Acquired Minisodes: A conversation with Guillermo Rauch

Guillermo, welcome to the Significance Summit.

Thanks for having me. Excited to be here.

For folks who are watching this video who are not familiar with you and your company, what is your full name and what do you do?

I'm Guillermo Rauch, CEO and founder of Vercel. We build the frontend cloud which helps developers publish the best experiences on the web.

What does that mean?

So you build an application or a website, and we make it really, really easy, and then you can deploy to our infrastructure. And what really stands out about Vercel-made applications is that they're really fast, really dynamic, and very sophisticated. So this is tooling comparable to, you know, what folks would have internally at Facebook and Meta to build applications. So we share the sentiment with Statsig that it's worth democratizing the tools that the biggest companies in the world would have to build their own products. But what about the rest of us? So Vercel tries to bridge that gap.

And you have some amazing companies and sites that are building on Vercel now. Can you give us a sample of a few sites people probably go to often?

Yeah, companies like Under Armour, Unity, Nintendo, OpenAI.com, companies that are again at the frontier of technology, pushing, growing, and They need these sophisticated tools and infrastructure, but they want to make it really, really easy for their developers to make progress.

So your story is pretty interesting. Before we kind of talk about the future of product development, how did Vercel come to be?

So I had the problem of myself wanting to publish a website when I started my first— actually, sorry, my second startup. And I had sold my previous startup to a company called WordPress. And when I left, I wanted to publish a website for my startup that used the latest and greatest technologies. They were open source, like Google and Meta had open source all this great tech. One was called React, the other one was called Kubernetes. And I sat down and it took me weeks to just set up all the pieces, like put everything in place to start working on that first page. And every big idea on the internet starts with a web page. So I was like, How can it take so long? And this is again after like 15 years of experience as a frontend engineer. So I kind of knew what I was doing, but it was so hard that I kind of got sidetracked. My original idea, I put it in a drawer.

What was the original idea?

The original idea was kind of related to like helping developers, but it was more around like real-time communication. And I realized I can't go into that depth of like real-time and whatever, when the basics of using the cloud efficiently were not in place. And I realized, look, it's hard for someone that's building a startup to go from 0 to 1 on the web. Imagine the biggest companies in the world. Imagine the famous car companies, the famous retailers. And so I did two things. One is I created an open-source project. Called Next.js. It makes it really, really easy for developers to get started. Today, Next.js powers some of the best products on the planet. And so I made it really, really easy, but also the scale and sophistication that you could get to with that project is pretty significant. So like walmart.com runs on Next.js and my side project and every little idea or the AI startups that are being born every day are using Next.js to power their businesses. But then I realized, look, it's not just about the tool, what about sharing that creation with the rest of the world? I was very inspired, and this kind of goes back to the real-time idea, I was very inspired by what Google Docs had done for word processing. We used to just email files with attachments, and then they came to the scene and they were like, look, you can go to a website and every document you build, every document you share, every document you edit,, it becomes a hyperlink that you can share with anybody on the planet. And I wanted to do that for applications. And so I had Next.js, and then I created this basically real-time collaboration system on top that makes it really, really easy to send the application that you're working on, not a mockup, not a design, but the actual application that you're working on across the world. And so the other day I met with a CEO of a really large retailer, and they were like, Wait a second, that means that I can be in the loop and I can keep up with the product changes that are happening to my doc file.

Sounds dangerous.

Sounds dangerous. And so the CEOs love us, right? You can actually be involved in the product development process. And again, too much of that could be dangerous, but think about the idea of folks that don't really understand what's going out the door. Sometimes you think about objects in the physical world and manufacturing companies, and there's a— You feel the product that you're building. You feel the thing that is going to be delivered and shipped to customers. When you're building a web application, I think before Vercel, I think folks didn't have that direct connection to the artifact that you're iterating on. And that's why, again, folks like Statsig were like the perfect peanut butter and jelly, because we're helping companies experiment a lot more. Vercel is actually letting you play with that application, with that idea, with that content change, before it goes out the door. So it de-risks, it makes developers more productive, it makes companies more product-centric and customer-centric, and yeah, so it's been fun to sort of build this tool and platform for the world.

