Why Great Companies are Defined by How Many Things They Say No To

Yeah. The first thing I ever did was creating websites for my passions and and my interest when I was a kid. At the time, I think it was, like, Dragon Ball Z, and I was learning just enough HTML to be able to put up a a website online. I would use a tool called FrontPage, and you can say that my passion has always drawn me towards the web.

Yeah. That’s a funny story because I would have never imagined that I would drop out of the high school that I spent so much blood, sweat, and tears getting into. So, basically, I could stay up all night during my early high school days. Like, I had these two lives. Was I trying to be the good student, and I was trying to also be the entrepreneur, freelance developer, open source contributor at night. And these things get more and more and more into conflict as it years, went by.

It had less and less time, and my work my my side hustles of open source and and work were doing better as well. And I was also starting to help my parents financially with I would, like, work online, get bounties for solving different problems in like existing open source or private projects. So by the time I was almost ready to finish high school, one of the open source projects that I was developing at the time called MooTools. So I had to make a very tough choice at the time.

I’m in Argentina. I’m 17. This company from Switzerland invited me to do a kickoff for a project on-site in Los Angeles, Switzerland. I had two subjects in high school that I needed to complete. My path took me to actually lean more towards work. I flew to Switzerland, then that company opened an office here in the South. And once I got to know the startup world here in San Francisco, it was just a one way street. I left everything behind, moved here, and here I am now.

It’s really interesting. Yes and no. So the very first time I made money from the Internet, I was 12 years old. The eBay of Latin America, MercadoLibre, had kicked off a revenue sharing program. I remember at the time I made like a 100 pesos. So that was the very first time I showed my mom that my skills could be monetized in some fashion. And over the years, the evidence accumulated that on the Internet, you could do amazing things. I think I was, like, 13 or 14.

I was already contributing to my family’s finances because the Argentinian economy has always been in shambles. And there’s been this huge discrepancy between the purchasing power of the US dollar and the Argentinian peso that is subject to hyperinflation. People in Argentina have always succumbed to hyperinflation. I would bring in, I don’t know, even like $20 from solving like CSS bugs on a freelancing website, and that would mean the world to my parents to make a bill or make ends meet here and there. I helped them out more and more.

So by the time I was 17, me saying, oh, by the way, you need to sign this paperwork so I can leave the country for this one trip. Like, wasn’t that strange. So on one hand, it was difficult to, like, abandon that really curated yellow brick road to traditional education. But on the other hand, over the years, like, everything was, like, showing us that this path of software engineering and the Internet could change our lives. So at the end of the day, even for my parents, it was easy.

Yeah. Yeah. A 100%. It’s like the more money, more problems. I think the one time that it showed up the most was my revenue stream was strong but unreliable. What it created was a certain sense of, like, we really can’t depend on this, but it’s really nice when it’s there. So it did come with some drawbacks, and it was also so much fun. There’s this intersection between like, what do you love and who what are gonna make money with. So overall, even though it came with its own stressors over the years, I think it was an amazing choice.

Yeah. That’s excellent because I truly believe that only immigrants can have a very in-depth appreciation for the unique qualities of the place that they immigrate into. And I do think that appreciation for what’s unique, different, special, awesome about a place or even a company is something that really empowers the individual. When I think about who would I recruit to be a part of this journey of building a startup, I really think that the people that appreciate what you’re building and the mission, even before they join the company, are the ones that are most motivated and driven.

I also believe this to be true of people that come into let’s think of a country as a super company. In fact, folks like to call The US an experiment because it’s so young. It still is a system. Right? So it’s like this massive startup that draws people in. There’s still a high bar to even get in, first of all. Like, it was really hard for me to even immigrate. So you’d really have to have this appreciation for what’s already here and what that can set you up to do.

So you called it infrastructure. What you can do in your lifetime will be a function of the underlying infrastructure that you’re given. And if you’re not given that infrastructure, the best thing you can do is move to where you can find that.

I think so. And I think that despite the turmoil, there’s something very compelling about having the right framework that sort of governs and keeps the system in balance over long periods of time. So really what I’m betting on is not the present state of a local micro measurement that you could make about the system. I’m a bigger believer in the rules, checks, and balances of the overarching system. Something that a lot of folks will point out about Argentina that’s broken is the justice system. And I noticed that even though, like, things can be in the short term unjust, I think the system coalesces into fairness and justice much more frequently here than where I come from.

