Video: App Modernization Best Practices in a Serverless World | Duration: 2296s | Summary: App Modernization Best Practices in a Serverless World | Chapters: Welcome and Introductions (71.995s), App Modernization Strategies (155.72499s), Enterprise Refactoring Drivers (363.39s), Refactoring Challenges Explored (553.64s), Developer Platform Advantages (888.04s), Serverless AI Workloads (1304.565s), Real-Time Architecture Challenges (1742.355s), Conference Wrap-Up Announcement (2167.43s)
Transcript for "App Modernization Best Practices in a Serverless World":
Welcome and hello to Cloudflare's webinar on app modernization best practices in a serverless world. Today, have a treat for you. We have Devin Dickerson from Forrester and David Liu from Cloudflare, and I'll play the host today. Devin, would you like to introduce yourself? Sure. First of all, thanks for having me. I'm Devin Dickerson, Principal Analyst at Forrester. My primary research coverage is cloud native development. I also look at application development targeted for edge environments and I lead our research on serverless in particular. Thank you. And David? Yeah. Hi everyone. I'm David Liu from Cloudflare, specifically working on product marketing for our developer platform. Okay. Thank you. So today, quick agenda for you is we'll have David kinda give you an introduction to what app modernization looks like and our approach at Cloudflare. Then we'll get into the meat and potatoes of this with a fireside chat with David and Devin, and we'll recap. We'll we'll talk about topics such as refactoring projects, the Cloudflare developer platform, and the most common workloads that we see in this area. Okay, so with that, David, why don't you kick us off? Yeah, absolutely. So on the topic of app modernization, it's really about transforming existing applications and their infrastructure, as well as building new ones using modern technology. The three common migration projects that we see within our customers, as well as variety of industries, starts with rehost, then replatforming, and then ultimately refactoring and building new things. And in terms of rehosting that first step is truly a lift and shift project where you're migrating your app and infrastructure as is from your on premise servers to the cloud and sometimes vice versa. But we mainly see that lift and shift from on premise to the cloud and there's typically no code changes to the underlying. Once a company has lifted and shifted, you're starting to then optimize these deployments. And so this is where replatforming comes into play. You're optimizing your app infrastructure with containers. It could be moving towards a serverless infrastructure, all for multi cloud and hybrid environments. And then what we're seeing now is companies are at the refactoring stage. They're building new applications on modern architecture, on modern technology. And this is super interesting for the Cloudflare Developer Platform and our conversation today. And so as part of that refactoring exercise, at Cloudflare, we see 12 workloads that a lot of customers are doing, right? And so within our customer base time and time again, these 12 workloads are repeating. And so it goes from AI training data and storing that on Cloudflare with zero egress fees, all the way to creating the next Zoom or Google Meet with real time audio and video workloads. And we'll explore some of these near the end where Devin and I will chat more about this. And the reason why companies are building on top of the software, especially those workloads that you saw is because of our serverless architecture. And what we really want is for companies to eliminate that 50% of their developer time that's not spent developing, but wasted on infrastructure management. We wanna free these developers to focus entirely on just building and shipping code rather than constantly having to re engineer table stake infrastructure such as disaster recovery, elastic scaling or sophisticated load balancing. We're really trying to make it something where our platform is smarter, data driven and automating away the complexity that developers face today. Awesome. Okay. So let's get into the fireside chat portion and bring in Devin. Devin, when we're thinking about refactoring as as projects, one of the first questions we we obviously get are in the enterprise world, what are the primary drivers pushing such large enterprises towards application refactoring efforts today? Yeah, that's a great question. I think that there's a few things at play actually. So one, and this is a constant, but it comes up again and again. And this is just the realization that the application kind of is the business these days. Right? It's not really an exaggeration to say that for most companies, the application is the business. It's your digital storefront. It's often your main channel for customer engagement. Some customers never deal with a person at a company. They deal with a website, a mobile app, increasingly maybe some kind of agentic experience. And so it's more than just a direct reflection of your brand. It is the brand. And, you know, it's been a long it's it's been decades since websites were simple brochures, things like that. We today, they need to be dynamic, interactive experiences and the user's entire customer journey from discovery, purchase, support, it all tends to happen through an application. So there's that need. Then we're seeing and I think this leads into something else, which is I I call it the expectation chasm. There there has been increasingly a shift, and I think it comes down to just the the difference between the user experiences people have when they're using the apps they use every day for their personal life versus what they use at work. But there's it's this user expectations gap. And many legacy systems, especially business systems, can't deliver on users' expectations. And these expectations are always on, always fast. It's not about your direct