Agent Experience: Building an Open Web for the AI Era
40:55 Share

One of the things I've started thinking a lot about is this concept of AX or agent experience. I wrote an article about that and have centered a lot of Netlify's efforts around how do we build the best AX.

And by AX, I mean back in the 90s, Don Norman coined the term UX or user experience, the holistic experience of a user of your product, not just the user interface, but like how do they interact, right? And as I said, when Netlify started, we built our whole philosophy around a focus on DX. What's the holistic developer experience around building with this architecture for the web?

We're starting to see agents play a major role in building the web, writing a lot of the code, coming up with the assets, doing all of that, collaborating with humans, but multiplying this factor of what you can build. So we're starting to think a lot about the agent experience of our platform and the agent experience of the web.

Hi everyone and thanks for listening to the A16C AI podcast. As you might know, there is, if not a consensus, then at least a strong undercurrent of feeling that the web is not as exciting of a place as it once was. Blame might lie in any number of places, including walled gardens, cookie cutter designs, and a level of complexity that keeps newcomers from building interesting modern sites. And while generative AI has proven incredibly exciting for many reasons, it has many of these web experience critics fearful that AI generated dreck will soon overtake the web completely.

However, as Netlify co-founder and CEO Matt Billman explains in this discussion with A16Z General Partner, Martin Casado, AI agents could actually have a profoundly beneficial impact on the future web. As Matt sees it, a new set of standards around what he calls agent experience

could open up entirely new capabilities for websites and design and bring millions of new web developers online. This doesn't mean just more people building the same old sites faster with new tooling. Today, for example, Netlify says it's seeing 10,000 sites a day launched using AI-generated code. But rather, they'll be incorporating entirely new elements around 3D assets, textures, personalized experiences, and other things we can't yet predict.

You'll hear Matt and Martine dive deep into all of that, as well as, among other things, what new developer primitives might emerge after these disclosures.

As a reminder, please note that the content here is for informational purposes only, should not be taken as legal, business, tax, or investment advice, or be used to evaluate any investment or security, and is not directed at any investors or potential investors in any A16Z fund. For more details, please see a16z.com slash disclosures.

Matt, it's great to have you here. Thanks for coming and joining. Very excited to talk about AI when it comes to the web and what you're thinking about there. So maybe for some background, just give a quick overview of yourself, and then we'll jump into the company and then what you're saying. Sounds great. I'm Matt Billman, CEO and co-founder of Netlify, and have been building tools for developers and specifically for web developers for a couple of decades now. Netlify being the one I've worked on the longest, and that's today serving 6 million developers reaching...

billions of people every month. Yeah, probably worth going through because at the time when Netlify came out, it was actually a pretty radical departure from how we'd done web development and now it is the standard. So in many ways, you were this pioneer of the web. So maybe talk through what Netlify does and what it introduced. It is very cool. Netlify helps developers go from code in their local code editor...

to something running on a URL on the web in production, reaching as many users as you can possibly reach with minimal friction. The original disruption or invention around that was this insight that when I started Netlify, the way people would architect for the web

would be large monolithic applications where the data access layer and the business logic and the backend logic and the front end was all sort of part of one application bundled together.

And I started really building this belief that on the backend side of things, we would start splitting that monolith into different APIs and services where some are your own, running in your own data centers, and some are other people's services like Stripe or Contentful or Algolia, all these APIs.

And then the role of the web developer would really change to actually build the actual web UI with web technologies and so on as its own self-standing application decoupled from all that backend infrastructure stuff. Saw these first sort of headless services pop up and could see that no one was really thinking about, okay, so what does that mean for the head? Like, how are we going to build the head to all of these? And that really became the start of Netlify figuring out what's the right end-to-end platform

to build the head for all of these web experiences. And back then, we kind of had to coin a term around it, JAMstack, just to start having even a nomenclature to talk about this is like a different way of architecting, right? What does JAMstack stand for? So it stood for JavaScript API and Markup.