So I think really you’re not betting on, like, what do you read on the news or on the front page of a newspaper on a given day. You’re betting on how does the system evolve over time and which system seems to be more likely to honor its foundations and its checks and balances. Those are the things that I look for when evaluating, let alone engineering systems, but also countries.

I think a lot about our mission as working backwards from what outstanding products look like and work like. You come to Vercel because you wanna build a great product, and you want our infrastructure and our tools to enable that. So I’m constantly thinking, okay, what does incredible look like and work backwards into the implementation? And that’s why actually we posited the market that frontend is so important. Even though it was overlooked for years and years and years, it goes back to, like, when Haven started. My dad was like, why are you wasting time with, like, JavaScript?

You should be learning hardcore back end languages and things like that. But in reality to me, that’s so upside down, it’s so inverted because the thing that matters the most is the look and feel and the capability of the application, the device, and then how you make that happen. I couldn’t care less as an end user. Right? So let let’s bring that back to how I evaluate folks that wanna join the company or companies that we acquire or folks that we’d reach out to is that what are they putting out into the world that is tangible?

What can I use? What hyperlinks have you shared with the world about your creations? This is a beautiful thing about the Internet. Right? Like, you could be anywhere in the world, and a hyperlink travels to me within a 100 milliseconds. So it really ends up being that you’re communicating your capability through your creations. I wanna see more of that in the world.

That’s that’s what I said. It’s not so bullet. Although, you know, there’s something even about, like, your writing. Your writing could be that frontend to your mind. So the ability for you to convey your thoughts, the discipline to actually follow through with your writing. What is the quality with which you present yourself and your experience? I mean, front, it could even be the way that you talk and the stories that you tell. I actually encourage a lot of folks to with the attribution problem in particular.

Right? Like, I was at Apple for twenty five years. I can’t even say what I worked on. And but there’s still a narrative of where did you move the needle? Is it measurable? How are you able to relate and express the business impact that a certain contribution you made to even maybe some internal of a distributed system or compiler? Like how did that actually permeate?

And if you can connect the dots all the way to how you made someone’s experience better, because I think this is the other thing is at the end of this rainbow, there’s either you’re making some end user experience better, whether it’s a visitor or a user of your app or some internal process like I made the life of a developer better. So you have to be able to connect the dots from technology to sentiment and even the way you talk about it.

There’s no depth. That’s a great pushback. What I do sometimes is I mean, this is probably an advantage that we have at Vercel. Right? Like, you wanna talk details? Let’s get into the details. Right? Like, let’s let’s go into the specific down to the Cisco of how you implement it and solve a certain problem. Let’s talk about what cloud infrastructure you use. Let’s talk about what options you wait on the table. Let’s hear your awareness about what vendors and what alternatives existed. I think being able to dive into like the absolute depth of the problem is something that everyone should be able to do.

That doesn’t mean that you have the complete texture. I think really what you need is the conceptual understanding. You might not know exactly how, like, writing a custom module for the Linux kernel works, but you should be able to understand how the parts fit together if you’re evaluating a systems engineer that’s joining your company. That texture is what allows you to discern better whether someone is just telling a good story or there’s actually depth to the story.

I have an easy one because I have betrayed that rule in the past. Sometimes I’ve placed more weight than I should have on this stellar linked list of awesome brands that you worked at. I did x at let’s just name random companies for the sake of it. Apple, Google, Amazon. And you look at that and you’re like, well, four companies in a row couldn’t have been wrong about this candidate. Right? Again, I think you do have to have a curiosity about what that candidate can bring to the table.

And when I’ve overweighted brands on the resume and I think this for other folks, this could happen with universities, I think. That’s one way where looking back, I think I could have done better with a few candidates. But but, yeah, I think at the end the day, it comes back to, can you evaluate the opportunity in its more objective and measurable merits than the approximations of merit?

No. I think you’re right. I I think the only risk there is that the pool could be constrained, but you don’t have to get every single opportunity right. You just have to get a few great ones. And I think as a way of constraining the state space, the search space, I think that’s a very good technique. I would add to that from my own personal experience, like someone that had an experience being an early member of a startup and was able to learn a lot about, like, I had seen decisions a, b, and c get made where I would have chosen the exact opposite.