competitor sometimes. It's the fastest, most seamless app that they use every day. That's what you're up against, even with your internal business apps. I think that drives a lot of refactoring efforts. And the last one I'll throw out is there's always emerging frontiers to deal with, and this has always been a key driver of refactoring efforts. Today, the elephant in the room, AI and agentic applications. And so the nature of these core applications that are either customer facing or business facing, they're undergoing a seismic shift. I have conversations with enterprise leaders every day and the focus is really gravitating towards AI. And I think a lot of what we're seeing too is there's this massive wave of experimentation, building POCs, prototypes, everything from internals, internal stuff to customer facing agentic apps. A lot of legacy infrastructure and legacy application patterns are just not meant to extend into these types of use cases. And so I think I think when you just look at the world around us, it's changing faster than a lot of legacy apps can cope with. Love it, David. Can please continue? Yeah, I really like the point that, you know, Devin made where it's like the legacy infrastructure just can't keep up with these new technologies, right? Especially on the AI front. We get a lot of customers coming in where they want to have an internal AI agent or a chatbot. And one of the simplest things that we take for granted is when the chatbot has a memory of your previous conversations, right? And so it's stateful. Actually, I learned that that was like a super difficult problem with legacy infrastructure is like maintaining state. And so we're just seeing like more and more customers come to us because we have primitives out of the box where you can maintain state for some of these new AI led projects. David, I actually think that's a really incredible point because one thing that I didn't know until I started getting my hands dirty with the development tools and the SDKs associated with building agents and chatbots and those kinds of things is that some of the things that you get out of the box with the commercial products built on the frontier models, access to tools, stateful chat interactions, memory. It gives you the illusion that these things are just built in to any agentic or AI experience. They're not. When you go start going through the SDK docs and and start building these things yourself, you realize that all of all of the ways and methodologies to do this are opinionated and require actions from the developer. Abstractions are really key here. And what you get out of the developer platforms and tools ends up being extraordinarily important because some of the things that we that your users are already leveraging in their personal lives with their chatbot experiences with AI, they are not built in when you build these applications yourself. And so I think that was a really interesting point you made. And David and Devin, either one of you, what are the other kind of pitfalls that you see when organizations have started such projects, but maybe have not thought through all of the implications of refactoring their applications? I can go first. The biggest issue that I see, the biggest pitfall is organizations go into these refactoring efforts without the understanding of where they sit currently with tech debt. And one of the biggest anti patterns that we see, because look, like David mentioned in his intro, there are different ways to start on a modernization journey. Sometimes you replatform, then refactor. Sometimes, you know, there there's different ways to get started. But what you don't wanna do is replatform your tech debt. And and this is something that unfortunately, we see quite a lot and it can be very difficult to unwind. Because bringing your legacy tech debt to a modern platform, actually can leave you in a worse state than where you started. Because in your legacy platform, at least you knew where the bodies were buried. Trying to kind of jam your legacy application into a modern platform can create all sorts of issues. So I would say that's probably the biggest pitfall. Yeah, and I think like the other pitfall is actually the talent as well. Like is your talent pool ready to refactor? We see that, you know, the technology is there to refactor, but a lot of times when I'm on conversations with customers, it ultimately boils down to whether their engineers, you know, can think about or know about the modern architecture practices. Do they know about the new techniques in terms of coding or developing this application. And so I think there's also some internal kind of obstacles that companies need to jump through. And one of them is like making sure that your workforce is ready to adopt the refactoring project. I think that's actually pretty huge. And I did some research recently on modernization maturity, and we kind of looked at that exact kind of conversation. And it turns out that managing and maintaining legacy applications requires unique skill sets that sometimes deemphasize what is needed when you go to a modern platform. And, you know, the human factors cannot be overstated. It's understanding the best practices and patterns, but also there's a willingness and a readiness of team to get in that mindset of taking advantage of the new technology investment to improve applications and not just holding on to what was done in the past. Yeah. And so I would say cultural factors plus those skill sets you mentioned cannot be overstated. Great. Okay. So we'll take we'll make sure to emphasize cultural patterns, tech debt, making sure the organizations are ready, the expectation gap that, you know, people within an organization have because of their personal lives and what kinds of apps that they work with in their personal lives versus what they're creating for their customers, all going into the refactoring portion. Okay. So next, let's talk about the developer platform. So