This sort of really centering on this idea that the stack had moved up a level, right? And suddenly the actual stack was what we would deliver to a browser. And the developers working in that world would just be working natively in that stack. You know, it's funny. It's because when you first came, talking about the Jamstack, it was actually a very radical idea. You know, it's funny. At the time even you were hedging, you're like...

Maybe more dynamic sites you won't use this for, but for the more static sites, you're going to implement it all in JavaScript. You're going to use third-party APIs. Even then, I don't think you were the pioneer, but you didn't have a sense of how disruptive. I will tell you now as an investor, if a company comes in the door and they're building a product, it's almost certainly going to be

you know, built natively for the web using JavaScript and TypeScript. And like the majority of functionality is like some third-party API. So in many ways we've gotten. Yeah, it's been pretty amazing to see. And in that way, the term Jamstack itself has almost lost its meaning. I was going to ask you, do people even use the term Jamstack anymore? I mean, it's funny, right? I think it's almost shifted to now analysts and enterprises start talking about it. Yeah, that's right. Yeah, the un-

But the developers are like, wait, isn't this just how you build stuff, right? And it's always forgotten that it was like a shift. And I think the other thing that was key for us that'll relate to what we're thinking about now in the AI space was then seeing, okay, if we believe in that architecture as the right approach, and we believe in the web and the open web as like something we really want to build around and make stronger as a platform,

The key to that will be the developer experience that we as a company can deliver to building with that architecture for that platform. So the first...

massive focus of Netlify was really that developer experience, right? What is it that makes this architecture and this web platform really amazing for developers to build for? How do we take out the friction and onboarding and understanding and scaling and operating and all of those pieces? That was really the driving force for Netlify. And I think a big part in shifting the whole architecture of the web in this direction and sort of

reigniting to the web also because it was also in that sense a very different world. Now you say all the startups are building on the web. Back then when we were first raising our seed round for Netlify, the most common question from investors was why the web? Yeah, totally. No, I remember. Isn't it all just mobile apps or social media? Isn't it just going away? And we thought, okay, if we can make the developer experience here great enough,

The web as a platform will win. Yeah, I actually went through this when I went through our old investment memo in Netlify. And it was like, basically, this is like long developers. So you believe like developers are like key and long web. But this basically means web. And both have turned out to be true. So to go back to that period. So you come in, you're like, listen, you know, more and more like web development is development. And so more and more product developers are web developers.

JavaScript is like the primary language, which is the J and Jamstack. And now going forward, like you've got every aspect of those old monoliths is like a separate company as an API. You've got like clerk for auth and resend for email sending and a bunch of backend services, say Supabase or Convict. There's a whole ecosystem that's played out. And a lot of that was because you had seen this shift where the web developers were kind of taking over, right? That's kind of the shift. So we're here to actually talk about a different shift. Yep.

Which is AI. But again, not like AI in a general sense, AI in a specific workload that needs to use the web and interact with the web. And so I know you've been talking a lot about this. So maybe can you just talk about your thesis about AI, how it interacts with the web, and then how, again, the web platform has to evolve to meet that? Yeah.

I've been working around the general thesis of the market where obviously like in our space of the web, AI is going to be the main disrupting factor and change everything in a lot of different ways. Some that are really hard to predict and then some that are really easy to predict. And probably the first one that's really easy to predict is just acceleration. We will have more code written faster, more assets, more content. No one will say AI will lead to slower code or slower code, right? That's just really obvious. So we have to

lean even more into systems like ours that allows really high throughput, right? Because you just get more volume going through faster. And this is just because developers are using tools to write more code. Totally, right? Developers are using cool content authors working tool. Artists are using tools, right? Like all of that. Yeah, so the cost of creation, whether it's code or content, is converging on zero. So you get a lot more of that. So you get more. We see this in space, right? Totally, right? That seems like kind of...