And I was like, okay, I wanna start something because these non obvious decisions that I would have taken a different path, I can now take the right path. And I felt like that was almost like an unfair advantage in the marketplace. It really is all about finding whatever alpha you can across a number of dimensions. Right? And to me, I was like, okay. Like, I wanna do hiring very differently. I wanna do product development very differently. I wanna just move faster, period. I felt that even without even getting into the weeds of the technology, I would set up different foundations for success for what I would have started.

To your point, I had my first startup which had a moderately successful exit, and my second startup built on the learnings of the first startup. So it is a good heuristic.

No. And it’s this is for the same reason I said, I think the approach you described is good because it won’t get you a 100% of the great companies, but that’s not your job. Your job is to even just find a few of the greatest companies. Right? And I think that’s also true for product development. The inspiration for me is, like, Google search where, like, there is nothing to do except for an input and, like, everything is so simple and, like, the magic is in the implementation and there is all these unfair advantages of access to data that gets constantly refined and all that.

But I do think that if your job is, okay, I wanna find a few of the greatest products that will be produced in this generation. I do think that a lot of those will be this deceptively simple things where the interface of the user is just magical. The company took on the burden. So one of my earliest investments was Auth0. It was one of those things where I was incredibly skeptical, and yet I was like, I’ll do it. Like, this is all about asymmetric upside. Right?

But I remember the thing that I rooted myself in was a simple API call that hides a tremendous amount of complexity behind. So I never wanna deal with, like, Mastercard and Visa and, like, all these payment gateways and all this nonsense retries and workflows and latency. And on the other side, you have, like, this awesome simplicity of, like, here’s the API call. Here’s the curl command that you can run to get started. So I do think that a lot of companies will follow that. Now where it gets more complicated is the famous, like, what got you here?

What gets you there? What got you to Series A or what got you to Series B is not necessarily what gets you from Series B to IPO. We and I mean, like, in the entire, like, tech industry, we can become so formulaic to our own detriment. I think in the early stages, a company is trying to find its wedge into the universe. I actually remember a fascinating conversation that I had with the original CEO of Snowflake, whose point of view was that open source was overrated.

And by the way, I was super open source build Next. Js. Even before Next. Js, I I built socket.io. I contributed a lot of Node. Js since, like, 0.1. Mongoose, the most popular for MongoDB. So I’m, like, drinking the Kool Aid of open source. I’m drunk on it, and I have this guy saying, well, you know what? Like, to me, source solves a concrete problem, which is market positioning and developing market awareness. It solves a zero to one problem. But open source is in this area solve the one to 1,000,000 problem.

Right? And if you look around, this is certainly true for every successful open source company. Right? Like, the reason that GitLab, Hashi, you know, a lot of these companies are successful is not just because they did only open source. So I think it’s really important to understand that we are operating in a sequencing. In the beginning of my career, I tended to look at, like, series seed a, b, c, d as so silly, like, arbitrary. But I do think we have to think more in terms of sequencing, and there’s a long journey ahead, and you have to be very adapted at each step of the sequence.

Yeah. I I think the product development strategy that gets you to the and, again, like, the reason I like this a, b, c, and d things, they’re arbitrary, but they help us anchor, and they help give us a common language to evaluate and pressure test different frameworks. The simplicity of your product from zero to one might not be the simplicity of your product from one to two and two to three, or let’s call it series c to a to b to c to d. This is very clear in that companies become multiproduct over time.

And it’s really easy to look at AWS and say, woah. What a cluster f they have going there with, like they have all these products and, like, they have all this complexity. Yeah. But, like, that’s where they are at now. And, certainly, if you’re trying to compete with them, you’re not gonna replicate their approach. You’re gonna try to bring a simplification to the market. But over time, you have to reevaluate your priors because the challenge, the market, the growth, like, things are changing over time. So you asked me the question of, is what I should be looking for only really simple products?

And to me is the answer is very much stage dependent, very much market dependent, and there is no universal truth to the question.

Yes. My answer to that is we don’t need to fight every battle. The job of a good VC will be to find two or per fund, right, a handful of the best companies in the world. My job will be to create a few of the best products that I can bring to my category. I don’t need to solve every single problem my customer has. I can partner, I can push solutions to user space, I cannot enter every market. I can wait and hold. Meaning, whenever there is a temptation to solve a problem with urgency, sometimes you’re better off waiting.