perhaps we can talk a bit about the value of Cloudflare's modern approach to refactoring and the advantages of a serverless platforms over traditional cloud solutions when it comes to certain types of apps at least. Perhaps, Devin, you would you would you would start here? Yeah. So, I like to think that everything in life when it comes to understanding hard problems comes down to looking at it from the right level of abstraction. And so when I think about platforms, what I think is I'm inheriting an opinion on what level of abstraction it's best to look at certain problems from. There are situations where it might be best to go all the way down into the voltages, if you need to optimize to that extent. But for the vast majority of use cases, especially customer customer facing use cases, I think that embracing the maximum level of abstraction that your use case allows, so that you can focus not on solving the problems that the platform provider is solving for you, but instead on that slice of value on the top that actually differentiates your business. You can look at industry leading applications, great success stories. 80% of what that application is, is probably non differentiated. Right? Let the platform take on as much complexity as possible, so you can focus on that 20% or if it's a 10% or 5%, whatever it might be. I think focusing taking advantage of the innovation of your platform provider, so that you can focus on what you do best is why I often lead customers towards embracing these kinds of platform level abstractions, especially when we're building cloud native applications or deploying applications that take advantage of edge infrastructure. Yeah, that's a great point. And, you know, see that actually today there are companies out there where they're like vibe coding platforms, right? Where anyone, my mom could go in and type like, hey, create a recipe generating application for me that has a certain style and look. What's interesting with those companies is that they have a hard infrastructure problem mainly when it comes to actually when you finish live coding and you wanna see the result of your vibe code, where are you gonna run that code? That untrusted AI generated code, right? And so what's been super interesting with Cloudflare is like we have a containers product that is an open beta. And when we launched that, we had a lot of vibe coding platforms come to us because what we were fundamentally solving was this infrastructure problem of deploying untrusted code, isolated, secure so that their development team did not need to think about that, right? They concentrate on making the Vibe coding experience better rather than thinking about infrastructure management. How do we make this serverless? We've got millions of customers across the globe, you know, how do we make their experience better? So, yeah. So anybody out there vibe coding, pay very close attention to what David just said, because I actually just wrote a report, Forrester's Guide to Some Quick Start Cards for People Who Want to Get Started. And one of the biggest takeaways that I had in that research is people can go from zero to a thousand lines of code faster than ever. And it works great on localhost. When they want to turn that application into something that has an IP address, that is where the trouble starts. Don't save that to the end. Think about that in the beginning. And in fact, you'll do yourself a huge favor if you do your vibe coding on a platform that considers an end game deployment. Because this is probably the worst problem to try to solve at the end. Yeah. And I'm curious, Devin, if you seen the news lately on the T app where it was fully vibe coded and the storage component that was vibe coded caused some some issues. Wondering if if you had any thoughts on that. Yeah, I did see the story and I think it's the first of many. I'll say upfront, I'm a proponent of vibe coding because I think that there's a tremendous potential. And as somebody who has been an advocate for powerful developer tools for a long time, I think that it opens up the total addressable market for developer tools to planet earth potentially. And I really like that. That said, we are we are just at the beginning of horror stories associated with production apps with real users, not considering things like security, not considering best practices for architecture. And this is an area where I actually think working with platforms that live in the cloud, that live at the edge, will do, tremendous favors for folks who maybe are not as sophisticated on those issues. Great. So, of the takeaways that I wanna make sure we emphasize from you all were making sure that you can maximize platforms that take away the complexity of the architecture on which your app is being built on, especially if it's cloud native, especially if it needs to be closest to users or is customer facing. We also talked about it's a very memorable way to put it, but, you know, can I consider the path of an app from local host to being having an IP address and, you know, use platforms that can provide the security and the networking, to make that make that easy? So this is this is very useful, information in that context. Okay. So the last part here and we'll try and spend the most amount of time on the on this part here is in terms of workloads. So we've talked a bit in in terms of platforms, and we've talked about what is refactoring. Let's try and apply that to specific workloads. And so perhaps, David, you can kick us off in this conversation as we talk about the first few around generative media, AI, and kind of the AI training data and storage of that as well. Yeah. So before we go into it, just wanted to show some some diagrams here just to set the context for everyone because what we see here in this particular diagram is the generative AI, especially generative media workloads that we're seeing on Cloudflare where people are using our infrastructure to run models