obvious and undeniable pretty much. At this point it's undeniable. I really have never met anyone that said it will be slower. It will be faster. Then at first we used that to essentially build exactly the same

that we've built the last 10 years. So we use that to build more of the same kind of web experiences, websites, web apps that we've built in the last 10 years. It just lowers the barrier of entry to those. We get way more of them, but they're right now fundamentally the same. Next big element I think is kind of UI 2.0. When we start discovering what can we actually build on the web that's

that's fundamentally different, that you couldn't have built two years ago without this. And there we're just in the very early stages of figuring out what does an elevated production, if you really take all of this acceleration and don't put it into more, but put it into something new, what does that look like? So that will lead to a lot of experimentation and change, and especially in driving. Either it will just lead to complete flat-throne change and people move away to the web because we're not doing it good enough,

Or it will lead to massive innovation on the web. And how can we build really compelling experiences with this technology that takes people there? Anecdotally, other than, say, maybe like mid-journey and Discord, all of these are web technologies like ChatGPT. Totally, totally, right? Ideogram, like all of them. Cloud. I mean, it seems like that the first place that they land is a website and then maybe an API, which is basically a rest of web service. But there could be questions around, are we going to...

consume our customer sort of experiences there? Or is that fueling out to like a TikTok and like an Instagram and to different platforms that just use the web as like this content blend content repository that's not really like... I don't think that will happen. I think if we do the right things, that won't happen. We can actually allow people to build really compelling experiences. But I think it will require that we start building things on the web

that are different from the things we're building right now. And does that have a form? Is it stuff like richer, unstructured content and media? I would say that's the area I think it's important to predict. You can start predicting what elements can go into it, right? As a platform provider, I'm thinking, what are the primitives I need to...

enable to allow people to invent that rather than having to invent it. But obviously, much more imagine that I think we haven't started figuring out what it means when the cost of video production, 3D production, WebGL also starts going to zero, right? So do you think, just to be very explicit about this, because we're talking about abstracts, so do you think we end up with an actual 3D rendering layer and we actually have, instead of a DOM and text, it actually is...

images and videos and 2D objects. And this is like the metaverse. I hate the term we all do, but like, is that what, are you thinking that there's a possibility? I think now we can see, now I can start seeing how you could build that, right? Which I couldn't really like, again, two years ago, it would just be too expensive for a normal company to build anything with like,

real 3D assets and stuff like that, right? Like, now I can start seeing, like, in a couple of years, why couldn't a small web team build that? That's very, very, it's so funny. Until I sat down in this chair, I just haven't, you know, so in the past, this has just not really worked, right? Like, I mean, you had, you know, Flash, which is pretty good. It ended up being kind of some... Flash was there, right? Like, but had...

So this will actually lead to one of the main points of making, right? One of the things I've started thinking a lot about that will tie a little into this is this concept of AX or agent experience. I wrote an article about that and have centered a lot of Netlify's efforts around how do we build the best AX experience?

And by AX, I mean back in the '90s, Don Norman coined the term UX, user experience, the holistic experience of a user of your product, not just the user interface, but how do they interact? And as I said, when Netlify started, we built our whole philosophy around a focus on DX. What's the holistic developer experience around building with this architecture for the web? Now,

We're starting to see agents play a major role in building the web, writing a lot of the code, coming up with the assets, doing all of that, collaborating with humans, but multiplying this factor of what you can build. So we're starting to think a lot about the agent experience of our platform and the agent experience of the web. And that both ties into what are those agents going to build and how do we give them the right tools to make it easy to build really amazing experiences on the web.

But it also ties into sort of the part you talked about with Flash, because one of the things that was very rigid about Flash and that felt totally out of line with the web as a platform

was that it was like this closed up... It was a walled garden. It was like a walled garden, right? The web has always... You were in the web page or you were in Flash. You were in Flash, but you weren't in the two of them, right? And you couldn't really... The web has always had this concept of a user agent that used to be a very basic agent, right? It was just your browser, right? There was always this idea that you could approach the same website with many different agents, whether it was like your normal browser or screen reader or like...

crawler from a search bot and so on. When we sort of pulled out all the flesh and said, okay, we need everything to be these open formats that computers can understand and so on.

