Vibe Coding, Tech Jobs, & Democratizing the Web with Vercel’s Guillermo Rauch | Fortune Term Sheet

0:00Allie:This episode is brought to you by Accenture. When your advertising operations fall out of sync, everything else follows. Spotify and Accenture are working together to reinvent the rhythm of ad sales, using automation, analytics, and smarter workflows to simplify campaign delivery and access better data across the business. The result? Less time spent on operations, more time connecting brands with the moments and fandoms that matter most. Learn more at Accenture.com slash Spotify. This episode is brought to you by Facebook. So you were scrolling on Marketplace, and there it was, the bike you'd been searching for.

0:38Allie:You sent a message, and it turned out the seller was super chatty, kind of funny, and an avid cyclist. The next thing you know, you're in a cycling crew. Well, a community cycling group. The thing about Facebook, you might find more than what you're looking for. From a browse to a bike ride, this summer, find more on Facebook. I literally get very sophisticated attacks, novel ones, every single week. You personally? Yeah, personally. Hello, hello. Welcome to Termsheet. I'm Ali Garfinkel, Senior Finance Reporter here at Fortune. And every week on the Termsheet podcast, we go over the latest news and insights in venture capital, private equity, and startups.

1:21Allie:Here's where you can watch and listen to us speak to the most exciting figures in the private markets. There's always a lot going on in the private markets, and this week was no exception. Here's what's been especially on my mind. Last week, crypto exchange bullish went public in a rollicking$13 billion IPO. Now, after the success of IPOs like Figma and Circle and Chime over recent months, it's starting to look like we are so back. Important caveat here, some of these IPOs, not all of them, but some, are happening at discounts to the highest valuations. For example, Chime was at a more than 50 % discount to its highest possible valuation.

2:00Allie:That said, for a long time, I talked about us sort of being at the top of the roller coaster. But now it looks like when it comes to IPOs, at least, we're pretty much on the ride. Elsewhere, I've been thinking a lot about the state of startup hiring. Last week, I published a story trying to codify some of the changes in the tech hiring marketplace. And there's a lot going on. On one hand, you have entry-level hiring dropping precipitously, which of course makes me think of the haves and the have-nots across startup hiring right now, because the haves are the AI researchers, the AI talent. We're talking go-to-market, engineering, anybody who's worked at OpenAI Anthropic, one of these giant AI companies, and they are highly sought after.

2:39Allie:The pursuit of that talent is extremely competitive, not only from$100 million checks from Meta, but from startups really across the ecosystem. At the same time, there's sort of another interesting dynamic at play. It's kind of a subplot. It's called 996. It's the push to have people working in an office from nine to nine, six days a week. This is also happening sort of amid a push towards RTO for a lot of startups. And one of the things that this really makes me think is that the market dynamics are getting very strange. On one hand, you're pursuing this talent. And on the other, you're trying to get them to work as much as possible.

3:14Allie:It does make sense. But personally, I think this is all a little bit of a trend. And it's going to be one of those things, again, back to the haves and the have nots. Entry level workers in tech maybe have less leverage than they've ever had, whereas a very specific set of talent maybe has more leverage than they've ever had. Now, I'm so excited to introduce our guest this week, Ghirma Rush, CEO and founder of Vercel, a cloud platform that simplifies web development. He goes by G, and I'd argue his role in tech really transcends the unicorn company he's built. Someone once told me that every great technology has to have its PT Barnum, like a Steve Jobs or in the case of AI, Sam Altman.

3:51Allie:And I would argue that G is the PT Barnum of Vibe coding. Now, Vibe coding is the process by which I can use language like I'm using right now to become a programmer. G is super candid about what this means for the future and what it means to truly democratize the internet. Here's G. G, welcome to the podcast. Great to be here. I'm excited you're here. With you, I really want to start at the beginning because I've always been fascinated by your origin story. Tell us the quick version of the G origin story. In the beginning of time, I was born in Argentina, Buenos Aires, the outskirts of Buenos Aires.