Right? Like, this is something that Apple does remarkably well because they don’t react to every trend that emerges. They pick their battles. And markets, yes, sometimes there’s a opportunity that you have to catalyze at at a given moment, but sometimes you’re better off letting the truth emerge from all of the experiments that are constantly being run, and then you pick your solution. And this actually matters more as the company becomes more mature. Because, again, you don’t always have to be first. You have to be right.

And that’s what I think companies over time optimize more around.

So number one, culturally, there’s an expectation at Vercel. We will go deeper into what we consider to be complete or ready for a product then I think most people are willing to go. We spend a lot of time on what we would call the internal testing of a product before we feel is ready to get to market. Setting up a really, really high bar for what gets to the customer’s hands. This is really tricky because I think some people might confuse having a high bar with being slow.

And I really think this is the crux of the problem. I actually had a an entrepreneur reach out the other day, and he was saying, like, the number one problem that I have is I wanna move really fast, but my customers are telling me that I’m breaking their stuff too frequently. So my advice was you can find ways of reducing the blast radius of your experimentation such that when it’s the time that you bring the product to everybody, you’ve accumulated enough evidence of success that you have an overwhelming amount of confidence in what you’re bringing to the market.

So it’s easy to dismiss when, like, these keynotes happen and and say, like, this is the best phone we’ve ever made and like, you can dismiss some of those statements as just a an effective product marketing team or you can actually earn that confidence yourself. And when you’ve done all of that pressure testing sometimes I talk about this awesome video that I come back to on YouTube. I send it a lot of people of Boeing bending the wings of the airplane inside the factory beyond the point that is actually realistically possible that a turbulence event will bend the wings.

They will bend them so much that it looks like the plane has molded into, like, a cylinder. But that gives them the confidence that, you know, you can do millions of flights a year and tolerate any storm and tolerate any weather event. And, yeah, the wings will bend, but you’ve gone above and beyond to certify and sign off. This is the plane that I want people to fly and feel safe inside.

I think UIs will change. If we can go to a chatbot, and that chatbot is acting as a simplification of what used to be a very complex dashboard of lots of menus and submenus, and like, you’re like, where did they go to find this and that? But at the end of the day, the information has to be surfaced to you. So I think it might just be that, I’ll give you an example. I just saw this incredible approach to surfacing data that you just interface with the system with natural language.

And then the system is basically guiding you to where the data is. Right? Like, it’s surfacing what otherwise would have been maybe like twenty, thirty different clicks. Now the AI is just like serving on demand. With some of our products, we’re basically going in that direction. Right? Like, what used to be dragging dragging and and dropping dropping and like spending like hours or going into a text editor and like going through like all these auto completions and like errors and now it’s just you go into UI that expects that you type in English.

But our point of view is that now people will be creating a lot more of those UIs. It’s a new tool in your toolbox.

I think it’ll be the opposite because my hypothesis of how I look at the entire AI space right now is very simple. There are a lot of jobs to be done that people will have to perform no matter what. Decisions that need to be made. There’s data that needs to get visualized. There is communication that needs to be made. Like, I need to send you an email. I need to make a phone call. Like, a lot of those things won’t go away, but how we do them is going to be profoundly transformed.

Let’s say that I want to create a new email client. Basically, what Superhuman did, I want to do again. What’s really fascinating about what’s happening today is that when we were creating software over the past ten years, part of our input into the design space, meaning part of what I thought was possible when conceiving a piece of software, didn’t involve AI in the past, didn’t involve the fact that we can have this reasoning machines that we can invoke on demand, call them the large language models or the AI as a service.

Now that I have those, how profoundly different will my design be? So if I have to create the next Notion, the next superhuman, my bet is that it’s certainly gonna look completely different. Because before, I would reach for what I would call, like, software one point o solutions. And over what makes someone really good at product design, frankly, is that they looked at lot of designs over their lifetime. They looked at a lot of data. They have a good instinct. It’s almost like a good Go player.

