20Product：在 Agent 世界中如何设计和构建产品｜为什么 AI 会杀死许多 SaaS 产品｜在拥有1亿开发者的 Agent 世界里，哪些产品会繁荣、哪些会消亡——与 Netlify 联合创始人兼 CEO Matt Biilmann 对谈

I think we'll see the next 100 million software developers emerge around tools that are agentic, but I still think it will be a profession to use those tools to do stuff, to build stuff for other people. If I look at Google's products, I now have all these genesis buttons all over, and I never use them. They just haven't managed to productize it in a way that feels super compelling.

Matt, I'm really excited for this. I really like your writing, and I've loved your writing for a while.

Thanks.

Thank you so much for joining me.

Thanks. It's great being here.

I would love to start with this: how does building for agents, not humans, change the way that we build?

1. Agent Experience Starts With Humans

The first thing I would say is that when I think about building for agents, I still start by thinking about building for humans. In essence, you could say that there are 2 versions of that 10-year future. There's one where the agents just escape, where they're off doing their own things and we are sort of left behind. I hope that's not what happens.

But there's another one where agents are tools that all of us use to do stuff with, where agents are fundamentally always doing something because some human is trying to accomplish something through an agent. Now we are in the very early days of building a discipline around agent experience and really starting to think about, when a user tries to get an agent to do something through your product, what happens?

Can I be direct?

Yeah.

I don't understand. How does building for agents change the way that we build?

Let me give you some examples of stuff we've done that's been successful with agents. One thing that we built specifically to improve our agent experience was an API flow that would allow an agent to create a site and deploy a site without any user signing up for Netlify first, and then hand over that site to a user later with a claim flow.

Simply from the realization that when we have a normal developer, the first creation of the site will always be triggered by that developer doing something with Netlify. They'll pick a template and start, link to a new GitHub repository, or deploy from their command line. But once you have a code-gen agent, they might just ask that agent, “Build me a web app with a Sudoku game, whatever, and now share it with people on a URL.”

So we gave those agents the ability to just deploy to Netlify, with no sign-up from the user, show the user the URL, and say, “Here's what is built for you.” Later, the user can go claim it and say, “Okay, I actually want a Netlify account,” or, “I have a Netlify account. I'll want this to stick around. I'll put a custom domain on it,” and so on.

We've seen a lot of companies now start adopting similar flows in the authentication space, the database space, and so on, to make their tools available to agents without an explicit user sign-up and setup step first, delaying that step until afterward. That's one very specific example, but one that's been really impactful.

Does designing for agents in this way reduce the importance of UX, design, and taste when building?

I don't actually think so. In the end, we're designing for humans. I just don't think anyone can have a great developer experience 5 years from now if they don't have a great agent experience.

That also means that a lot of the traditional UX disciplines and understanding not just the UI design, but the UX design—the holistic experience of a user trying to do something with your product—are still really important. Obviously, a lot of the details are going to change. A lot of things we do today are going to require very different interaction flows and very different ways of displaying that UI to people.

If I'm a founder today—

Yeah.

—how do I tell my product team where they should focus, what they should build, and what they should do? If I'm setting strategy for them, what the fuck do I tell them, Matt?

You embrace agent experience as a discipline in the same way that you would do it for developer experience or user experience. I think we'll see that being the approach.

Rather than starting with some theory of, “We need this spec now,” start by understanding, in your space, what the agents are that humans are actually going to be using with your product. In our space, it's on the one hand tools like Cursor, Copilot, and Windsurf. On the other hand, tools like Bold, Lovable, Create XYZ, Refine. But in other spaces, that might be a different set of agents.

When I interview people today, Matt, the common public company short is Salesforce—

Hmm.

—for a number of different reasons.

Yeah.

But do agents merely turn Salesforce into a data repository?

2. Custom Apps Challenge Generic SaaS

That's definitely the big threat to Salesforce: to just be reduced to some kind of system of record. I'm mixed on whether, long-term, that's a really, really bad thing for Salesforce, or if it could also be a good thing for them.

If you really start becoming the main system of record that all of these tools rely on, I don't think you should underestimate just how valuable a position that is, because all those flows and apps and everything will run on top of it. Sure, it might change a lot of what Salesforce looks like today into what it looks like in the future, but they definitely feel it very strongly as the biggest threat to them: to be reduced to that system of record and not the actual workflow platform.

