For our final episode of the year we chat with Netlify CEO Matt Biilmann about the way AI is reshaping software development and the trends he's excited about for 2024.
Biilmann says we can't ignore the impact GenAI is having on developer productivity. One of their engineers created a GPT that automatically generates stories for React + TypeScript components, and after seeing how successful it was internally, Netlify made it open source for the public.
We also chat over the results of their recent State of Web Development survey. The key takeaway is below:
The 80% of developers that have integrated AI into their workflow are quickly reaping the benefits. Seventy percent report using AI to automate manual and repetitive tasks and 42% are using it to improve internal knowledge sharing and increase productivity, freeing up more time for impactful work and enabling faster launch times. Over 50% of developers also realized new opportunities that AI created, such as generating new web projects with a single prompt or reading API documentation.
However, AI experimentation is not without its own unique challenges. Developers are concerned about receiving incorrect answers and information (65%), security issues and leaking confidential information (52%), a lack of regulation (48%), and a decrease in code quality (45%).
So much opportunity, but plenty of risk as well.
Last but not least, Biilmann tells us what he's looking forward to in the near future, specifically apps that can reformat their UI on the fly to be more customized to each user. He calls this UI 2.0, and it sounds a bit like what Google showed off in its recent Gemini demo.
Congrats to our lifeboat winner of the week, Petrus Theron, who answered the question: How can I make a public struct where all fields are public without repeating `pub` for every field?
[intro music plays]
Ben Popper Transform your business with Microsoft Azure. Easily move workloads with Azure Migrate and Modernize, and build new AI solutions with Azure Innovate. Access comprehensive resources, expert assistance, and cost optimization tools. Visit aka.ms/azureheroofferings to learn more. Make sure to use that link to let them know the podcast sent you and support the show.
BP Hello, everybody. Welcome back to the Stack Overflow Podcast, a place to talk all things software and technology. I am your host, Ben Popper, Director of Content here at Stack Overflow, joined as I often am by my cohost with the most, Ryan Donovan, Editor of our blog, the man who pushes send on the newsletter. Ryan, how’re you feeling today?
Ryan Donovan Oh, not bad. Not bad.
BP We are going to be chatting with Netlify CEO Matt Biilmann today. He's been on the show a bunch of times, makes a lot of sense. He is a central figure within a central company in the Jamstack community that's driving forward the state of web development. So they released a survey that came out earlier this week, and it's got a lot of stuff to say about what's going on with composable architecture, and then more pertinent to us here at Stack Overflow, what's going on with the adoption of AI tools and how that relates back to developer productivity as well as improving internal knowledge sharing which is kind of what Stack Overflow is all about. So without further ado, Matt, welcome back to the show.
Matt Biilmann Thanks. Thanks for having me.
BP So you do this survey every year. It used to be called the Jamstack Community Survey. Do you have a sense off the top of your head about how many people participated this year?
MB Yeah, this year around 7,000 developers participated in the survey.
BP And they're from all over the world?
MB Yeah.
BP Nice. So we'll start with the stuff that's near and dear to your heart and then we'll move on to the Stack Overflow stuff. Composable architecture unlocks new opportunities for enterprise. You can talk your book a little here. When we talk about tangible business results or cost savings or faster time to market and composable architecture, give me and Ryan some concrete examples, either clients you've worked with or just, “Here's how you would work with a client in X industry about how composable architecture is paying dividends at, let's call it, enterprise level.”
MB I think when you think about how we approach it and what it has always meant for us, there's this notion in the whole industry always around should you buy something or should you build something? And when you think about companies thinking about how they should build their digital experience for their customers and web properties across corporate sites and e-commerce and the dashboards and so on, they've also always had that kind of question of, “Hey, should we buy some platform that then determines how we deliver our customer experience, or should we build the whole thing ourself to really be able to craft how we represent ourselves to our customers?” And I think there was a long time where, especially in the large enterprise, but also at smaller scale, companies were really going for buying some monolithic piece of software that bundled together the user experience and the business logic and the data access and content and everything. And now we're seeing a clear divergence from that to saying that we need to be able to actually build the user experience that our customers interact with, because otherwise we're going to essentially look and feel and behave like any other company that bought the same piece of software that we bought. So a lot of this composable is really just about how we go from that big box with our whole user experience bundled into it, to being able to say, “Hey, I want to be able to build a great experience for our customers across different surface areas and devices and so on, and I want to be able to have the best-in-class content management and the best-in-class e-commerce behind it and the best-in-class search behind it,” and so on and so on.
