How AI is Changing App Development with Guillermo Rauch, CEO of Vercel, Part 1

All right. I am here with Guillermo Rauch, CEO of Vercel, the makers of Next.js. So great to have you today. How are you doing?

Thanks, Eric. Really excited to be here.

Thank you so much. So obviously, you know, your company makes one of the most widely used and well-known frameworks in the internet. So we've got a lot of fun stuff to talk about today, but I, I'd love to hear about your journey and what led you to ultimately lead this company.

Yeah, for sure. Just yesterday I opened X/Twitter and I see Argentina trending. I was like, oh, what do we do again? So there was a big topic about Argentina getting a new president and it had me thinking a lot about my past and my journey. It was a good reminder. I'm in San Francisco, California now. But I originally grew up in Argentina. I taught myself how to code from a very young age. I got into the world of open source. I kind of built up a career from scratch. I designed it myself. I taught myself JavaScript, which is sort of the, I call it the English language of modern programming because you learn that language and you can do a variety of things. You can launch products, you can launch servers, you can do backend, frontend. IoT, you can program the edge of the network. It's the gift that keeps on giving. So I became sort of an expert in that domain that led to getting job offers internationally, even when I was like 17, 18 years old. I found my way to the Bay Area where I discovered two things. One, that my true passion was to start companies. And two, later on, and which has become now my true calling in Vercel, is that you can, you can build companies that sell infrastructure and developer tools. I've always liked increasing the productivity of the, even when I was developing by myself, I would build tools to make myself more productive. So it's been a blessing that I was able to transform that idea of building developer infrastructure into like what's now become a very fast growing business in Vercel.

Wow. Yeah. Yeah. I love it. I love what you're talking about around sort of JavaScript being that, uh, that universal language. I was actually talking with a CTO last week and recalling, I'm a little bit long in the tooth now, but recalling that, you know, I've learned over a dozen programming languages in my career. I mean, starting with BASIC and Pascal and C++ and Java and Perl and Python and C and so on. And it was because I needed a different language for all different kinds of problems. And then, you know, later on, kind of JavaScript sort of ate the world and now you can essentially do everything. Now, not to say that it's The best at everything, but obviously Swiss Army knife, it really gets the job done.

I compare it to English. Like it's a weird language in a bunch of ways, right? Like it has all these quirks. In fact, for non-native speakers like myself, like we struggle where like we intuit how you would pronounce a certain word. Oh no, there's an exception in that one. That one is actually like rendezvous, whatever. You know, like there's all this DNA from multiple languages, you know, Germanic and French words. And so JavaScript is a lot like that. It has all these weird quirks with like types and, but it proliferated and proliferated and proliferated. The creator calls it almost like an evolution rather than intelligent design. Like it's just been, it had a very fast sort of genesis in like a 2-week sprint at Netscape where they conceived it and they just evolved and evolved and evolved. And it's because it's been in the right place at the right time. I think that like is like the success in business sometimes. You have to be at the right place at the right time, riding the right wave. And JavaScript rode the wave of being deeply embedded in the web browser. And that's kind of how I approached it. I was looking at all these incredible apps that were launching inside the web browser and transforming the world. I noticed when I was a kid, how Google was adding all these like really cool features, like an autocompleter in real time when you would search. And then they launched Gmail. Gmail should have been a desktop app in theory, but they were able to put inside the web browser. And what all these innovations had in common was that they leveraged JavaScript to create this really rich, interactive, and real-time experiences. So they started sort of reverse engineering how everything worked, because that's the other beauty of being inside the web browser is that you can peel back the covers and understand how things work. You can like right-click, inspect, and now you know how everything works. And I started like noticing how, okay, like Facebook works this way, Google works this way, Gmail works this way. And I kind of became obsessed with this idea of, I understand how these people are doing it. Like, what if I could give the gift of all these platforms that they've built, all these technologies they've built to everybody else? And that's what's kind of made Vercel, I think, so interesting because there's maybe a handful of companies that have figured out how to scale on the internet. And those are Amazon, they're like FAANG or whatever, like now it's like MANGA, I guess, Meta and Amazon and Netflix. But outside of that, you know, the rest got the scraps really. And this is why you have the cloud as sort of this monumental advancement in computing because Amazon, who figured out how to run amazon.com so well, said, okay, people, we're going to share some of our infrastructure with you. But what they didn't share is all the know-how, all the insights, all the tools, all the frameworks of how to like, it's almost like they're just giving you the bare metal. You know what I mean? That's how you can think of AWS. And when you think about Vercel, we kind of built everything on top so that you can actually take advantage of this incredible step forward of democratizing access to all the incredible infrastructure. So for me, the story has been JavaScript became sort of the lingua franca of how you program nearly everything. And to your point, it's not perfect., and now you can kind of build a cloud that's designed for those people that are looking for productivity. They're looking to ship innovative products and they don't want to think about all the lower layers. They don't want to think about Assembly and C++ and Rust, all the things that sort of we use under the hood.