I think that's pretty likely what would happen. What we're seeing now is definitely that the cost of building custom workflows, custom UIs, and custom applications is going to dive down dramatically. Typically, that will mean that demand for doing that will go up radically, and people will start building much more custom applications for everything.

Then the question is, where does all that data live that those applications need to interact with, and where do the rules that govern that data live to make sure that it's also safe for these agents to do stuff to it? Where does the authentication, permission, and auditing layer live, and so on?

I don't think it's as binary as that.

I think you can retain Salesforce as your system of record—

Yeah.

—but then also build applications around it and use other SaaS providers around it. You said before, “Generic SaaS tools will increasingly be replaced by custom-developed internal apps.” I’d love to debate this one with you because if so, I’m going to be out of a job. Why do you think that?

I think it’s a balance that will really start tipping, right? Today, if you’re a company and you have a specific business need, hiring a development team to build, deploy, and operate an internal application just for you to solve that business need, the math of that very rarely makes sense. In the build-versus-buy equation, the buy equation almost always makes sense.

That balance is going to start really shifting toward building being much more of a viable option. And that obviously—

But that’s not a generic SaaS tool. That’s a specific tool that a specific company needs for a specific workflow that they have. I agree with you that dev shops that build specialized custom solutions will be replaced by your Bolts and Lovables of the world, but you said generic SaaS tools, and we have companies like Klarna saying, “Oh, we’re replacing all of our SaaS with AI.” They’re not. They’re absolutely not.

No.

And they won’t. Why do you think they will?

Again, I think it will be more and more common to start doing that. If you think of generic SaaS tools, they are also solutions to a specific business problem that businesses have. They’ll just say, “Okay, this problem looks very similar across tens of thousands of businesses, so we’ll build the generic version that works across all of those businesses.”

They don’t each have to build their own, and that comes with a set of trade-offs. You get some benefits from getting a generic tool that encodes some best practices and so on, but you typically also have to adapt how you work as a business to the tool you just bought.

With a lot of the SaaS tools that I’ve bought over time while building up Netlify, it’s been, “Oh, we’re kind of doing things this way now, but this tool says we should do things this way, so we’re going to start adapting to the tool because we can’t afford to just build our own stuff.”

That’s where I mean that more and more things that are right now in the balance, where you would buy a generic tool that works for lots of companies, will become viable solutions to say, “Let’s just build our own. Let’s just solve our specific problem instead of leaning into a generic solution for tens of thousands of companies.”

You’ve sold to large enterprises before.

Yeah.

They move slowly.

Yeah, yeah.

They’re not always hugely educated. Respectfully, I think it is a very small customer segment that will be doing that in the next 15 years.

No, honestly, I probably agreed with that a year ago. Then I started seeing just how fast tools like Bolt and Lovable and so on are starting to change things at the other end of indie startups, everything like that. Now my timeline is more like 5 years than 15 years before it’s going to be really relevant.

It’ll start by tapping into areas like everywhere where you have no-code tools right now, like Retool and stuff like that. Starting to have agents that use code together with people to build there will start giving much more flexibility. You’ll be able to build much more.

As that evolves, we’ll start seeing that spread into more and more areas. I think internal tooling teams will start solving some of these problems, with some budget gradually shifting from the SaaS spend budget into either an internal tooling team budget or an external providers’ custom-built-for-you budget.

How do you think about maintenance? The joys of having independent providers is that they’re maintained and updated—

Yeah.

—continuously—

Yeah.

—by great teams. When you have 200 tools across a large organization, you’re not going to do maintenance and security updates on all of them. How do you think about that?

3. Zero Operations Become Essential

That’s something I think a lot about because this whole Jamstack architecture that I proposed back in the beginning of Netlify’s journey starts becoming really important for this as well.

If you start going from buying 200 SaaS tools to running 200 internal specialized applications—or maybe 400 internal specialized applications—you have to do that on a platform that is zero-operations, first of all. That means you start having this new stack where people like you can make a lot of money by investing in the right places. It’s all of the components that go into that kind of zero-operations and serverless stack of building blocks.

You definitely don’t want a code-generation agent that you’re using to solve one of these problems to give you a massive Kubernetes cluster and be like, “Good luck.” You’ll want it to deploy to platforms that are zero-operations and serverless, with security built in and so on.