We did also pull back the level of control over the experiences we could build. As much as I hated Flashback then in many ways, there was also a real loss in the level of creativity we could actually build there. If we can build the right agent experience around the web, that will make it possible again to build really wild and creative experiences, but still allow any person interacting with them

that also have agents helping them to interact in whatever way they want, right? And scale down that interactivity or just give me the data, just give me the information or whatever, right? Or scale it up. I think that's one of the things that I think is sort of underplayed in where we could see the web going in that people still just think about those kind of like text interchange formats almost and not about also the web

covering a need for shared experiences. If your wife is buying something and she sends you a link to it, she wants you to see the same that she's seeing and she ideally wants that page to also convince you that you should buy this. And people sometimes forget that shared experience layer. And that's where I see the web having a really big opportunity, but it requires us to also rethink a bit like how can we actually deliver more compelling experiences on the web. It's kind of interesting, like there's two ideas here. One of them is something that

well, dropped on the AI thing. And there's a separate one, which is almost like a reimagined layer. And you see them together. And is there... Is it...

Do you see them together because this is a chance to reimagine these things? We've got better tools to do it? Is there something more fundamental about like this new layer that's AI specific? To lead up to that question, so an obvious question would be is AI agents are trained on the same user data. And so they know what we know. So why not just keep everything the same and then they'll be able to navigate it already. They can use websites already. They can use APIs already. So like, what do you think the impetus is to actually move it to something different?

Is that inherent or is it just an opportunity to do so? I think it really depends on what we want the world to be like and the web to be like in certain ways. There's some people that are talking about this as essentially sucking all creativity out of the web in a certain way, right? Like, okay, who even cares about how we build and design any of these experiences? It's just about giving the agent access to the information and then you can go ask ChessGPT on it and it will tell you what's there, right? And I think that's like a path where you then end up

with the experiences moving somewhere completely different, right? And you start potentially having these walled gardens, whether they're built with web technologies or not. I see what you're saying. I'm interacting with a model. The model has nothing to do with the web. It's totally separate. It's totally separate, right? So it's just like a system for agents, right? But I think there's also a different way of viewing this as a real opportunity for human creativity to do much more than we could do before. And that's when I look at a lot of these

CodeGen agents that are now a persona of ours at Netlify, right? Like in a building and deploying to Netlify together with users. I see that as like a massive

of creativity to build. Suddenly a lot of people, I think the next 100 million web developers are going to come from that motion of working with agents that do a lot of it. And I see it both in a more democratizing fashion where it just becomes easier to like, you want to build an app, just do it. And then I see for companies to then stand out from that and look like they've built something that's not just an app that whoever built in five minutes with Bolt, right? Like they're going to have to

start using this technology to create another level of experiences, right? And again, there I see some very AI specific things that we couldn't do two years ago. Some things that we couldn't do for cost reasons and some things we couldn't do because computers just couldn't do it, right? And the cost reasons are obvious, right? Like 3D modeling, art assets, textures, like visuals, all of that.

Then interaction paradigms, it's just like a computer couldn't do things just because you expressed intent. Computers were purely like all of our user interfaces so far have been purely transactional interfaces. I click a button, the computer does something transactional and deterministic. That constraint has basically gone away.

But I feel we've barely scratched the surface of what you can actually do with it. We know that marginal costs are going to zero for like creativity and stuff like that. But also AI allows you to do things that you couldn't before with computers like this creativity, right? I can create a story. I can create an emotional connection. And there really isn't good platform support

for that right now. I mean, there is platforms for that, but it's not native, right? I mean, like today when I use these things, I like, whatever, I spin up 3JS and it's like some library where I put things, but it's not like a native, you know, so like those primitives just haven't been part of like the ethos for the web. What you're saying is, A, you have a new type of

agent, which is going to be AI or an AI enabled developer. They have new capabilities. So this is both an opportunity to reimagine the way the web lurks, A, and then B, not only is it an opportunity, it's probably going to demand it. And if we don't do it, then that new consumption layer may end up in a walled garden. And so the right thing to do is to make sure the consumption layer doesn't end up in a walled garden and we do what the web is good at and we make sure that it's the center point for these things. That's a great point. Yeah.