BP Yeah, I like that idea of standing out. That seems like something that makes a lot of sense both on the web and especially mobile web where it's so intimate– you and your phone in bed thinking about what you've been shopping for.
MB And when we think about a lot of customers, think about banks. Maybe 5-10 years ago, you would fundamentally pick your bank by your experience in the local bank department and how were the employees there and how did it feel like. They would put a lot of work into those offices and make them feel a certain way and so on. Today, it's probably really rare you go to one of those offices, and it's probably not the moment where you pick your bank, it's probably a moment where there's something wrong or something. And almost all your interactions will be through the web or through your phone. It'll be a digital experience.
BP No, I notice that a lot. I think that's a great point. I feel like my bank and my retirement and my other financials, they're some of the apps that update the most. I'm getting release notes all the time. They're making tweaks, they’re making improvements, they've clearly hired top-notch developer teams to focus on it. That's what it is.
MB Because it's going to be the driver for what you see as their brand and who is your bank. And that's kind of the same for almost any business now and that's why it becomes so important to control your own destiny there and be able to execute there. And then when we look at this, we really look at how do we enable people to do it fast?
BP Yeah. Now that I think about it, I care a lot about the app for my insurance provider– is it hard to use or easy to use? And now my car– dealing with things for my car, I care a lot about the app.
RD So I wanted to touch on the sort of headline you threw at us when you contacted us, which was ‘the end of the monolith architecture, the death of the monolith.’ Why do you think monolith architectures have survived this long? We've had service-oriented architectures and composable stuff for a while.
MB I think they've survived long because it's true that if you don't have something to glue together the different services in a predictable manner, then it's suddenly up to your team to build and maintain all the connection points and all the complexities that come from having different services. So I think that's obviously a big part where we think we can make a big difference. It's a part where I already think we have the model. When we talk about Jamstack, that was kind of a separate idea of starting to take the web UI and decouple it from content, from back end, from services, and make it its own thing. And right in the beginning when we started Netlify, there were people doing that and doing it really successfully, but it was pretty rare because there wasn't any sort of platforms around it at the time. If you wanted to move from just building a monolithic website to building a self-standing UI, either statically built or a single page application or anything, you suddenly also had to just implement all the stuff you took for granted in the monolith around how do we actually pick up a change and redeploy, and how do you manage release management, and how do you orchestrate the whole thing and so on. And once we kind of built the first front end cloud product to say, “Here you just connect your Git repository and now that whole piece works,” that kind of unlocked the migration there. That kind of unlocked this whole new ecosystem of modern front end frameworks and just in general treating the web experience as its own thing. And I think with composable and monoliths it's very similar. If you have to go from having bought a traditional DXP engine or the full Salesforce Commerce Cloud and so on and you're like, “Okay, we can see that our developers are really struggling with building within these constraints. It's really slow and so on. It now no longer feels like we can really deliver the right experiences,” the challenge –and we see this in the survey as well– the challenge is really how do we actually go from that to being able to introduce new modern content and data sources, decouple that front end, be able to still allow our non-developers to interact with it and understand where it lives and so on. And if there's no platform-level answer to that and it's all just a question of, “Good luck, buddy. Figure it out, glue it all together, build it from scratch, maintain it, operate it,” then obviously you’re going to end up with a monolith and say, “Okay, but we have the whole thing in a box.” It's still appealing even if the end result might not be exactly where you want it to end up. So we have to reduce the friction if we want to unlock that.
BP Right.
RD It's almost like it's a middle ground between just assembling everything yourself and having the monolith, and I think you'll see a lot more connector stuff developing.
BP And I think another thing you noted in the survey is that 44% of developers said they would have built with composable architecture if they had a clear path. So almost half can't because they lack the resources or the technical talent or they can't get the approval. So getting off the monolith is not easy. We at Stack Overflow can attest to this. So I want to shift gears for a second and do the second half of this episode about what's going on with AI. You had a whole section– “AI success requires modern architecture.” 80% of developers say they have integrated AI into their workflow and are quickly reaping benefits. They use it to automate manual repetitive tasks, improve internal knowledge sharing, increase productivity, freeing up more time for impactful work. 50% say they are realizing new opportunities such as generating new web projects with a simple prompt or reading API documentation. So let's start with the new opportunities bit. Matt, if you have had experience with this or a client has talked to you about this, tell us about a little bit of how this works in real life.