You’ll want it to use serverless API services and databases that are all managed with their own security teams and operations teams as the Legos it builds on top of and composes together. Because otherwise, you’re right: if it’s just fully custom-built Kubernetes cluster stuff, all of it, the viability of that whole approach goes away. The maintenance cost just overtakes everything.

Then you could speculate: can the agents maintain it? But I think we’re definitely in a 15- to 20-year horizon before they could be relied on for anything like that.

If the agents build on top of a stack where your deployment layer is fully zero-operations, serverless, and so on, where you’re using systems for hit list content or using Salesforce as your system of record for a bunch of stuff, you’re using managed databases, whether it becomes Superbase or a NeonDB or some of the big cloud providers that win out in that layer, the agent uses these building blocks that are, in themselves, maintained by someone.

Then you start being able to take on that level of building without exploding in terms of maintenance costs.

When you look at the different stages, we mentioned Lovable and being able to create very easily. Prototyping is now step 1: have an idea?

Yeah.

Do design tools like Figma die in the next wave of building?

4. Prototyping Moves Into Building

They’re definitely going to have to think a lot about this full stream of how prototyping is often going to shift from the design tool into building something. I still think that, for a certain level, as we talked about earlier, as long as brand matters and things like that still matter, it will also matter to have teams that work with tooling operating at the visual level and the interface-design level.

But if Figma doesn’t manage to connect that to the full end-to-end chain of prototyping and building and so on, then this could be a pretty existential threat to companies like that over time, or at least a threat to their current revenue model.

There’s no doubt that the way people start projects in general is already starting to shift a lot, both from traditional templates, bootstraps, and front-end frameworks on the one end, and from design prototypes and visual design on the other end, toward prompting as one of the key ways to start a new project.

You mentioned prompting. Do you think chat is the right interface for this next wave of AI? I’ve had many guests on the show before say the worst thing with ChatGPT is that it made chat and prompting the default interface. Do you agree?

Yes. I don’t think chat is the be-all and end-all of interfaces. It’s a very simple transactional flow that’s easy to work with, and I think chat or conversational interfaces will be a thing. It’s also a thing between humans: we all use Slack, Messenger, WhatsApp, and all of these chat tools for good reasons. It’s not like that will go away.

But I think there’s a whole other set of user experiences and user interactions that we have in the non-AI-powered world that we haven’t really figured out how to do in the AI-powered world.

One of the things I’ve been talking a lot about is this path where, right now, when we look at all this custom development and all of these apps like Bold and Lovable, they’re generally making it a lot faster to build exactly the same things on the web that we built over the last 10 years: the same kind of applications, the same kind of marketing sites, and the same kind of user experiences that we built over the last 10 years. We’re just driving down the price of building that.

But the next frontier for me is figuring out what we can build in terms of actual user experiences when we have LLMs and AI available that we couldn’t build 2–3 years ago. I think there are a lot of interactive user experiences that we can now build that are definitely not just a chat prompt saying, “Type what you want to see or do,” or, “Talk to me.” That couldn’t be done if we didn’t have the ability to create all kinds of assets automatically, from 3D modeling to sound, music, visuals, and so on.

That couldn’t be done at a time when the core constraint of all user interfaces was that computers are transactional, deterministic machines. You click a button, and the UI does something that you know that button will do every time. You type a command, and the computer does something predictable. Generative AI kind of removed that underlying constraint, and we haven’t really started building new kinds of UIs that take advantage of that.

I really agree that chat is not it. Chat is part of it, but if I follow a link and land on an e-commerce site from a large brand that’s trying to really differentiate itself and be something different, I definitely wouldn’t want just some box saying, “You come up with what you want me to do.” I want those online experiences to guide me, to create an experience for me, and to take me somewhere. That’s never going to happen through chat-style prompting.

Why? Do you not go into a shop and ask for what you want?

Yeah, but think about how much a fancy brand spends on that shop, on the display windows, and on people who don’t ask first—on everything that’s supposed to draw you in to the point where you’re ready to ask, “Can I have a look at this thing?” That’s a very expensive industry, to make all of that happen.