That is... I see. That's like the greater play, right? And where in a similar way of when I oriented like all of the first sort of iteration of Netlify on DX, right? If we're really leaning into AX now, it's because I think that's the requirement to make that happen, right? Like we got to make...

the web great enough to build for, right? To make it win as a platform again. So you're kind of at the center of this, which is a lot of these new tools, news, Netlify, right? And so whether this is like text to make a webpage, text to like make a basic app, a lot of them are choosing Netlify to land on. So like you, I think you have some visibility that the rest of us don't. And so maybe can you talk a little bit about what types of companies are using Netlify for this? And then what makes them different that kind of gives you a bit of a visibility into this

hopefully better future for the web. It started intentionally with a strategy of like, what do we need to do to make Netlify the best deployment platform for AI, right? And got crystallized now in this thinking around AXS, like agent experience as the way to unlock that. And we started out by building when ChatTivity launched the GPT store

It's kind of a bit of a walled garden approach of some of this, right? But again, we saw the possibility there and build a GPT for deploying whatever you asked ChatGPT to make to Netlify with zero friction. So no log in anything, right? Like just deploy, the agent can deploy on your behalf. And if you like it, you can go claim it and add it to your Netlify account. And started experiencing there with just that flow, right? Like of how is the flow different?

When it's not me as a user going to Netlify and saying I am deploying something, but when it's the agent you interact with saying, I'm going to deploy something, here it is, do you like it? It starts changing where it originates and so on. And we started having to build some different APIs for that and some different flows to unlock that. And then we started seeing...

those being picked up by Devin from Cognition Lab was like an early integrator. I remember that, yeah. Yeah, and then latest, like there's been different ones of them, but by far like the most impactful have been Bolt.new that sort of really exploded into the market at the end of last quarter and which they set a new standard for how far you can go with like a

prompt-based agent to build apps and so on. And by now at Netlify, we're seeing about 10,000 sites a day.

created from AI code generators. Wow! Right? It's pretty wild. And do they look different than if a human being did it? It's just like faster by... Right now, as I said, I think they're just building the same things that human beings can build but faster, right? That's the phase we're in right now, right? And they're still not as good as teams as human beings, right? Like they typically need more work, but they are enabling a lot of people that could never build a custom application

Yeah. To build one. Do you think that people that are using these things are primarily technical or non-technical? It's really a broad spectrum. That's also really interesting. There's a large set of the people that were going to local tools before, that suddenly code has opened up for them. In a way that just wasn't before. Some of these you start seeing

becoming technical through this journey, right? So you see them start prompting Bolt to build something. They download the repository. They don't quite get it, but they can use Cursor to ask questions and start figuring it out. And you suddenly see new programmers being born out of this, right? And being creative. Or old programmers learning new language.

I literally learned JavaScript, TypeScript, and Cursor. I was just like, how do you do this? How do you do this? It's a great tutor. It's amazing. I will say one thing we see with professional developers is why would you start an app for Scratch if you

can get all the basic pieces in place just by prompting a bold or a windsurf or a cursor or something, right? Then why go through downloading some framework, NPX installing it, setting up everything yourself, right? That doesn't create much value. There we see these as replacing the boilerplates and the templates we've always had, right? And to some degree, even the frameworks. Do you think that this spells like a slowdown to frameworks? Like, is this the next thing? Because...