What made Lisa Dole a formidable adversary to AlphaGo is that there was this instant, this pattern matching. But now it’s almost like the board has shifted. What made you a good software one point o designer, it needs to be actualized. It’s not that you’re not gonna be good at a software 2.0, but you need to say, okay, now when I sit down with my conceptually empty canvas, now I have this other set of tools that I can use to solve the problem. And it’s exciting because startups don’t have any priors.

They don’t have any legacy software in which they have to retrofit AI onto. It’s a great opportunity for incumbents potentially because they can now say, I added AI, ship it, and they can increase their TAM. They can say, I’m ready. But it’s also a really scary situation to be in when someone can come in and produce a completely different design, if that makes sense.

I I think I agree over a long enough timeline, but that doesn’t actually result in practical advice for what to do over the next three quarters or even couple years. I think the most successful innovations tend to meet the world where the world is. So when I think about the iPhone, famously, there were two concurrent experiments for what would become the operating system of the iPhone. IPod OS evolves, and it was a fresh new code base. It was already on mobile devices in the sense of, like, it was already being deployed to, like, smaller miniaturized hardware that was portable.

And concurrently was the strategy of let’s retrofit macOS to fit it into the iPhone. And, obviously, macOS being forked into iOS is a strategy that won. And it made the world where it was. The first killer app was being able to downsize websites and put them into a smaller screen and then find new evolutions within that which became responsive design and so on and so forth. And then new native things happened within the context of that platform, namely new applications that weren’t possible by just retrofitting, new applications that took advantage of the medium and new capabilities that were unique to the medium.

But even fast forward to today, like twenty years later, we’re still in that world of there’s this hybrid of the DNA of what already existed with all of this new DNA of what’s now possible because you’re on a mobile device, and they continue to coevolve. I think this will be the coevolution of AI and traditional software as well. But, again, going to, like, what makes practical advice possible, I think a lot of those chat type interfaces will have to meet the customer where they are. If you say to the AI, please help me make an advertising that’s gonna be deployed onto x in Meta or Facebook and Instagram, it’ll have to give you some UI as feedback of what’s proposing back to you.

And then you’ll have to interface with the system further. So I think it’s still all UIs all the way down. Now the key magical part of this is that they are UIs that are unique to the AI problem space, and those UIs will definitely be different.

One of my hypothesis is that we actually, in some ways, have to give people more UI. Let’s actually look at what’s succeeding in the AI world today. We have mid journey. Mid journey gives you this text interface, but as a as a result of your text prompt gives you four choices of what you could possibly like. That’s what I mean by it’s actually giving you more UI, not less. It’s uncertain about what it is that you actually want, so it needs to give you more choices.

This is very different from software one point o, which is the world of determinism, the world of rule following, the world of predictable algorithms. If you give me a form and I put a, b, and c, I give you result d a 100% of the time. And if it’s not d, someone gets paged because we monitor it. Right? Now software one point two point o AI, I don’t know if I’m what I’m giving you, you’re actually gonna like. I don’t even know if it’s offensive because, like, there’s all this, like, craziness that happens when, like, you give this AI that is, like, not supervised, you gave it a lot of data, and you can’t even can’t even comprehend the amount of data that it has.

So you actually have to create UIs that tame the craziness of the machine. You have to create UIs that give the creator feedback. So another easy to dismiss thing about the brilliance of some of those products is that because they’re giving you choices, they’re now feeding data back into the system to make it better in the next iteration by once again relying on new kind of capabilities that are more in this realm of probability rather than certainty. So I think that’s a very important departure is that now we’re entering the realm of I can assist you in the creative process.

And, again, that means that you have to think about software differently.

It depends on what stage of the problem solving workflow you’re in. If it’s in in a very creative side of the process, I do think having more options is better. I like to give the Rick Rubin reference of, like, his contribution to the world is telling the artist if they like something or not. He says in that interview, like, I’m not an expert in the specific music techniques. I don’t know how to play an instrument. The feedback that I give to the artist is what’s valuable.

What I like and what I don’t like is what’s valuable. So if the tool is very much in the creative stage, I do think having options to go through is really good because I I was mentioning, in my experience was made a designer of a certain piece of software or UI really good is just how much they’ve rejected over time. What makes a company good is how much they reject over time. How many things they say no to, how many seemingly attractive on the surface ideas they rule out before they decide, yes, that’s the one that goes to market.