I think the more the world goes digital, if you want to build any kind of brand affinity and storytelling, you also have to find ways to do that. Today, honestly, most e-commerce sites have been stagnating in terms of development in that field for the last 3 years, and everything looks pretty similar, with some changes.

But I think when the cost of building an e-commerce site that looks like any of the current brands today goes down to just asking Bold to do it for you or asking Lovable to do it for you, those brands are going to have to come up with something much more engaging if they want to keep standing out and keep building relationships with people as more than just a one-click shopping experience among any other.

You’ve mentioned Bold and Lovable. When we think about these generative AI services that are scaling revenues like we’ve never seen before, I don’t think—

It’s wild.

Wild. Everyone says these generative AI tools are sugar-high revenues. You can’t rely on them. How do you distinguish between sugar-high revenues that lack sustainability and meaningful, durable revenues that can be relied upon?

5. Agent Tools Create New Developers

When I look at tools like Bold and Lovable, the category itself has something where it seems obvious that there will be a lot of money spent there. When you see some of the examples of what people can do with those tools that they just couldn’t do before, it seems obvious that it’s worth something. You see how fast someone can build a typical website, app, or dashboard.

It’s so funny. This is the promise of Webflow and Squarespace for years: anyone can build it. The truth is, Shopify, Webflow, and Squarespace—you and I both know—my mother—

Yeah.

—can’t build on them.

Yeah.

It’s still too tough.

No.

She can use Lovable.

Yeah. But even then, I don’t know. I’ve liked this example of a UX designer from Niantic who is building these things on Bold that just seems mind-blowing. He built this whole 3D object editor, an actual tool where you can build 3D objects in the browser with AI-based prompting and everything running in the browser. He built that entirely with Bold just by prompting, and that’s so mind-blowing.

But it’s also an example where I’m sure your mom probably couldn’t do that with Bolt, because I don’t think I could do that. I’m pretty sure I couldn’t do that with Bolt. That takes real skill, having built the skill set to work with these tools.

I’m sure we’ll see the same thing: not everybody is going to use these tools to make stuff themselves in the end. Even if you can potentially, as a small restaurant owner or whatever, go to Squarespace today and build the website for your restaurant, you’re probably busy enough that what Squarespace will do for you is make sure that the cost of getting a freelancer to build you a website is much lower than it would be if you had to get that freelancer to program the whole site from scratch.

I think we’ll see the same with tools like Bold and Lovable. I think we’ll see the next 100 million software developers emerge around tools that are agentic, but I still think it will be a profession to use those tools to build things for other people. There will still be a skill set involved in knowing what the agent should actually build for someone, what works, how to get there fastest, and how to push the limits of these tools.

But it will lower the cost of each project built with those tools dramatically from what it would cost today.

So there are developers using Cursor, there are developers using Lovable and Bolt, and then there are non-developers using your Lovables and your Bolts. When you look across that stack, how many do you think remain among the developers using Cursor and your alternatives?

Because I’ve been in this space of building tools for web developers for way too long, I’ve seen the same discussion forever. Early on, when I was CTO of a company in Spain, we were mass-producing websites for small to medium-sized businesses in Europe. I was a CTO building the whole platform that our freelance designers would use to do that work.

Tools like Squarespace, Weebly, and so on started coming on the scene, and everybody was saying, “That’s going to be the end of all web development. People are just going to use those tools.” That didn’t happen. It just lowered the barrier to entry. It meant that every little corner store could have a website, but it also pushed up what you needed to show up with if you were somewhat more professional.

Then I started this whole Jamstack movement around decoupling the actual web front end from the back end and starting to build at that level. In the years since—probably 2016, 2017, and 2018—we started seeing all these boot camps spitting out React developers. There was this sentiment that they weren’t real developers and didn’t know the whole stack, but it turned out that they were really valuable and that working at that layer of the stack was important.

Now we might have an additional 14 million developers in the world from that layer. But the interesting thing is that the layer below, made up of Ruby on Rails, Java, and .NET developers, is probably still at least as large as it was before that. The new layer just got added on top.

My prediction is that something pretty similar will happen. We’ll probably add a new— in this case, I think more like 100 million—software developers who mainly operate at this very agentic stack. Some of them will go a couple of layers deeper, just like now. You’ll have some who are more full-stack than others and who go a few layers below that level of the stack.