4:33And I grew up in a humble but very engineering focused family. My parents were both engineers. They got me into software very early on. and I developed a passion for the web, for open source, and for basically building businesses online. So what I do at Vercel today is sort of an intersection of all my interests for the longest time. Empowering developers, empowering developers with great tools, all of those things were things that I, even as a kid, loved doing, breaking things apart, hacking things, and then presenting them in a better way. And eventually, of course, I moved to Silicon Valley, start companies.

5:13I started a startup, sold it. Later on, started Vercel, which is the biggest mission I've ever had to date.

5:22Allie:I was going to say, what do you consider day one of Vercel? The most poignant memory that I have that was like, this is the key founding moment of Vercel was I knew I wanted to start another startup after I sold my previous one. and then I sat down to work on what you would do today to start any business, right? That you started online. So you want to make a website. And I sat down and I've been at this point, I've been making websites for, you know, decades. Um, and it was just so freaking hard. How bad, like hair pulling out bad. What are we talking about here? You know, the, you know, the idiom of like, I'm not mad and disappointed.

6:06It wasn't even, like hair pulling i was like so disappointed with the industry and on one hand i knew that the future was here but not evenly distributed so i'll explain exactly what i mean by website like there's lots of ways of making easy websites i wanted a production grade application that was going to be the foundation of my company right so i needed to set up cloud infrastructure which is basically, you know, configuring AWS or Azure or Google Cloud. And then I also needed the tool set for creating a web application. So at the time, React was starting to pick up, open source by Meta. It's like the tool that every web engineer wanted to use.

6:50And so what I mean by future was there, just not even distributed, is like it was kind of becoming clear that these were going to be the foundations of the modern Internet, but they were so inaccessible. even to an expert like myself. And I thought at the time, if this is so hard for me, taking weeks, months, taking all this configuration, doing what we tend to call out of ourself, undifferentiated heavy lifting. It's like heat loss. Like we're all doing the same things. We're all doing the same exercises and they're in no way distinguishable to anybody that's accessing your business. Like if I go to, you know, fortune.com, and you set up the best cloud infrastructure in the world, that's not going to come across.

7:36It's not a measure of success that's going to come across all the way to the consumer of a product. It's going to be something that's more like, you know, like maybe like engineering pride or makes you feel good that you jump through all those hurdles and whatever. So I wanted to build a cloud that was very easy to use and that was very focused on those end user outcomes. Meaning when you go to this application, you're just, you know, taken aback by how awesome it is in some way.

8:07Allie:I would say, what do you think it was about the inaccessibility that so ground your gears? Yeah. Accessibility for me is, and obviously this word is a little tired and people say it a lot, but it's truly around democratizing access to something. Once upon a time, I guess, and Vercel has obviously contributed a lot to this, only a handful of companies in the world had cutting-edge tools and infrastructure to build great products that could reach millions or billions of people. So you had, like, the manga, people call it manga or fang or whatever, right? Like you had Facebook, Apple, Amazon, Netflix, Google.

8:48These companies are not just known for their products. if you were to ever join those companies as an engineer, you'd be taken aback by how good you would have it. Sometimes we call in the industry the developer experience. And when I say developer experience, some people might think, oh, it's like when the Google lunch, buffet, cafeteria, like the I can get a haircut at work. Yeah, the value became known for those things, but it's actually the fact that you can just ship things as a developer. You join, you go through a very brief boot camp at this company. They basically onboard you to how the sausage is made, and then you become very productive very quickly.

9:35A lot of these companies celebrate what is the time to your first ship. meaning you join a company like Netflix and now you can put out an experiment or a feature to billions of people within weeks of working there. And so that's what I wanted to democratize. I wanted to give people a platform for invention that was very accessible. So much so that with two clicks of going to Vercel.com, you're going to get a Netflix-grade platform, a Facebook-grade platform and now after all this investment and time and energy i would even there say many ways better and one of the ways that it's better is that it's open it's built on open source you know what happens to a lot of companies like google is that they build their own sort of like little like silos of the world and you become a google engineer you don't become an engineer that has like you know skills that are like durable and and and maybe like transportable to another company So Vercel is really powering a generation of people that can just build and ship things and democratizing what other companies consider to be a competitive advantage.

