do that?
DANE KNECHT: Yeah, on LinkedIn as well at Dane Knecht on LinkedIn or Twitter at DOK2001. And, of course, learn more about the developer platform at Cloudflare.com.
EIRA MAY: Awesome. Thank you so much. We'll see you guys on the next episode of Leaders of Code.
[Outro music]