We’ll have that happen, and then you’ll probably have more or less the same number of developers who fully understand the layers below and are working on them. Those are probably the ones who are going to be thinking a lot about the agent experience of what they’re building, but they’re still going to be really necessary.

I totally get you. When you look forward to that future, open versus closed is a very big question. How do you think about which approach dominates in an agent-first world, and why?

6. Open Platforms Shape The Agent Economy

I think every time there is a platform change, we see this battle between open and closed. This is another kind of platform change. I think open will win again, in part because the ones that were best positioned initially to become fully vertically locked-down platforms were really the large language model makers.

There was a moment when it really looked like we would just have a very small number of those that would be incredibly important and would be able to build a whole stack around them, and that would be it.

And then suddenly, model training costs went down, open models came around, and DeepSeek came around and so on. Now it seems a bit the opposite: there will be a lot more to do to build agents that can swap out whatever model is best, combine models together, and so on. It will actually be an advantage for agent builders not to be tied to a specific model and to be able to jump around between whichever model is best. I think once that starts happening, we're going to have a much broader, competing ecosystem of agents. That's going to mean the benefits of an open model, where people can bring their own agent, will win out over a closed model, where everything is fully verticalized.

I'm not sure. I think Google is the most underestimated because the endpoints they have with consumers are so broad, and they have so many consumer touch points. And then they also—

Yeah.

—own the chip layer and their own compute, and it's a very strong walled garden if they keep it within all of their G Suite products.

Yeah.

And so I'm not sure. Then you look at Microsoft, its suite, and Copilot—

Yeah.

—and everything that they can build within enterprises, and the security provisions within enterprises being nontrivial, to say the least. I'm not so sure. I think closed has so much value. Again, am I wrong?

I don't think you're wrong in the sense that, as I said, the web really managed to fight back, and it's fully relevant today. There are still a lot of closed, walled-off gardens, right? I think we'll see the same.

But when I look at where we are right now, if I look at Google's products, I now have all these Genesis buttons all over, and I never use them. They just haven't managed to productize it in a way that feels super compelling compared to just going—

That's really interesting. Is that not just a Google product strategy? I would say that Facebook is the same. Llama and the productization around Llama—

Yeah, yeah.

—has been, bluntly, very, very poor. At the moment—

Totally, totally.

—it's just an integration into WhatsApp.

And that's what I mean. Most of the tools I use that are agentic sit outside any of those walled gardens. We're starting to see things like, for example, a Chrome extension to use ChatGPT within Google Docs instead of Gemini, just because people want it. They say, "Hey, I use ChatGPT all the time. It has my history. It knows me now, and so on. I want to bring it there."

You start seeing those things emerge as pressures to do stuff with these products, with other agents, because the vertical strategy, at least right now, is producing more bland and worse results than the myriad of companies building agentic experiences. So right now, my signal is that the open world will probably create more compelling innovation that people will want to bring into the product they otherwise use. If they really can't, they might start using other products.

Do you think we'll see a specialization of LLMs? You see Anthropic really excel in developer tooling in particular and code, but honestly, OpenAI has just wiped the floor with them on consumer touch points and consumer brand.

Yeah.

I think you just answered it in a way. We're already starting to see models get a brand for being good at different things. When you start building an affinity for a particular task, you think, "Okay, if I'm doing this thing, I'm probably going to go to this type of model, and this company seems to really serve me there."

Again, it's so early that who knows? But the sign right now is that we will start seeing large language models have different strengths and weaknesses, and you will go to some of them for specific use cases. Anthropic has definitely stood out in its ability to solve programming and coding-style tasks in a way that I think is very bullish for Anthropic, because that's such a groundbreaking part of what makes LLMs multiply what they can do by a lot. The better they are at programming other computers to do stuff, the further the reach of what they can actually do.

You can invest in Anthropic at $60 billion, OpenAI at $300 billion, or Groq at $50 billion. Which would you buy and which would you sell, and why?

I'd probably go for Anthropic. It's a really interesting thing from the outside: a lot of their motivation started around safety concerns and how to make sure the models don't take over the world. After an initial panic, it now seems pretty clear that LLMs are probably not going to escape and take over the world.