10:42Allie:And I assume that's part of why you put, you can just ship things on a billboard. Yes. A lot of people have, like maybe the older generation, say, wait, are you a company? It's like, are you FedEx? Are you like shipping things? So, you know, kind of a compliment taken, great company, FedEx. But yeah, I think we're seeing this with AI pretty acutely right now, right? Everybody can ship. There's basically two mottos that we repeat a lot. One is that anybody can cook. It's a saying from Ratatouille, but in the AI world, people talk about models as their cooking models. And I believe that with AI, everybody can now contribute to software, which is part of that genesis of our cell was, hey, let's make it so easy.

11:31that we involve a lot more people in this software development process. So there's that part. And then you can just ship things is that the world is built off of a lot of things that started out very small, that started out as prototypes, that started out as experiments. I used the word experiment very consciously earlier. When you're an established enterprise, you might argue, well, can you just ship things? Like, are you just, you know, going to ship my social security number to the internet? No, you have to be very careful with certain systems. But that doesn't mean that you can't experiment at a small scale and enable people, product engineers, PMs, designers, et cetera, to convert their ideas into reality more quickly.

12:20And that's, those are the two pillars of a lot of what we do oversell.

12:23Allie:So, Ratatouille is actually kind of a great segue to vibe coding, as it turns out, because the central idea of Ratatouille, right, is that not everyone is a great chef, but a great chef can be everyone. And the central idea of vibe coding is maybe not everybody is a great coder, but a great coder can actually now be anyone. How has the discourse around vibe coding evolved? Yeah, it's been super interesting because it starts off with a tweet by Andra Karpathy, noticing that even though he was an expert, he was getting into a mode. He was getting into, you know, a mindset of, I'm not even looking at the code.

13:02I'm transfixed by the output. Immediately, that actually really stood out to me. At this point, our vibe coding platform, V0, already existed, but it really resonated with me because I've always been a person that likes to work backwards from the end result of products. I've always liked to start with the user experience. And then I start thinking about the infrastructure, the data structures, the algorithms, the computer science. And Vibe Coding is all about doing exactly that. It starts with the intention of the user. If you go to v0.app, you can just put in a prompt. And then you say, what do you want to see built in the world?

13:43And to your point about the mirror with the movie, it doesn't mean that everybody's going to ship products to production directly from a prompt. But it does enable everyone to get into the conversation. One of the emerging things that we've seen is enterprises are really excited by, quote unquote, vibe coding. because everybody that was not an engineer has traditionally been a little out of the loop, right? Like you have an idea on how you want to make this product that we're using now to record this. You have an idea about how to make it better. You just go and tell the AI, hey, draw me up this idea that I just had for how to better configure volume and camera preferences.

14:31And now instead of that being sort of this dead document, it becomes a quick app that you can share with people. So for us, Vibe Coding is evolving to become more of a communication tool. It better delineates conversations that teams have when they're doing their day-to-day work. And again, as every company becomes a software company and every business becomes a digital business, these tools are more important than ever before, especially for the people that are making that migration from where they are to where they want to be online. A lot of what the world deals with in software is like daunting migrations.

15:11You asked me about, you know, what was grinding your gears? What was like frustrating about that world that you want to leave behind? And I'll tell you, one of the things that really frustrates me and frustrates a lot of people is you can have a vision for where you want to be, but getting there feels daunting. migration has become a bad word in our industry like i don't know you want to think about the next big way of writing your stories and you're like well ali let's talk because we're in the we're amidst a platform migration from one content management platform to another content management platform and it's going to take several months and things are going to be kind of broken in between And so pipe coding is promising to become even more than just a quick way to prototype.

16:01But agents are now allowing us to modernize software at scale to do a lot of the work that engineers have always considered to be daunting. Very much like I mentioned, you know, configuring the cloud and the infrastructure and all of that is heat loss. There's a lot of our tasks that we do day to day that are heat loss. That are very much like a translation process. I think, you know, being a native Argentinian Spanish speaker really benefits me because I've been able to sort of over time grade the different translation systems. And I can tell you that if you ask ChatGPT today, translate this thing into Argentinian Spanish, it's going to do a perfect job.