Yes, that’s the one that earns the position being in our product lineup. So it’s really going through lots of iterations of yes and no that it gets you to the point where you can say confidently, this is what I like. So I do think that accelerating the iteration loop in turning down things could be really healthy provided that the trajectory is good. What really matters in these AI systems is that the system is learning when you’re turning things down. Because to your point, I like the pushback because the best assistant in the world will have rich context about what you already like and will over time propose better variations of the solution that tailor to your preferences and tailor to the context that you’re in.

But we can’t be too absolutist in that process because otherwise we’re just gonna give you garbage. And notice that a lot of the successful AI systems so far have been in the realm of being suggestive rather than authoritative. Copilot suggests a completion to your text. Gmail Smart Compose suggested what to reply to the email. Notion AI is saying, I can go ahead and, like, complete or give you ideas or finish the bullet points for you. When it gets into the realm of authoritative, like agents, like, I’m just gonna let an agent go into the world and act on my behalf without supervision.

We haven’t seen a lot of products that are successful in that in that space.

I think we absolutely will, but it requires the close feedback loop with the person that that agent is acting on behalf of. It requires the person is still in charge. The metaphor that I like to use is being the editor in the newsroom. You’re having the agents bring you the ideas, and you’re saying, nope, go back to the drawing board. This is a feedback. Nope, this doesn’t make the cut. And you’re the one that’s also setting the creative direction. This week, we have to go after AI safety.

That’s a hot topic that people are interested in. Next week, we have to go into this. We have to go into that. So you’re still in charge, and this is why, ultimately, this is the healthiest version of an AI future after all. Right? Because we don’t want an AI future that inverts us and puts us in the we’re we are the agent. The AI is calling shots. That’s what people actually get scared about. So I very much like this idea of AI in the assistance of the creative process and AI just making processes more efficient.

I’ll give you a good example. Right? Like, incumbents are in a very problematic position when, as I mentioned earlier, the new AI alternative is just really disruptive to their historical approach to solving the problem. The best example would be Microsoft Word, the infamous bad design or a word shipped with a really thick toolbar that exposed every option that used to be nested within menus. They made it visual. 30 of the screen real estate was tools. I remember a lot of the reactions at the time was like, holy crap.

Like, I didn’t even know you had all these options, Word. The way that a company like that was looking at the problem of making the best possible word processor was very additive. It was a more is more approach to the world. What does this mean? That adding more utilities gets you promoted. Adding another way of changing the color of the text or even adding more colors. Now look at Notion, for example. It doesn’t actually even give you every color on the spectrum. The designers of Notion said there’s eight or 16 colors that mostly work well for the purpose of writing this wiki style documents and whatnot.

So they actually said, we’re gonna do less. We’re gonna have a lot fewer text editing and text formatting options. And that’s gonna give people agility. That’s gonna make them more focused. Now there’s an extreme version of this, which is that, can I actually not implement any of that? Because AI exists now. And my software one point o simplification strategy, which was creating fewer options or creating options that are more general, pales in comparison to an AI that can actually just, like, do exactly what you want even without having the options.

I actually didn’t software engineer any of those utilities and icons and blocks and whatever. The entry players can win because they do a lot less and because it’s so disruptive than the incumbent has to say, holy crap. We have to delete everything? Like, I feel so invested. I crafted every icon. I crafted every menu. I crafted every piece of documentation for how to use this. I recorded all the videos that taught you how to use those. So I think that’s the generational disruption that happens in software that is extremely problematic for incumbents.

I I do think that a lot of these companies can be underestimated. One thing I’ll give some kudos to Microsoft for is that they told the company they had to work and focus and have a look toward AI much earlier than a lot of other people. So they have years of strategically thinking in the direction of AI. Now, a lot of these companies have the right strategy, the right foresight, the right internal talking points, the right investments, and they still get disrupted because the new way of doing things is so radically different that initially they don’t even look at the alternatives as a competitor.

The best example would be potentially mid journey to Photoshop or to something like that. Like, I’m actually very impressed with Adobe’s incremental addition of AI. It was swift. It seems high quality, well considered. It gets the job done. Like, it’s strictly in a measurable basis, like, better. But there’s a chance that, like, again, there’s such a platform shift towards now the interface is going to mid journey where all of the other tools just don’t exist. They literally don’t exist. Like, all of the drag and drop and magic wand and selection tool and color switcher.