I think it turned out that a lot of the mechanisms you needed to build to make sure the model's output was safe also apply to making sure that the model's output is usable for a practical purpose. That's where Anthropic has really built strength: in actually being able to guide its models to do things that work in production, write the right code, not break the syntax, and so on. If they can keep doing that, it makes their models really applicable in the real world.

So would you sell OpenAI or Grok, then?

I'd probably sell Groq. I still think it sits a little more in a middle space where, as you said, OpenAI, on the other hand, has really captured the market. When you think about going and asking an AI for something as a consumer, it is still the brand you go to. It's the first one you think of, and I think that carries very real value.

It's very hard to build that with the scale of users they have and with the ingrainedness of ChatGPT in the language as what a normal person would talk about if they talked about going to talk to an AI.

Speaking of carrying value and going full circle, we spoke a lot about the agent experience. How does value accrue in an agent economy? If an agent builder has to pay you, Bolt or Lovable, and OpenAI, how much margin is left?

Every layer will shift spend from somewhere else. That whole layer of Bolt and Lovable, as well as Cursor, Windsurf, and all of these tools, will shift budget that would normally go into more hours of developer time to build something. There's a lot of that budget that can shift into building more things instead.

Tools like ours will shift budget from maintenance, operations, DevOps, and the actual cost of running a thing after you build it. There's a lot of budget there to shift into paying for a platform that works. If that's true, you'll still have roughly the same amount of budget available for anyone buying your thing or getting value out of it. The pieces of budget that will move are the hours you would previously have had to pay developers to build something.

Do you think we'll see companies be willing to shift labor spend to agent spend? What I mean by that is, you're seeing like, uh, trucker calling, you know, calling up brokers for truckers, Happy Trucker or whatever it's called, the Andreessen one. Are we willing to move labor costs to agent spend costs?

I think so. Companies will be willing to do that, and it will be disruptive in some areas. The first area we've started to see it in is probably something like call center workers, where obviously you don't see companies saying, "We have to keep our call centers as staffed as they are today." They do say, "If we can shift call center hours onto just LLMs, that's going to be more predictable and more—"

And we don't see per-seat pricing in those use cases. We see resolution pricing—

Yeah.

—which I think is probably much more reasonable.

Precisely. We're starting to see real changes in those kinds of pricing models. That's a very concrete example of a willingness to shift around labor. Part of my theory is that a lot of the labor will again shift into building agentic digital workflows, products, applications, and so on. That brings me back to my thesis of at least another 100 million new developers.

I think Stripe will be massively more valuable—

Yeah.

—in a world that we now enter—

Yeah.

—because I think there will be so many people who are enabled to create stores and entrepreneurial endeavors, and who will inevitably use Stripe as their default payments mechanism—

Totally.

—that would never have been entrepreneurs making money on the internet. The GDP of the internet will increase seismically as a result of that.

I think there's a ton of opportunity in that space. Stripe, as you mentioned, is one of these—

It has all of those characteristics I mentioned. It’s zero operations, and it can be a building block in any custom-built system. It’s a little more abstract than any of the end-to-end e-commerce shops that you buy ready out of the box. It’s just a building block for all the underlying transactions, and it’s zero operations. You really don’t have to operate it or secure it if an agent builds some of it.

All of those sets of tools have huge potential to be really valuable as we start custom-building more and more solutions. Anything that sits in the space where you have predictable business processes that are still pretty manual is just an area where we’ll accelerate how much of those can be programmed away.

Final one, and then we’ll do a quick fire. What behavior or action in the wave of AI that we’ve seen has been most surprising to you?

I think it was probably the panic over AI safety when ChatGPT came around. It was like, “Oh, they’re immediately going to kill us all, take control of nuclear missiles, out-compete humans, and so on.” When you actually go and look at these LLMs, they’re very, very far from any kind of internal agency outside of, “Give me a text that says this, and I’ll try to predict it.”

That was probably the most surprising to me: the extreme panic around that for a moment, with all kinds of attempts at really strict regulation way too early.

It doesn’t even do a Twitter thread. It’s not exactly going to do a global invasion.

No, exactly. Exactly.

I feel safe for the time being. Listen, Matt, I want to do a quick-fire round. I say a short statement, and you give me your immediate thoughts.

Yeah.

What do you believe that most people around you disbelieve?