Well, you mentioned the integration with Statsig. In what way do you work together?

So what we're both conjunctly trying to bring to the world is the idea of minimizing the cost of an experiment. There's almost like a paradox with bigger companies when you serve enterprises, right? Like, they've already figured out their business model. They've already— like, you guys have covered all these amazing companies, you know, like Ferrari and Porsche and Mercedes. They already have like kind of a great product. So now they're trying to figure out their .com, and there's actually significant risk with changing their .com. One of the big companies that we serve told me, look, before Vercel, we only used to ship a couple times a quarter. And I live in San Francisco, I'm in Silicon Valley, I know folks that use our platform to ship dozens of times a day, hundreds of times a day, and I'm like, wait a second, you ship only a few times a quarter? And they were like, no, but wait, it gets better. I actually ship only 3 quarters of the year because when Black Friday—

Oh, you don't want to jeopardize holiday.

Black Friday comes around, we do a big code freeze. And so, there's a lot of fear sometimes to changing your software. And so what is the recipe for eliminating that? It's not to slap the fear out of the customers, to give them the tools and workflows so they can iterate on those changes and progressively roll them out to the world. So the Vercel rendering infrastructure makes it very easy to simultaneously basically hold hundreds, thousands of permutations. And now with AI, infinity of permutations. That can be served to the customer when they access your .com. And the important part is to track the progress of those experiments. What you want ideally is almost like a knob, right? Like you say, oh, that idea that we publish on the home page is selling like hotcakes. Ramp it up. Let it rip. Oh wait, orders are going down. And potentially through integrations, we can do this automatically. Dial it back. Get the feedback back to the product teams. So that feedback loop, we call this iteration velocity. It's not just about moving really fast. It's like the infamous move fast and break things.

It's revised to move fast with stable infrastructure.

Enter Vercel, right? We are that stable infrastructure. And so what happens in practice is instead of moving fast in whatever direction, which is I think what a lot of companies are used to doing, Realistically, a lot of companies that have not figured out their dot-com might hire an agency. The agency goes to town, bills infinity hours, and there's a sunk cost fallacy. We're gonna ship that new dot-com no matter what, even if it performs, doesn't perform, et cetera. So the better way to do it, and this is why I make this distinction between velocity and speed, is velocity has a clear trajectory, a direction. You have to monitor every step of the way that you're getting to the right place, that you're getting positive feedback from the world. And our other sort of maybe controversial statement is we take a very incremental way of building software. I always say there's two kinds of migrations, incremental ones and failed ones.

So—

So when we come to companies, we say, let's not boil the ocean. If we have to help you rebuild your software platform, your storefront, we can take a slice of that workload. We can say start with a page or a set of pages. Actually, one of our biggest customers started with their checkout. They were like, okay, Vercel, if you're for real, then let's optimize the checkout first because Vercel makes it better and faster. I should see a conversion lift. And we can run that in itself as an experiment. So as customers were approaching the checkout, we can sort of bifurcate traffic. We can say, okay, Part of this goes to the A part of the test. Part of this goes to the B part of the test. And now SATSIG, appropriately named, gives the confidence to the companies that they're making statistically significant conclusions, that they're informed by data and analytics, while Vercel helps the developers implement the systems that, if you try to do it yourself, is not trivial. In many ways, some people call us the easy button to the cloud. Why have to figure out all these pieces and glue them together if you can kind of get the blueprint? From Vercel.

Love it. Well, maybe let's go a little out of order with how we've been going with other folks and ask you our second question first. You just mentioned data. Can you think of a time in your career at Vercel or before where seeing data, seeing a statistically significant amount of data on a particular question has completely changed your mind?