One way to think about frameworks, whatever your framework is, it reduces the barrier to get something working. Just something, whatever, but then you start to go on top. But AI also reduces the barrier to get something working. So you could argue that it just basically replaces the framework, but the human still needs to describe what it wants and build on top of it and so forth. There's been some interesting threads and posts from the team at Bricks.

around just how they changed their React codebase to be more LLM friendly. Right? No, I didn't know. Describe it. So like they moved away from using hooks, for example, because they said like hooks is a great abstraction, but it hides the complexity. And we found that if we use Relay,

in single-page component files, it becomes much easier for the LLM. Because you expose more of the semantics for the LLM to see? And the LLM can see the whole thing, and it becomes very easy to take it in context and do one-shot operations on it. Wow, that's very interesting. That was just interesting. Again, it's early in a lot of these days. So wait, does this mean for people that are implementing APIs and you've got an API posted, you actually want to have more of the semantics shown so the LLM will learn it in talks in the future?

in context rather than like... It's totally possible, right? Like I think it will change all of these things. Relay was such an interesting example because I think Relay has like very limited usage in the industry compared to like Apollo or all these other... What is Relay? Relay is like a specific GraphQL client that sort of really build around static analysis and being very smart of lifting your queries up to the highest instance and so on. The typical...

people I talked to that have used it. Once people have really used it, they're sort of like evangelists about it. Everybody needs to use this. It's the best way, right? And at the same time, it just has this like barrier to entry, right? Like where the developer experience means that it's just, that's never going to happen, right? Like normal teams are never going to adopt it.

But it's one of these interesting examples for me of this shift from developer experience to agent experience. It's suddenly it turns out that agents are really great at working with Relay, and it makes it easier for them to reason about the whole data access together with a component.

It's funny, Convex has a very similar thing to this, which is just like these LMs are very good at it just because of the way that it is. And they're another company that I can see is really investing into the agent experience and building actual like evil frameworks for agents. So to be very clear, when you say agent, this is probably a human being with an AI? I think. Is that the right model? Totally. Or do you make a distinction? Is it useful to make a distinction? I think it's useful to make the distinction that when you think about

An agent is an agent, right? Like it really just means a computer that acts with autonomy, right? And takes actions on its own, right? But I think agent experience almost always ends up being like, how is the experience

of that agent in helping a human do something. Again, there's like two versions of the world, right? There's sort of one version where they all just run away from us and we have this economy 2.0 where it's just like AIs interacting with each other and we can't really do anything about it, right? Or there is a world where all of these agents are doing things

on behalf of some human in the end, right? Because someone wants them to do something, right? And then it becomes, the full experience also becomes like, what is that human's experience of the agent doing their thing? Maybe I'll just be just a more explicit example.

on this question which is, if I just draw from my own experience. I use cursor, I do a lot of web dev for fun. So it's actually very meaningful that I am involved because whatever gets produced I have to understand and like stitch into the rest of the code. So I think somebody that's like designing AX in this context would want to know that I'm also involved. So there's the single question of,

can I make this useful for an LLM? That's one question. There's a second one is, can I make it useful for an LLM human pair? So I'm just curious from your mental model, do you assume...

there's a human working along with the agent in a way that is actually meaningful to your design or not. Independent of whether these things take off or don't take off, I actually think that there's still like a human. Totally. We're thinking very human-centric in this, right? Like there's pretty much always a human involved. And it also means that the output of the LLMs need to be understandable to humans, right? Like it can't be completely disconnected, right? But it will.

change what's easy for a human to work with. I think that's like back to the point where setting something that's just easy for a human because of human's characteristics of how we adopt it and pick it up might really change if the agent can do all the basics for you and set it up right for you and do all the things that were complex.

And then for you to dive in, right? Like now it's like a really well functioning system and you get it. In that way, the expertise of the agent is very different than the expertise in a human. And sometimes it can act as that guide to the human that makes some of the things that used to fail in the market, not because they were worse technical designs, but because they were harder to adopt.