Yeah, no, that's great. So, so I'm actually a backend engineer and I, I got to say probably, you know, at least half of my job. Used to be supporting the frontend engineers, right? They need an environment stood up. They need me to stub out an API. They need some dummy data. Like most of, we would definitely have to do a, so usually what we would do is we'd kind of have a sprint that's like, let's dial in the information architecture. Let's kind of figure out the workflow of what we're trying to build. And then while frontend is working with design to figure out what the UI is going to look like, I'm over here trying to spin up a backend so that they have something to point to and connect to so they can actually stand this thing up and then, okay, here's the code and let's go deploy it and all that. And so, yeah, I would love to be able to just hand them off and not have to deal with it.

Absolutely. Yeah, you nailed it. That's the work. And imagine doing that for every project, every new idea that you have. You have to spin up these teams to do all the sprints and do all this repetitive work to stand up all that infrastructure so that you can get started deploying and iterating and think about it now in the context of this incredible fast progress that the industry is making. So now you would be spinning up a new environment so that someone can experiment with LLMs and AI. And if you take too long to do that, the competitor gets the AI feature out faster. So what we realized in the process of creating this company is it's not even just about making a thing fast like that. That's almost like we started. I noticed that most websites were just too slow or they would very trivially go down. Like think Black Friday, which is coming up now. Everyone at Vercel is like monitoring the situation closely because it's, we're expecting this massive traffic spikes. But the reality is that it's not just about being able to run a thing well. In a fixed point in time. Like, okay, I can run ericwise.com really, really fast. It's about giving you the platform and infrastructure so that you can iterate on it very fast with the right direction. And that's what kind of has become the true hidden superpower, hidden secret of the Vercel platform.

Nice. Yeah, I mean, so at, you know, at the end of the day, it's We want our developers to be focusing their time innovating and creating and not all the time on all the overhead and all the grunt work that it takes to take that little bit of innovation and actually get it out in the world.

Totally.

Yeah. Amazing. So, um, what, uh, you, you mentioned AI and obviously everything in the world is, is AI driven these days. Where do you see the opportunities or the, the changes coming as it pertains to developing?

Applications? Yeah, I think just this weekend, literally 2 days ago, I think I arrived to a term that I've been kind of like refining and approaching and been looking for. I've been calling it AI-native UX. Hmm. So AI-native UX summarizes for me, so AI-native user experience summarizes for me the new North Star that I want to unlock. For our customers, and I think conversely also the industry should be focusing on. An AI-native user experience is a product or application that you're creating that is designed from the ground up to emphasize the benefits of AI. So looking back to the platform shift that we saw from the introduction of mobile, Right? Like the iPhone comes out. The very first few apps and even the demo that Steve Jobs gave of the iPhone, it seemed like they were retrofitting the stuff that collectively we all knew about. If you actually look at the first screenshots of the iPhone, they show Safari, the web browser, rendering Wikipedia in the desktop version. And Steve Jobs in the demo shows you how cool it is that he can zoom in and out.