that we host for image generation, text generation, etcetera. And what was what's key here is actually, even though we do generate this new type of media, you also have to think about how you're gonna deliver that media to your users. And so that's where the CDN part comes in. So it's both a serverless approach, but we're also leveraging, you know, Cloudflare's expansive network to deliver the content. And so that's the generative media aspect. And then going to the AI training data storage, just want to set a little more context before we hop in and get Devin's opinion on this is here you see we have a object storage product called R2. And what a lot of companies are doing when they are using R2 to store their AI training data is they're able to move this data freely from R2 to AI training providers, to cloud services, Cloudflare services, etcetera, etcetera, all without paying this egress fee. Right? And so those are like two very popular AI centered or centric workloads that we have on Cloudflare. I'm curious, Devin, you know, how do you see serverless inference and storage come into play when it comes to app modernization? Yeah, and so that's a great question. And I'll say a few years ago, serverless was really a major buzzword, it had huge developer communities. It started to dip a little. And now, I wrote a report on this last year, serverless is trending again hard. And I actually even looked at Google search analytics. And last year, serverless was searched for more than at any time since the term was first created. And so I think serverless on these modernization use cases, particularly as we look at some of these inferencing use cases. There's just something inherently serverless about these use cases. And I'm thinking about it even from the user experience point of view, not just the underlying back end technology. There's when you have these types of interactions or when you automate a piece of a workflow, with an agent, these are things that don't run all the time. So it's it's just a natural fit for serverless. I also think, and this comes from my customer conversations, there is a tremendous amount of talk right now about cost optimization with AI workloads. And I think one of the levers that organizations have access to that they can pull and tinker with is more on the type of architecture that you build for and run on. Because we, you know, when we when we think about, like, more like FinOps conversations, sometimes the architecture and design point of view is left out of that. And I actually think that when it comes to these emerging services, you're when you're thinking about token costs, things like that, optimizing around your architecture choices is really important. Also, just want to point out because you mentioned object storage. Another thing that I think people, some people are surprised by is that when you're building for some of these AI use cases, you're often still building around in need some of the technologies that you were building with before AI. And it should not be a surprise to anyone that object storage is probably more important than ever as we lean forward towards these use cases. Most of the reference architecture that I've seen from customers that involve either building agents or AI tools for their customers involve object storage somewhere in the stack and often involve serverless technology and serverless workloads somewhere in the stack. Yeah. Like going back to your comment about user experience on the serverless inference side, right? A lot of the conversations that we have with customers who are in the generative media generative AI space is user experiences is number one. And the reason why is it's backed by data where the companies tracked how long it took for an image to be generated and then delivered to the customer. And they realized that if it took too long, the customer would turn, right? You would just shut down the app. It's taking too long. I don't want to wait for it. And so a lot of these generative media companies really care about speed and delivering that AI generated image as close to the user as possible and also running that inference as close to the user as possible just so that they don't lose that customer. Yeah, a 100%. Yeah. And then object storage. Yeah, it's still there. And I think it'll be there for a long time. Right? It's just a matter of like, I think the serverless approach to the object storage. Again, bringing it close to wherever the downstream operations are, right? Yes. And this is an area that could be a major cost driver as folks start to wrangle with these new use cases. I would say your choice of object store, your data architecture and how you think about where you deploy things. All major factors that could that could influence your costs, especially if you unleash agents into these workflows. And agents start to do, I'm not saying direct reads and writes, not suggesting anything in particular, but agents start to get unleashed in a way that can influence data operations, costs, things like egress suddenly become major factors. Great. Okay. So, just to recap that part of the conversation, that was fascinating. User experience being critical and therefore serverless platforms again and then serverless platforms probably providing another tool in their toolbox for FinOps teams as they think about object storage, data architecture and deployment costs as we think about the AI and agentic AI world. Okay. So the next one, David, is really around this area of real time architecture. And so if you want to go through that and we can have a discussion there. Yeah, so here it's a durable objects architecture, but in fact, the way you abstract away from this, like Devin mentioned, right? Abstraction is this is how, let's say Google Doc would work or would be architected, right? So if Devin and I were working on the same Google Doc, you can see here you have like two clients going in and accessing like the app server, the SQLite