16:45Like problem solved. You know, my aspiration to become a full time English to Argentinian Spanish translator, done. I believe that a lot of the work that we do as software engineers ends up being a little bit like that, that you're just translating from A to B. And then there's a whole bunch of work that is very much creative and accruing value to the business. And that's a work that, you know, with Vercel's AI cloud, we want people to focus on that work. And then we'll delete all of the stuff that is sort of repetitive and daunting.

17:17Allie:Well, and this actually gets into some questions about the future, right? For example, how do you see the future of computer science jobs? So there's one common meme that's emerging, which is there's no jobs anymore for junior engineers. And it's kind of become like a funny thing, right? Because people say like, oh, if you have a computer science degree, you're cooked and make jokes about it and whatever. There's an inkling of truth insofar. there is a kind of work that a junior engineer could do, which is somewhat deterministic, like I mentioned, like it's repetitive. It's a set of easy tasks to sort of, you know, start contributing to that code base, that first ship when you join the company.

18:03It's very true that an AI, an agent today, can probably do that job better than a person that's fresh out of college. So I have an idea, hey, like change this thing to the website, make that your first ship. You can do that with Vercel today and we're probably going to outperform every junior engineer. But the other side of the world that I think is being left out of the discussion is that we've never hired junior engineers to just do those tasks. Those tasks were just the proxy for how we ramp them to the code base, to the jobs to be done, et cetera, and how we get them up to speed. So they can actually ultimately really understand the business, really understand the customer, and take on larger challenges.

18:51What's going to change quite profoundly, I think, is that you're going to go from being a person that writes code to being a person that orchestrates agents. the act of writing code does seem to be getting automated away. This is not new in sort of the computer science world because over the years we've been building compilers and frameworks and technologies whose entire job is to raise the abstraction level. Writing code is becoming easier and easier and easier to the point where we don't write it anymore. But the human that's in charge is sort of orchestrating all those agents and sort of being the, you can think of it as moving from being a junior engineer to a junior engineering manager.

19:38And that's somewhat of a completely different skill set, right? Again, your first day at the job was write some code. Your first day at the job would be like, hey, like round up this team of agents, make them productive, and make them do something that we need to do for the business.

19:56Allie:When you think about the future and how vibe coding can sort of revamp our world, I know there are concerns, say, about cybersecurity, right, or concerns about jobs. What do you see? Yeah, on one hand, anytime there's a new revolutionary technology like AI, I think people's alarms go up a little bit. So people are more fearful of what could the consequences of this technology be. And I think in the world of coding, it's rightfully applied. To put it this way, even one line of code being in the wrong place can be devastating for a business. one line of code can be in we're talking about like 20 characters can be responsible for authenticating a user can be responsible for enforcing a permission can be responsible for applying the right price to a product that's being sold so i have an enormous amount of respect for the problem building public cloud infrastructure at versell the security has been a concern for now 10 years so we also have a lot of the methodologies processes and infrastructure to create secure software by default i mentioned one of the things that motivated me to start up our cell was i wanted applications to be delightful and fast by default i also wanted them to be secure by default security is really hard a developer we talk about a junior developer it's like fresh out of school they might not know all of the best practices in every detail, every dot and every eye that goes into securing production-grade software.

21:35And so the security problem has already existed before AI. I don't know if you heard about the T-Dating app that, you know, there were a lot of women on this app uploading photos to do identity verification, their driver licenses, their selfies, very private photos, obviously, and they all got leaked to the internet. I did a huge analysis on X of the attack because I, I not only at Versailles, we invested a lot into cybersecurity. Like I'm also personally super interested. I mentioned one of the things that motivated me to learn coding in Argentina was the idea of like the hacker, the movies, the, you know, getting into systems, et cetera.

22:14So I nerded out really hard on this because I actually wanted to get to the truth of like, is Vybe Coding responsible for this? And so the answer is actually no. It was an engineer that unfortunately didn't know better. And so I think therein lies the answer to this problem, right? Like the securing of code in the AI age is very much like the same problem of an autonomous car, where we will compare, you know, how many accidents is Waymo or Tesla robotaxi causing versus the benchmark of how many accidents are human drivers causing. And so when you think about security in the cloud, there are a lot of accidents caused by human drivers.