So people just go and spend their time elsewhere. I think this is how desktop software sort of became less important than mobile apps when that transition happened. It’s not that the right tactical determination was stop improving the desktop software, delete the desktop software. And this is what Meta ended up doing well. It was that the user minutes and the attention and the growth was gonna go to a different way of expressing their mission in the case of Meta, which was like connect the world. They needed to go and almost to start over.

You needed to go to mobile and start over because everything that you’ve built for desktop, again, it still exists to this day, very much like AM and FM radio still exist, but the growth is elsewhere. The growth is in this new approach. So you can have an incumbent that does that really well, but historically, most don’t, which is when the platform shifts and they need to do things very, very differently.

I think it’s an acceleration of what was already happening with cloud. If you look at Vercel, Vercel offers you two very powerful things for your organization. Number one, the most obvious and the most visceral to a lot of developers, which is that you give us your frontend project and we host it. We give you infrastructure to scale it autonomously. And that’s traditionally a consumption. You pay for what you use business model like Snowflake. But what Vercel also gives you is the iteration velocity and agility, very much like what Google Docs or Figma give you, which is like now I can collaborate really fast with hyperlinks, and I can share around my entire organization what everybody’s working on.

It’s almost like a Jira. My narrative violation is that businesses that only do one half of the equation are gonna become less and less popular over time. Because when I buy software, I expect to buy a comprehensive platform that solves business problems. And solving business problems is never about just procuring a very specific material, a cog. I want this to do more for me. I want it to be like an operating system for my organization, especially as you hit scale. You’re like, oh, I’m using Vercel for this little thing here to like host this, but I’m using Azure here and Google here.

Give me the foundation to level up my product development organization. That’s what most of my customers actually want. So the way that we’ve built against that is that we’ve done a combination of like SaaS that is by seat in purchasing platform capabilities with infrastructure that grows when your visitors use a software. I think AI from the outset is doing this because AI is about getting some workflow or some tool, but it’s also spending GPU cycles on very expensive outsourced intelligence, so to speak. And the more I use it, the more it costs.

If I use a lot of GPT four, it doesn’t matter that you’re charging me like $20.30 bucks a month. You have to like somehow model piggybacking on your infrastructure for intelligence a lot. So I think we’re gonna see a lot more of these hybrid models where you give me a baseline of utility, but it can also burst to utilize more of what you can automate for me, if that makes sense.

Right now, LLM is not as good as GPT four. Not even close. But with the ecosystem saying, this is what we’re gonna bet on. This is what we’re gonna throw all of our AI researchers. This is what we’re gonna throw all our hardware AI chip accelerator budgets. This is what we’re gonna throw in all our software acceleration budgets. This is where we’re gonna throw all our documentation. This is where we’re gonna throw all the user space frameworks, like Lama Index and Landing Train and this and that.

Like, now you have this entire community that’s saying we’re gonna bet on this thing becoming better. The counter thesis to that is that famous Jeff Bezos quote on why AWS got so far ahead of the competition. Even though the competition knew full well that cloud was a strategic place to go, is that even a two year advantage in technology can be lethal. Two years in tech is very hard to overcome. Five years in tech is just exponential. Ten years, good luck catching up to that company if for whatever reason you ignore their advantage for ten years.

This has certainly been true in our space. When we were betting on dynamic frontend rendering technology, a lot of folks were betting on static. And I feel like a lot of folks have now said like, oh, like Vercel, you’re growing so fast, you did this and that. And like, they think like it’s some act of magic. But going back, we just were willing to like bet on what was a little less popular, especially among VCs in the valley at a few crucial points. So, yeah, I can tell you like LLM will win or GPT4 will win because I have evidence in favor of both.

But I can tell you that every day we’re getting more data and it’s one of the most fascinating open questions to me right now.

Correct. Most important thing.

My prediction is just like what happened with web two point o where because of the platform shift toward mobile and social, you have countless successful IPOs from that generation. You have the Ubers. You have the Airbnb. You have the Facebooks. We’re gonna see the exact same thing happening with AI. Now, I can tell you exactly what the ratio will be, but I would be willing to bet that 20% of investments will be productive ones. Now, means companies that then get merged and acquired. Like, we might buy an AI company.