If the agent suddenly helps adopt them, then the benefits to the design suddenly kicks in, right? So, you know, you as Netlifier are seeing a lot more tools using AI, and you think that the web has to shift as a platform to be better suited to these workloads so that they can both create kind of new experiences, but also it doesn't get...

pulled into a separate new walled gardener emergent platform. And so you call that AX. And then can you maybe talk a bit practically? Are you like just posing the problem like AX is a problem or do you actually have a solution? Yeah. So I think I'm proposing AX as a discipline in the similar way as DX or UX as a discipline, right? Once it starts being an intentional focus of yours,

and you start really observing how the humans and the agent work together with your product, then you start finding solutions. Then there's a lot of things that becomes obvious pretty early that are concrete, all flows and authentication flows and just all of those patterns of like how can an agent actually take actions on the web, on your behalf, in your products. Those becomes like one problem set from developer tools specifically. There's all these like how do our primitives work

map to what an LLM is really good at versus just what a human is good at, right? How do our documentations map to that? For the web, there are questions of how can we simplify the way agents interact with the web, where right now the state of the art is like a headless browser and taking screenshots and having a multimodal LLM take action on that, right? That's probably not

the best way forever, right? So how can we simplify that? I'm proposing it as like a discipline similar to UX and DX, right? And we've already started doing it, right? So we know some of the core things

that you need to start doing, but I think it's also still so early in even the web. We also have launched the resource, agentexperience.ax, to sort of start cross-industry collaboration because there's almost nothing in terms of like on the web, like how do we define what an agent can do on a certain domain? Like how do we, like that's all, we all need to build that together as a set of standards, basically. And so if you're working on, let's say, a web project

related company to date, like an infrastructure, like I run the infrastructure fund. So like an infrastructure company that provides say an API for a web service. So is your view that the primitives have to change and everything changes in a review that like the existing infrastructure,

The existing components need to think about AX, but they can just be modified to think about AX. Because DevEx didn't change the primitives. It was just a way to think about the consumption layer. So is this a way to think about the consumption layer for existing tools? If you have a company today that sells to the JavaScript community that is core infrastructure, what would you guide these founders and project managers and engineers to do to take advantage of AX or prepare for this future? I will say I think that there's always an interlock where...

the primitives and the experience change together. So if you take the existing infrastructure, something like Netlify, right? Like part of the primitives we made

available through this was, for example, global CDN hosting. Before that, a CDN was something you put in front of something, and suddenly we said, well, if we want to make the DX around just building decoupled front-end easy, we should be able to just put them very close to the user. We needed to build new primitives around that. Serverless became a primitive driven by the same need to, if you're really just

just writing that front end backend for the front end layer. You want something that's like fully stateless, non-traditional and so on, right? So AX will probably be similar in the way that yes, it's not like the core underlying

You'll still need an old TP-Navigation. You still need a good old HTTP connection, and you need storage and compute and all of that. Networks. Right, like networks and so on. But you don't think as you get closer to the app, you're going to have new primitives emerge. Yeah, as closer you get to the app layer, the more you might have new primitives emerge. Because that wasn't so obvious for DevEx. Normally, it was like the products won because...

the product consumption layer was more... Like, I guess Netlify's kind of...

Think about something like containers as a primitive, also really started from a DX perspective of, hey, it's suddenly- Yeah, the portability is- Yeah, yeah. It's a good point. Suddenly portability, suddenly you can run something locally, and then we came up with Kubernetes to distribute them and so on, right? Yeah, right. But namespaces in the Linux kernel had been around for a long time. Totally, totally, right. It's almost like you take the old primitives and then you evolve it for the new thing, and you have new- Yeah, okay, that's a great point. Yeah, I like that. So you think that we're going to see a similar shift

I think so. For example, one example, right? Maybe if you had asked me two years ago about serverless databases, right? I would tell you that I don't really think... I've looked a lot at it and I think it's kind of like...

scale to zero for a database just doesn't matter that much. Once people have data in it, they never want it to scale to zero. Like, why would you do that? You want it to scale up, but you never really scale it down in reality. But suddenly with these agent-driven, like with that level of