23:01Juro developers, tire developers, you know, I'll be the first one to admit, you know, when your brain power is not at full throttle and you're a little tired, you might slip and not have that one line of code that I just talked about, right? And so then you have, to also do justice to humans, AIs also make mistakes. That's why a lot of people thought that T-dating incident must have been caused by bi-coding. How could someone be so sloppy? But when we did the forensics, it turned out the timeline didn't add up. That bucket of photos that was unprotected had been created before bi-coding even existed.

23:39So, you know, not AI to blame in this case. But yeah, people are rightfully concerned. AIs can also make those mistakes. The approach that we've taken at Vercel is we hard steer the models and we write benchmarks and tests and evaluations to make them meet and exceed the human standard. So we check that all of the best practices of all of the experts of the internet that we've hired of ourselves, the people that started those frameworks and platforms at Google, we've had them sort of like write down the tests. It's almost like you're grading AIs all the time, right? Like anytime a new model like GPT-5 comes out, people are wondering like, is it smarter than Grok?

24:25Is it smarter than Anthropic? And so on. And so you get these benchmarks, right? Those benchmarks are sort of like, think of them as like the SAT. They're standard. Everyone is taking the same tests. But it's really important for companies to have their own tests. So for Vercel, we keep tests of, hey, is this the right AI agent to use for this kind of problem? And when we write code and review it, we say, hey, like, what is the percentage of real problems that these agents are finding when we do security scans and so on? So my very optimistic take here, and also from the data that I've seen, an example is the zero preventing thousands of security vulnerabilities per day that the LLM would be, you know, kind of tempted to introduce.

25:15I'm very optimistic that very much like a self-driving car, we will prevent a lot of these security incidents. And the AI will be so alert and so steered and aligned and focused on this problem that we'll end up being in that area. I think we can exceed the human potential or at the very least be a trusted advisor of the human that's doing that final review and saying, I sign off that this code is secure.

25:46Allie:Gee, you've also invested in more than 100 companies on your own. What role does all of this awareness sort of bring to your experience and time as an investor? Over time, I collect frameworks of ideas. so very much like Vercel creates frameworks to help people build applications you know I didn't mention this but one of the things we're known for is we have a framework called Next.js we have a framework called the AISDK and developers love it because instead of giving them like a bag of lego bricks we give them almost like prefabricated structures that are easier to reason about that's what this frameworks this developer tools do and over time my success as an investor has been predicated on i built frameworks for how to reason about successful companies instead of always starting from scratch and looking at every detail of like where was this founder born and you know like what school did they go to look at broad strokes does it fit within a framework of how i see the future evolving and so when i started versell and right around that time i started to notice that companies that hid enormous amounts of real world complexity behind very simple apis for developers were going to be very successful and so at the time i noticed stripe was doing that for payments twilio was doing that for communication And then I went on to invest in three companies that fit that criteria.

27:25One was Auth0, which was doing that for authentication and identity. Same formula. World of authentication, extremely complex. Presentation to developers, handful of lines of code. And that company ended up being acquired by Okta for$6.5 billion. then I invested in another company called Clearbit that did that for lead enrichment or augmenting information about converting an email to who is this email who is this person what can I sell them so basically like an information API and they're being acquired by HubSpot did really well and then the other one perhaps the biggest one of this bunch was Scale AI, which was doing basically the API to knowledge and evaluation of data that could be used for training models.

28:28It's very much an extension of what I want to do at Vercel, right? So it's about taking very, very, very difficult processes, whether it's the cloud, whether it's frameworks for creating AI agents and condensing them, removing all of the complexity away from the would-be buyer. And I think the bet was that buyer was going to be increasingly a developer and someone building software. And not necessarily like, we didn't even have to build up a lot of the traditional marketing artifacts. You know, the very complex PDFs and webinars and things like this. We just skipped right through, hey, developer, here's how to implement the software.