That doesn’t mean 20% will be unicorns, but I think we will be see a lot of positive ROI.

People underestimate just how much their own time goes into thinking about important problems that need to be solved, but you just don’t have the bandwidth to solve them. What what do I mean by this? If I look back on some of the most interesting investments that I’ve ever made, there are things that I would have loved to do myself. I just didn’t have the time and I wasn’t in the right place. Maybe I was working on another project, maybe I was working on a different company to actually execute on those.

Just to give you some examples, I invested in this company called Scale AI. I really badly wanted to work on a product like Scale AI. I don’t know if I would have been the person to build it. And I think I got really lucky by I chose the right horse in that particular space. But I do think your own experiences are very validating, especially I think this is what’s making a lot of founder CEOs successful in angel investing is that the things that you see, the things that you perceive and this is also why I’m so bullish on AI as well, like there’s so many AI products I want our company to buy.

There’s so many AI products that I want the world to benefit from. There’s so many inefficiencies. There’s so much repetition that’s happening. There’s this mismatch between supply and demand. And and that’s how a lot of great investments happen.

I think the founder needs to be uniquely positioned in a number of dimensions, uniquely willing to go through the grind. That’s a very important side of it. Uniquely positioned to either have the background or learn really fast what it takes to solve the problem. That’s one that’s kind of messed up about angel investing. Like, you can lose a lot of deals if you just look for expertise. Sometimes, and this is also true for my hiring philosophy, it’s about the slope. It’s about how fast people can learn.

Here’s a universal truth. Ultimately, it’s all about the market. It’s all about people’s willingness to buy a certain thing that determines winners and losers. That’s why I wouldn’t underestimate your own ability to judge what it is that you want to buy, what it is that you ultimately need. Now you might be actually really bad at judging the solution, so you have to be extremely open minded. I made a quick comment earlier that the thing where I got lucky with Auth0 is that I just didn’t feel like a 100% confident about their solution, frankly.

And if the founders are listening, they’re gonna hate me. Like, I felt at the time, holy shit, outsourcing off to a startup? How does that make sense from a chicken and egg point of view? How does it make sense that your homes and your company’s door and the lock, you outsource to the lowest bidder, like a startup that has three people. So I was like, okay, it’s kind of crazy that you want to turn auth into a microservice. At the same time, I do know that implementing auth, it’s freaking hell.

So this is the other advice I would give people is that there are a lot of things that people overestimate their own capacity to solve in an excellent way. What do I mean by this? Every company I would advise or we started that I would start, I used to set up a chat server in IRC. I would set up an IRC server so that people could collaborate. I knew the right problem to be solved. I knew that in order for software engineers to be more productive and collaborate with the rest of the company, they needed a chat medium to collaborate with others.

I had the wrong solution. The solution was not to set up an IRC server. Very few people know this, but when Slack was coming up, a few competitors to Slack were actually giving you IRC as a service with some UI. So the market knew that there was a great opportunity around collaboration with chat for organizations, but the spectrum of solutions was incredibly bizarre. Some involved IRC hosting. Some involved a hybrid of Slack with IRC. Slack even for a minute had an IRC proxy to help people move from the old world to the new world, and then they discontinued that.

It’s the solution that becomes kind of like the tricky thing to where do you place your bet. I remember when Matt from WordPress, I sold my company to WordPress, said we have to install Slack and deprecate IRC. And at the time I was like, woah, are we really gonna trust this company with all of our communication? Same visceral reaction that I had with my own investment in Auth0. Are we gonna trust this company with all of our x? So you can almost see a pattern there.

When you know that the problem is so painful, but your only hesitation is whether you can offload it to somebody, there might be a huge opportunity there.

The way we talk about this is we still live in a world where there are a lot of bad products. There are a lot of bad experiences that you have on a on a given day. Our dream and our vision is that Vercel has raised the bar of every experience that people have with software, with applications, with products. It has to permeate all the way down to the average person doesn’t have applications they hate, processes that bog them down. Why? Because we empowered every creator, every developer to fulfill their creative dreams, and we give them the tools to turn those dreams into the best products that they can.

And I think AI will play a huge role in this. I think part of what we’re doing now is creating a bigger funnel, bringing this technology into more and more people, and putting the creative process in in charge.