Oh, absolutely. One that is very recent is one opportunity that we figured out with Vercel is, okay, we're making these applications easy to build, but you still need pretty experienced developers. And AI is becoming very, very, very good at generating code. So we created this product called v0. v0, appropriately named version zero, and also shares the V with Vercel. Thank you. It's pretty good. It helps you get started with an idea. So instead of having to start with a code editor, you start with this ChatGPT-style interface. Think of it as like a ChatGPT for web developers. And then you start saying, okay, I want to try out to build my conference registration form for my speaker proposals. And you just write that down in English and v0 will give you a generation. It'll give you the user interface. And if you don't like it, you just tell the AI, "No, no, no, no, no, like that's, it's too boring. Make it pop." You can literally say it, "Make it pop." Designers will get offended if a client comes in and you tell them, "Make it pop." But the AI will just gladly say, "I heard you, you're absolutely right." The AI actually always tells you, "Yes, you're absolutely right. I was so boring. I made it pop." So with v0, we launched the product about a year ago.. And we started learning a lot about what it means to have a data flywheel for the improvement of AI products. We actually set out to, one, we dogfooded the Vercel platform. We said, look, we're not going to spend a year building this, conducting user research interviews. We know that in the AI age, data is the new oil and is a differentiator between winners and losers. So let's put the product out and start getting data. And so one of the key things that you're going to hear about AI products is that they have almost like an acceptance metric. For example, the, the AIs that autocomplete code in your code editor, they try to monitor if you accept the suggestion or if you reject the suggestion. So we started really diving into the data and understanding, are people grabbing the code that we're generating? Is the code making its way into production? Are people returning to reference back conversations? Are they using this product daily? So those are the things that kind of by shipping a product that was pretty early in its lifecycle, we started prioritizing that data capture lifecycle, getting the thumbs down. A quick anecdote is a lot of AI products give you thumbs up and thumbs down on the generations. There's almost like a science to it because if you just get thumbs down and you don't get any further data, you can't really act a lot on the feedback. So folks have actually been iterating a lot on, okay, I thumb down and I can present the customer with some questions around why didn't you like it? And we started sort of understanding how to categorize the intent and the feedback that the people were giving us. And then we started using AI to process the feedback. And so today, a year later, we have a product that is infinitely better because we've captured that data flywheel and we set our sights on how can we make people return to this product and turn it into a daily assistant rather than something they use sporadically. Fascinating. Amazing.

Well, closing question for you. What are some examples, or maybe just one example, of a company or a person that you look at historically and say, they have done product development really well?

First company that comes to mind, and I'll tell you this, and I'm sure it's something you hear a lot, I like the early days of Apple. Mm-hmm. And specifically, the idea that the computer was gonna be a bicycle for the mind, I think is something that a lot of folks that are now investing in AI or enabling AI creators, I think about a lot. AI, if computers were the bicycle for the mind, it feels like AI is the rocket ship for the mind. It's like taking us into—

Didn't you say yesterday that the bicycle for the bicycle for the mind?

Yes, it's the bicycle generator. It's whatever bicycle you want, you can just like speak in human language and sort of get it out. And there was a sense of like this divide between creation and consumption. There was a quote at one time of like, give people wonderful tools and they'll build wonderful things. And when I think about Vercel's mission to enable the world to build the best products and ship them, I think about, look, I want to see more creation in the world. I want to see less of sort of the doom scrolling and like just vapid content. And I want to return to some almost like the early days of computing. When people were willing to hack with things and fork and a lot of the work that we do in open source is in fueling that spirit of like the world as it exists is for you to change and mold and deploy. And one of the beautiful things of building on the web platform, and this is potentially like the thing that I don't like so much about Apple, is that instead of a walled garden, we can build on top of the world's freest platform. Anybody can own their domain name, anyone can publish, at any rate, and we were talking about iteration velocity and publishing quickly, and one of the key things there is that, you know, you might wanna deploy thousands of times a day. You might want to get the AIs to deploy on your behalf millions of times a day. So I'm very excited to sort of deliver on a vision that's similar where you integrate the parts so well, the vertical integration between hardware and software so that you can deliver an awesome, easy experience for the customer, but also hopefully build this open platform that everybody can join in and build and publish amazing things.

And so if I am hearing you right, a mission of yours is to shift time spent on devices from consumption, the scrolling, to creation.

That's right. And with AI, I think we're at the cusp of something magnificent in that regard.

I love it.

Thank you.