MB I always look at it as the short term and the long term. Right now we are all seeing a lot of potential in generative AI helping us build the same kind of things that we've been building for the last couple of years with the same overall set of tools that we've been using for the last years, but faster. I think that's a lot of the initial, where we see a lot of just obvious opportunities to automate all the manual work and the boilerplate work, and what are the things that we're spending time on that if we could just automate it away we would all totally do it. And we see from the survey that's the first area that developers are thinking about. 74% think that AI can improve what they're doing there from where they're sitting. So it seems like the obvious low-hanging fruit. Of course, all this stuff that's boring and repetitive, let's get a machine to do it.
BP I have a developer here at my co-working space, Ben from Dev.to, and he uses it to generate tests. He's like, “It's great. Generates good tests. I glance over it once. I don't like writing tests. Very helpful. Improves my speed. Simple.”
MB Totally. We shipped, for example, an open source project to do that for your Storybook components. We use Storybook for our whole component library, and it's amazing when you have it, but it's extra work. You might need a new component. Now you've got to go make the Storybook for it with all the different examples and so on. So our team open sourced a project that lets you just put in the component you wrote and then it'll give you the Storybook story with all the components. Those are just obvious examples of stuff that's sort of very low-hanging fruit. It's right there, it'll totally accelerate how we work, and it's great. That's the short-term thing.
RD I think when we did our last survey and asked about AI, a lot of people were using it and finding productivity gains, but a lot of people were sort of wary and didn't quite trust it. Did you find any results around that?
MB We hear a lot that there is a fear of actually adopting that. At the same time, from the audience in the survey, I would say we really saw that the majority have tried using some AI tooling one way or the other. In that way, it's really exploded onto the scenes and people are trying to figure it out and they're experimenting with it, but we have seen a general also side effect of, “Hey, can we trust the output of these models to not infringe on copyrights and not get us in trouble?” That part of it shines through for sure.
BP I've got the survey here. It says 65% were concerned about incorrect answers and information, 52% had concern with security issues and leaking confidential data, 48% were concerned about the sort of lack of regulation, to your point of what legal repercussions might there be, and then 45% were concerned that this might lead to a decrease in code quality that you wouldn't notice right away but would become tech debt down the line. So definitely more than half, or around half, are thinking about this stuff. We talked about OverflowAI and what we're excited about with that and how it lines up with Teams. And with Gen AI, really one of the most obvious applications I think is incredible enterprise search. Put all of your data and all of your docs and all of your code repos in here. Let this foundational model learn this stuff, maybe fine-tune it on your data so it's a little better with that stuff, or maybe use retrieval augmented generation, but now you're going to have this unbelievable assistant. And so your survey kind of found that there's the automations like you said, Matt, which are the low-hanging fruit, but 42% are also using it to improve internal knowledge sharing. So I thought that was really interesting because we've sort of been looking at that same space.
MB I think that's really interesting and another short term aspect of, “Hey, this can really be useful to date for us.” We'll see it both for internal knowledge sharing, I think we'll also see more companies do similar things to how, like at Netlify, we introduced what we call ‘Ask Netlify’ that's both integrated into our docs and into our Slack app and so on. That's again, a model that has access to all of our docs, but also all of our forum posts and material from our support team, sort of an internal knowledge base.
BP The ultimate tech support, yeah.
MB And it's just asking it a question and it'll go to that and it's trained to answer those. That's another area where I think, again, there's this low-hanging fruit that we'll quickly see becoming really useful. Then I think there’s the other area that’s like, “Everything is going to completely change from this over the next five years.”
BP I think it's interesting because people were saying in that later section, “Well, what if the information is incorrect? What if I don't know how it's licensed?” That's one of the benefits of searching on Stack Overflow where you know how the code is licensed, or internally, at least having sort of a crowdsourcing system that's saying, “This is the best answer for this question and this information is up to date.” That gives you a little bit more, if the AI system can cite its sources which people are learning how to train them to do, that gives you a little bit of ground truth to work with.
MB Totally agree. As you said in the survey, everybody has that view of, “Okay, but sometimes these models come up with stuff that looks very real and it looks like they're very certain of but that's completely wrong. How do I differentiate between the answers it gives that are completely right and the ones that it gives with equal confidence?”
BP Right, it doesn't give a confidence score. It should give a confidence score. All right, Matt, let's have fun with our last five minutes here. What are you dreaming about? You're saying there are things coming that will totally change everything. You're saying we don't know what they are yet, but what are you dreaming about and what are the things you would be most excited to bring to the table? Or if you're able to give us a generalized view of where your roadmap is going, talk to us about what gets you super excited in this space.