Right, right.

So there's an element of like mobile native or iPhone native, which is like, oh, you can pinch. That's awesome. There's a new technology, new capability that most of us didn't have. But what he, the application layer is retrofitted even from Apple who are the pioneers and they're introducing this product, right? I think what's happening with AI right now is exactly that. We have this groundbreaking new capabilities, the pinch to zoom or multi-touch or GPS. All of those innovations are the LLMs, the retrieval methods for long-term memory, all of the diffusion models like DALL-E and Midjourney, like all of these tools that we can now use are enabling AI applications. But what most people have gotten to market so far is the retrofitted versions. Oh, I already have a UI. I already have a text editor over here. How can I shove some AI on it? Yeah. And that, I don't want to sound dismissive there. It is useful to add AI to existing UI paradigms. But just like what happened with the iPhone is that one day we got Uber. Uber was an application that was native to the platform. It could only be conceived with portable computing, like mobile computing. It could only be conceived with the GPS built in, and it enabled a magical experience. So I think what's going to happen over the next, literally at the pace that AI is moving, weeks and months and years, is that we're going to start seeing the Ubers of AI arrive. And our job at Vercel is to give them that platform where everything is already on a silver platter, ready to be deployed. You, what you bring in is your new ideas about how to disrupt existing application models with the use of AI.

That's really interesting. I haven't thought about it that way. What, what, what's a, is there a clear example that comes to mind for you or something that you've even seen that is AI native?

Yeah, for sure. So I'll give you first an example of like not AI native. So you have a text editor, let's call it VS Code or some other example, let's call it like Super Code and you add Ghost Text. So you add the autocompletion of AI. That's not AI native because we already were used to autocompletion in some form of another. Let's call it like old school autocompletion, right? From a dictionary, from a HashMap. From maybe some a little bit more advanced smaller models, but it's definitely been disruptive and incredible kudos to the Code team in Microsoft for bringing in a superpower of a large language model that almost seems like as you're typing the code, it's reading your mind. But again, the primary interface there was not conceived for AI. We just tossed some AI on it.

Right, right.

Conversely, I think ChatGPT is truly an AI-native application because the idea of a chat interface with an AI that solves this huge range of tasks that used to be specialized in all kinds of little apps. This used to be in Wikipedia. This used to be in some like, uh, little Google-based calculator or widget of some form. But it's still the critique that I would make there is that it's still the raw model being exposed into the internet. It's almost like you only took the AI capability and you tossed it into a web browser. It's successful due to its incredible minimalism, but it's still almost like a starting point.

Yeah. Well, it's like Google, right? Just the little search bar and that's it.

Yep. Which is awesome. And I think there will be a place for that, but I also think If you look at maybe a more recent example that I just saw in the last couple days is there is a new kind of way of rendering 3D graphics that is built on basically you don't know how to model, but you give it an approximation, you give it kind of like a rough shape. And then on the right-hand side of the screen, it gives you this incredible 3D rendering. Yeah. So this is so disruptive because the classic 3D rendering application was full of buttons and UIs and like complexity and the learning curve was months and months of like getting up to speed with what all these little buttons do and the kinds of like layering systems and categorizations and so on. So that is truly disruptive because the incumbent looks at it. And they don't even think it's competitive. They're like, well, that's just some other thing. You know what I mean? Those are the real threats that are coming, I think, to large established businesses that initially they're going to look at it as like, well, that's a toy and that's a toy and that's a toy because it doesn't do like even 1% of the things we do. We have all these features, all this functionality we've been accumulating for decades. But that's exactly why these applications are going to be so disruptive because of how little they do in comparison.

So I get it now. I see what you're saying. So I see this all over the place now where the incumbents, they've got all of their buttons and then they add another button with the little—

It's the AI button.