10,000 sites a day from those and us and so on. And this just being the beginning, as you start seeing this emergence of, "Hey, let's just write an app for that." And you get almost ephemeral applications and so on, because why not fire up an app to solve a problem? Suddenly that whole idea of an ephemeral serverless database makes a ton of sense and will give the best experience. You fire it up and if you need to scale, you can do it. But

you also, you actually suddenly need to scale to zero, right? So that's an example of a primitive I can see now making total sense that actually like five years ago didn't make a lot of sense. Let's say everything goes the way that you would like it to go. People focus on AX. What does the web look like? So the web is obviously like really thriving and it's the platform of choice to build on and experience on. And it is a platform for building creative experiences. And we kind of have seen the next 100 million people

web developers emerge from this. I mean, is it still JavaScript and .NET? Is it still the web as we know it? It'll change. So again, when I coined the term Jamstack and we started Netlify, right? It wasn't JavaScript, right? To be a developer, you were like a Ruby and Rails or Java developer or something like that, right? And then you started getting these boot camps just spitting out

React developers, right? And there was even like this slight derision around they're not real developers and so on, right? But you probably then in that time added something like, I don't know, 14 million new developers, pure like front-end JavaScript developers. And you still have all the other kind of developers, probably still a little more of those, right? But then just like 14 million new as those. And I think

In five years, if this goes as I think we have like another hundred million new developers that are primarily working, like the deepest of them will go several layers down the stack and we'll know everything behind it. And the shallowest of them will primarily just be working with agents, but there'll be real skill in

And we'll see the cost of building internal applications and applications in general go to almost zero, right? So there'll be tons of those and lots more internal applications than off-the-shelf SaaS products, right? And then on the flip side of it, we'll also start seeing that whole new levels of full-fledged creative productions on the web from the teams that need to actually like really build

deliver differentiation. That's the future I want to be part of building, right? Like where there's just lots more people building for the web, the web is a more creative place again, and it's a platform of choice and a fully open platform. Maybe as a bit of a glimpse in the future, like I know that you actually see a lot of these kind of like new AI projects. So maybe talk through some of the cooler ones you've seen. Yeah, I just always try to collect things that people build with AI that feels truly creative, right? Because

Often I think people have this sense that you're just like,

that the agent is suddenly the creative one. I think again, it's a huge unlock for people to build really wild creative things. There's this guy on TikTok building this sci-fi epos with the Romans and the Persians going to space and massive, all with these AI generated videos, but he's really set the style and tone and so on. It's so cool. Imagine that one person can suddenly do that and how much

more weird

experiences. Cool and beautiful and bizarre and wild and wonderful. You get like AAA Hollywood movie and it's like very dull because I don't know, there's just so much budget behind it, right? But you suddenly get to the level where someone on TikTok can build it and you can get like really weird and interesting ones. Yeah, and in this case, you would want like the web to natively support something like that rather than you having to go to TikTok. Totally. Ideally, we want more like creative stuff to also happen like in the web, right? And then I see people building like crazy stuff with these code generators, right? There's

There's this guy that's been tweeting his project that he's built entirely with Bolt. Yeah. Bolt.new, this code engine. Yeah, yeah, yeah.

It's a full 3D object editor in the browser. What? Like an actual application for doing that, that he's somehow gotten Bolt to build just by prompting, right? What? I mean, he's got to be doing some code editing, right? I mean, I haven't said behind it, but he claims to do these projects entirely in Bolt. And it's just so wild because, again,

I could never get Bolt to do that. It's really not just you go to Bolt and say, "Build me an advanced 3D editor for objects and here you go." You've got to figure out a lot of stuff. One thing you learn, so I worked with a bunch of these model companies. One thing you definitely learn is that there is a skill to prompting. This idea that it allows everybody to be creative is not quite right. It just turns out that some people know what they want and they have vision and they actually know the language of these models and they create much more sophisticated things.

I did not know we were at the spot where like somebody could use Bolt to create a 3D design tool. That's pretty cool. I'll shoot you the link for this. It blew my mind when I saw it, right? Wow. Cool. Amazing. Thanks so much, Matt. Thank you. And with that, another episode is in the books. If you enjoyed it, please do rate and review the podcast wherever you listen and keep listening for more episodes on how AI is remaking the developer experience.