database. It's all about maintaining state. So each action or change that Devin does, I see on my side and vice versa, right? So this is the real time interactive experiences that we see a lot of customers are building. So it's not just Google Docs, but it could be sports betting applications or just sports apps in general, right? Score is updated. Everyone needs to see that the score was updated. And then another real time workload is real time video and audio. Again, a lot of companies are looking to add some type of real time interaction, whether that's trying to build, you know, the next Twitch or streaming service, live streaming service. This is all done via Cloudflare's like out of the box development kits. And it's hosted again with that, all the little dots across the world in the three thirty cities, it really becomes important for real time applications because you want the least amount of delays, right? And so Devin, in terms of like real time interactive experiences, video, audio, I guess what type of challenges do you see typically present when you're trying to refactor and build these types of applications? I think the biggest challenge is that you'll never find a more savage customer base and we're all part of this customer base, than folks that are consuming real time media. So if you're watching a live event, for example, and you're getting a spinning wheel, you're not gonna be happy and it's going to be all over social media within seconds. Before you fix the problem, there'll be probably 10,000 people talking about it, right? It's one of those areas where trust is also extraordinarily fragile. I think that I would put the biggest biggest pitfall with these types of use cases is maintaining customer trust, avoiding that churn. Not every not every use case has the benefit of exclusive media rights. Sometimes customers have choices. And in those situations, latency is incredibly important, probably the most important. And it's not just in this use case, but I think that this use case is really interesting because it's one where you don't need any analytics. You don't need sophisticated dashboard to notice when there's a problem. And it's also one of those where the best experience is one where you don't even think about the platform you're using. You're just thinking about the content you're consuming. And in order to deliver that experience, a lot goes into it. It's I mean, speed of light, undefeated. And so sometimes the only way to solve this is by getting workloads and data closer to those consumers. Sometimes there's just there's nothing doing for the fact that it takes data time to percolate around the globe. So, yeah, I think that this is probably one of the best use cases when thinking about things like serverless and edge technology and how they can support these types of dynamic use cases. Yeah. And I think it goes back to what you previously mentioned where, again, you don't want your developers to be thinking about the infrastructure, especially on this type of workload or use case, right? We're also seeing a lot of traction here or an interest from customers, especially on the real time audio side, because they're trying to build agents, voice agents, whether that's your inbound sales development rep or inbound customer service rep that's powered by AI on the back end, but you still need this real time communication feel. I'm curious, like what you've seen in the market when it comes to like, you know, voice agents interacting with them in real time. Yeah, it's yet another example where even a relatively small degradation in customer experience will be noticed. There's also something else that I've seen that's interesting in multi agent applications. So if you have a situation where you've got an agent that's kind of the customer service rep, but there's all sorts of specialists in the background helping that agent do its job. Well, latency between those different hops, you have to think about this thing like any other interface. Right? Every interface in any system is a risk point. You show me a diagram of an application I've never seen before. I could probably write a 100 risks without even knowing anything about the application because all I'll do is point at the seams. And the same is probably true even more when you think about these agentic experiences, because now we have probabilistic systems in the mix. And these probabilistic systems require intervention and callbacks to make sure that we're not hallucinating any of the steps because it can blow up the whole workflow. Right? And so, we don't bad news doesn't get better with age. Bad news doesn't get better with latency either. So, want our intervention mechanisms and our robust architecture. We want those fail safes to kick in as quick as possible. And if some of the, orchestration is experiencing intense latency, it really is an uphill battle to enable these experiences with agents. Awesome. Okay. We've had an amazing chat. Thank you to both of you. Just to recap, bad news does not get a get better with age. Great one there, Devin. And also the idea that speed of light is undefeated, so get your data and serve your users closer to where they are. Great bits of advice there. So thank you again to both of you. We'll just wrap with a quick announcement of, Cloudflare's Connect, which is, is going bigger and better with our annual conference, setting the stage for transformative experiences that you wouldn't want to miss. It's taking place on October 2025 at the ARIA in Las Vegas. This will bring together customers, partners, and technology industry leaders for an awesome experience. Connect will be filled with dynamic keynotes and breakout sessions and technical tracks, certification, networking opportunities. So I hope that you can use this conference to empower you and your business while providing lots of learning opportunities. So see you there. Thank you again, David and Devin, and hopefully see you at Connect as well.