MB Real big picture, I think we'll see what I call UI 2.0. I think every single user interface we’ve built since the very first computers was based on the core constraint that a computer requires precise instructions and then it will do something predictable based on that. But all of our user interfaces are fundamentally built around that kind of interaction with a computer, and I think widely that constraint just started completely going away. So if we really rethink how we should build user interfaces from first principles when that's no longer the constraint, how are they actually going to behave? Because the endgame there is also not a chatbot sitting and typing in this and getting text back. You still want to interact with visual information that you can sift through and that you can interact with. You still want visual user interfaces.
BP I want to interject something here. We actually just did an episode with a cool startup called CommandBar, and the idea there is that if you see a user incorrectly using something or searching for something over and over again, you give them a pop-up –but it's not a pop-up because people hate those– but you guide them to the right place in the UI so they don't make that mistake. And I think what you're talking about is really this amazing idea of the platonic ideal of Clippy. I'm in the software, and if at some point I don't understand it, it's going to guide me to the right button. And if at some point it sees I'm clicking three buttons when I could be clicking one, it's going to say, “Hey, I'm going to walk you through this now.”
MB I think we'll see it go much further than that. I think we're seeing the early parts of helping software developers build UIs, but I think we'll also get to a point where you can kind of construct your app with sort of a whole toolkit for what kind of UI should the model be able to construct on the fly that you can interact with in a way where rather than having this strictly upfront defined UI, it’s you digging into a problem and working with this model and solving it and it'll give you the right kind of individualizations that you can interact with and dig into, it'll give you the right set of actions that you can reach out to and tweak and so on. I think there's a lot to be said to how on-the-fly they will be generated.
BP It's interesting you say that. I got the same thing from the guy at CommandBar. He was sort of saying that UIs will now be personalized to maybe the situation, like you said, or maybe the user. And then we got a hint of this when we watched –and it's not been publicly released– the Gemini demo. You ask it a question and it creates a UI that's appropriate to then working on that project, and if the project changes, it refactors the UI on-the-fly. So if that is something that is not just a demo, but really works, UI 2.0, I like it.
MB Yeah, totally.
RD So I was wondering if you all Netlify engineers and such use Gen AI for your daily work.
MB We do. I try to encourage our engineering team to adopt it as much as possible. We obviously are Copilot customers and make that available to all of our team. But actually, I still see a lot of engineers even at Netlify that are wary about adopting it. I do see a wariness where I still feel I have to even drive encouragement into ‘I want you all to adopt it.’ Because, one, I think there's a certain fear of, ‘Will these tools replace us as engineers?’ that I just don't think is going to be true. I think the other thing is going to be true that, as an engineer, your craft will to a large degree be defined by how good you are at interacting with these tools and using them for things. So I'm really keen on having engineers adopt them as early as possible and learn how to wield them to do things with, because I think more and more of our jobs as engineers over time will be to wield these things, understand their strengths and limitations, and get them to be effective on our behalf. And I think as that happens, we'll see the number of developers grow rather than shrink.
BP Yes, even us marketing fools will be developers on the side. I'll be developing a few little things. I made a dog park app in collaboration with ChatGPT, so that was fun.
[music plays]
BP All right, everybody. It is that time of the show. Let's shout out a user who came on Stack Overflow, helped to share a little bit of knowledge, and saved a question from the dustbin of history. Awarded two days ago to Petrus Theron, who provided an answer and was awarded a Lifeboat Badge for, “How can I make a public struct where all fields are public without repeating `pub` for every field?” If you want to stop repeating pubs so much, we have an answer for you and helped around 20,000 people, so we appreciate it. As always, I am Ben Popper, Director of Content here at Stack Overflow. Find me on X @BenPopper. Shoot us an email with questions or suggestions. Or if you want to come on the podcast, that's cool too– it's podcast@stackoverflow.com. And if you enjoyed this show and you’re game to move forward to UI 2.0, leave us a review, because it helps the program.
RD I'm Ryan Donovan. I edit the blog here at Stack Overflow. You can find it at stackoverflow.blog. And if you want to reach out to me, you can find me on Twitter @RThorDonovan.
MB I'm Matt Biilmann, CEO and co-founder of Netlify. You can find me on Twitter or Threads or Mastodon or wherever the platform of choice is for you, if you can figure out how to spell ‘Biilmann’.
BP I'm going to put that in the show notes. All right, everybody. Thanks for listening, and we will talk to you soon.
[outro music plays]