One thing is probably the sense that we’ll still actually be building UIs, applications, websites, and so on. There are a lot of people who think that once we have agents, everything will be disintermediated, and the agents will just do stuff with each other. But I believe that as humans, as long as those agents are actually doing things because humans are trying to do things, we need these shared experiences that we can pass around, link to, and understand.

We need to be able to say, “Hey, I’m seeing the same as you’re seeing,” and so on. That’s probably one thing where I sometimes see a lot of people thinking that all of that is going to go completely away, and I don’t think so at all.

You can buy and hold one public stock for the next 10 years. Which do you buy and hold?

It’s a hard one, but I would probably say Microsoft. If you can only buy one, you probably want it to be a pretty safe one. You want it to be one that benefits from all of the disruption AI is going to bring to the field rather than falling because of it.

Of all of those, I think Satya is the strongest CEO in that space, and I think he’ll make great decisions.

What happens to WordPress?

I think WordPress is in a challenging position where it’s kind of at the end of its innovation cycle, and I don’t think it’s in a great position to respond to all of this disruption. It’ll be seen more and more as a bit of legacy software.

What happens in that situation? Might private equity buy it and eke it out for cash?

Potentially, yeah. It depends. Maybe there’s enough community and pressure that it just keeps sticking around for a long time. But I have a hard time seeing the venture-scale acceleration of it ever happening. So either it kind of ossifies, or it’s got to have some kind of PE exit.

What have you changed your mind on over the last 12 months?

One big piece has just been the timeline for when companies would start writing real software entirely through agents. For me, the combo of Bold and Lovable coming out was really significant. I thought it was going to be 10 or 15 years, and now it’s more like the next 5 years are going to be really interesting because they pushed the frontier so fast.

What was the hardest fundraise for Netlify? Why that one?

It was the first one. Our seed round was lots of rejections.

How many meetings did you have?

I don’t know, 100 or something.

100.

I don’t really know, because it was all the angels and individuals that then led to funds and so on. Then there were all the funds, with different meetings and several meetings with each fund. It was a long process.

How much was your first fundraise, and what was the price?

It was just over $2 million—$2.135 million at an $8 million post-money valuation.

What’s the latest price for Netlify?

The last round was a $2 billion valuation, so we’ve come a long way. Back then, one of the key questions was actually, “Why the web?” Why were we building this whole thing for the web when everything was obviously mobile and social media?

When did you raise it at $2 billion?

The end of 2021.

Was that a really toppy round?

At that time, the climate was also very different than it is both now and in 2016.

Do you regret raising at such a high price, or was it actually a joy because you’ve got a lot on the balance sheet and you’re in a really comfortable position?

I would say a mixture of the two. I think everything was priced too high there, and it probably wasn’t entirely healthy. On the other hand, as a founder, it’s pretty good to take advantage of that and get cash into the business before hard times came.

As a founder, you should always raise at the best price you can and capitalize the company as much as you can because you want to be able to weather harder periods.

What’s your biggest short in the public markets? We mentioned Salesforce earlier, which is a common one with a lot of people. If you were taking them aside, who would be yours?

I’m so bad at thinking in that sense—betting against things rather than betting on what’s going to be really big. If I had to pick one, maybe I would just pick something like ExxonMobil, because if they go down, it’s probably a good thing.

Final one for you. When you look forward 10 years at the most exciting things that happen, what excites you most in the next 10 years of AI?

The capabilities it can give humans for creativity and for building amazing shared experiences. What it can do for builders seems really amazing. I’ve always collected examples of people who are starting to build things as one person or something like that, and it’s just like, wow.

There’s this illustrator who’s making these TikTok sci-fi epics with AI-generated video. It’s just like, wow, one person can sit and come up with something like that. Or, as I mentioned, this UX designer from Niantic who can build a full 3D editor in the browser just by prompting.

I keep thinking that if we do things right, make sure that open platforms remain really accessible and readily available, and invest in tooling for people to build stuff, own it themselves, and be creative, AI has so many exciting abilities to let humans be more creative and do more interesting things. That’s what excites me the most. That’s why I built tools for builders in the first place.

Matt, listen, I so appreciate you joining me today. I so appreciate the directness. I apologize for my debating at times. I hope SaaS continues in a meaningful way, but I so appreciate your time today, and thank you for joining me.

Thank you so much. This was a lot of fun.