It's the little magic fairy dust button that's like AI, you know, that's just yet another feature on top of that stack. But you're saying something that's completely built from the ground up. Yeah, I get it.

Yeah.

So another area where, so I do a lot of work, I do a lot of work with, with product design and, and user experience as well. And where we've looked at good opportunities to integrate AI is if we map out our user experience, map out our workflow, any point where there is a decision that has many different branching pathways, that has all this different logic that has to go into it, that's a good place to at least introduce something like a chatbot. Versus say, hey, we've got all these screens and then you've got to make some decision or go through some workflow and then you can go any number of paths down. Well, where there's fuzziness in that, you can have a chatbot come in and have a dialogue. Okay, based off of this conversation, I think you should go that way and then put you back on a workflow.

Yeah, I think there's going to be a hybrid opportunity for AI where like you're not redesigning the product you already have or the workflow you already have, but you're also not sort of spinning off a completely different way of solving that problem because you want to make it feel natural to the user. So I do think we're going to see a lot of this Sidekick approach where you basically give the opportunity to the, to the user to navigate that complexity, that branching through a chat. Type interface. I think it's probably the next evolution of what became the command palette. Command K brings in all these options, but imagine if you could create a truly AI-native version of that. And there's already examples of folks doing this for documentation where you go to a documentation website and now you can ask with AI. But I think it's also a very good tool for navigating the UI itself. Cause think about why documentation exists in the first place, right? Or even like the documentation of a lot of GUI applications, it involves, you go to the docs and they show you little screenshots of the UI. So how do I change my password? Oh, first click here and it shows a screenshot of the options button and then go here and then go here. What I think is really interesting and certainly a pattern that we're enabling our customers to do at Vercel. We open source an example called chat.vercel.ai, is you can take that open source frontend of a chat type assistant, but now you can bring it into your own application. And it's not just about asking questions of your product, it's also about, say, well, help me change my password. And then you help them change their password. You're not even teaching them, you're not pointing them to documentation. They just do the thing. So I think that'll send us down this path of first AIs are answering questions, but then they're going to facilitate agent-type behavior where you're still in the loop. So they might give you, okay, click here and we bring this UI for you to change your password. I think in the future, they'll go off and perform the action altogether for you. They'll say, well, confirm that you want to change your password. I already changed it. And now press here, save it to your keychain. So basically an evolution in how much we're automating and how less we're expecting of the user to become an expert in that user interface. Because when users become experts, they know exactly how to navigate everything. Oh, the, I go to my project over here and they drill down over here and they'll drill down over here and like. I ended like 7 levels nested into a breadcrumb to find another button to press. But I think it's just going to be a lot more agile to have the AI do that altogether.

Interesting. Okay. So yeah. So the current paradigm in UI development is this sort of static workflow, the static information architecture. And what you're saying now is that the AI will enable you to be more flexible or even take the UI out of it entirely. And have the AI basically just run the backend code directly. Is that right?

Correct.

Yes.

And I think in the short term, people will introduce a hybrid version of this because it's not like, you know, Salesforce is going to say, you know what, we're going to delete all of our UI. People hate it anyways. You know, it's kind of clunky. We're tired of changing the colors every month, every year to modernize. They were just tired of changing the fonts. They're going to say like, Okay, there's an alternative way to navigate all this complexity, to get information and to perform actions. Until I think the AI pathway, as it becomes more intelligent, more reliable, does more and more and more and more. But again, going back to AI-native user experiences, like it does raise the question of like, if you're starting a new application today, do you actually build Salesforce? Probably not. You're going to build a more AI-driven version of it. And by the way, there will be UI. I don't think it's all like text and text and text and text and text and text. And that's why, again, there's a huge opportunity here for folks to come with fresh eyes and to be building this new patterns that will— we'll probably going to look back on them as like, of course, that was the pattern to use.