Martin Casado (2:58)

Thanks for joining. It's good to see you. So this is Matt Billman, the CEO, founder of netlify, also creator of the Jamstack, you know, very well known to be an architect of the web and someone that tracks the trends. And you were with us a couple months ago and we were discussing AX and what that means and now, you know, like the, the space is moving so quickly, we're a few months on and so the point of this is going to be to talk about what we've learned in the last couple months, but maybe before that, maybe you can describe what AX is.

Matt Billman (3:32)

Yeah.

Martin Casado (3:33)

And then we'll just kind of get into like, kind of what we're learning.

Matt Billman (3:36)

Yeah. I wrote, I wrote an article about the first article about AX in, in back in January, which is like, feels like 10 years ago in AI time now, like as the term for agent experience and sort of really thinking about the evolution from user experience as like the term that like really could set a product apart from other products, not just based on a checklist of features, but based on like, the experience of like, everything around it. And then our first decade at netlify of really centering around developer experience, that's typically been a way to set platforms aside from other platforms by making lower friction for developers to actually build on them and extend them and use them and so on towards this idea that we suddenly, all of us that build product, have this new Persona using the product that's an agent driven by AI and that also sort of actually have an experience of the product. Right. Like, whatever that means. Right. Like, doesn't necessarily mean that there's like some soul behind it or something. Right. But it obviously has some. Like, it's like actually attempting to figure out how to do stuff with our product in an autonomous way. Right. Like, so suddenly the documentation about our product, the onboarding experience of our product is no longer just for like a person, it's also for a person asking an agent to do stuff. Right. Like, and that, that was what I sort of described in this article on, on, on agent experience as something that suddenly for us at netlify became like our product North Star. And that I've proposed at the time should be pretty important for anybody building products in the world. Today.

Martin Casado (5:29)

So listen, you, I mean, you read netlify as we mentioned, which is one of the world's largest hosting sites. So have you seen more agents than humans showing up at these websites or, I mean, I, I feel like there's a lot of stuff going down. There's like, there's that, but there's also, like, people are creating websites using agents. And so I just feel like there's, you know, just a lot to tease out here.

Matt Billman (5:47)

I think, especially in the dev tool space, like, I, I, I'll say that, like the term AX have had like outsized adoption in the developer tool space. And I think that's because for developer tools there's this like wild thing happening where if you take netlify, right, like three years ago, our audience, like our addressable audience was essentially 17 million professional JavaScript developers, right? And then there's like some adjacent developers that might use it as well, right? Like, but it's like deeply technical product, like based on writing code and shipping it, right? Like, so the like, main audience is 17 million people. Yeah. And suddenly computers can write code. And that whole question of like, do you actually need to write code to build software? That's becoming like, not really, right. Like, and you start having it an addressable audience for a tool like ours, that's more like everybody that can use spreadsheets today, which is more like 3 billion people, right? Like in the distance between 17 million and 3 billion is really, really big. Right? Like, but if you want to reach that large audience, you have to reach them through the agents they're using to write the code and to build the software, right? So I think that has meant that in the developer tool space, this idea of agent experience is like extra important right now and there's like extra attention of it. That doesn't mean that it's limited to like developer tools. And I think we'll see it more and more in all products, right? Like, as agents try to interact with anything. But it does mean that specifically in our space, it's like become something that's like almost essential to actually like reaching your potential as a company.

Martin Casado (7:33)

So from the perspective of netlify, it seems like there's two things going on. One of them is to what extent are agents making websites using the bolts or the lovables or the replits or whatever it is to generate the website? And then there's a second one is for a website that's already generated, to what extent is an agent visiting it?

Matt Billman (7:54)

Right?

Martin Casado (7:54)

And so how do you think about, you Know.

Matt Billman (7:55)

Yeah, yeah, yeah. There's those two parts to it constantly. Right. Like, and by the way, is AX.

Martin Casado (8:01)

Both of those or is it.

Matt Billman (8:02)

It's ax. Like, we at netlify, we talk about like netify, we actually talk about three kinds of AX relevant to us. Right. Like, it's all ax. Right, like, but there's like netlify's ax. Right. Like, what is the agent experience of using netlify?

Martin Casado (8:16)

Netlify, sure.

Matt Billman (8:16)

Right, right, right. That's like netlify.

Martin Casado (8:18)

Like you're like the. Yeah, yeah. Like you're like the website or whatever.

Matt Billman (8:20)

Yeah, totally. Right. Like, it's like our cli. Can. Can cl. How to use our cli. That's like part of netlify's ax. Right. Like, then there's our customers AX the stuff that our customers built. How can we help them make that, like, make a really great agent experience?

Martin Casado (8:40)

Assuming they're using agents.

Matt Billman (8:43)

Yeah, yeah. But independently on that. Right. Like, if they're building like an E commerce site.

Martin Casado (8:48)

Yeah.

Matt Billman (8:48)

What do we need to help them with to make it easy? Like, now Stripe has a new, like, payments option to buy directly through ChatGPT.

Martin Casado (8:58)

I see.

Matt Billman (8:59)

What does that mean, building E commerce sites and netlify. Right. Like, what do they actually need to do different to make sure that ChatGPT can also buy their products on behalf of users? Right. Like, there's that whole line of talk, like, stuff like, should we help people charge for crawling like cloudflare does and do people even want that? And so on. Right. Like, that's all in our mind, like the customer where, like, how do we help our customers do these things? And then the third side of is kind of what we call Industry X. That's like all the protocols and like general standards we have to agree on and as an industry, to make all of these things simpler. Right. And that's kind of how we think about it internally. So we touch all three of those. And I would say we've done the most work on netlify's own AX right now. And that again, kind of comes back to just like how. How massive that change in like the developer world is right now. Right. But we are obviously also doing a lot of work on helping customers build MCP servers and figure out how to make their sites accessible for agents or like deciding if a visitor is an agent or human and so on.

Martin Casado (10:11)

Do you have any numbers that like, are. You can highlight on, like, whatever? Three years ago, you know, people were not using AI to build sites and now they are like, like, do you have any, can you give any sense of like, I mean the extent that.

Matt Billman (10:27)

Is just that shift in like the mass of people that are suddenly coming in the front door. It's like that's just the first obvious one where like a year ago we were sitting at around 3,000 signups a day, which for a developer tool is like that's a lot. That's a lot.

Martin Casado (10:41)

That's a lot.

Matt Billman (10:42)

Yeah. Today we're sitting at around 16,000 a day.

Martin Casado (10:45)

Wow. Right, like, and you have a sense of, do you have a sense of where they're coming from?

Matt Billman (10:49)

Like that's the wildest thing is that it's everywhere, right? Like, because like at first we started having a whole bunch of these agents sort of directly integrating with her. Sports new being like the largest one, right? Like then a whole world of like same new builder IO, Mimix, Rocket, like Codemade, Windserve, like just like all of these different ones but all of them together. It's kind of like a 4% of the signups. So the rest just happens organically. Right? And we actually bolt that new had like used our claim flow which is like become one really common technique for agent experience. This idea that you can like let an agent use your product before the human knows the product exists. And then when the human wants to get involved, they go and claim whatever the agent did and sort of open an account. And at first like bolt new which is one of these like very consumer oriented code agents, right like was using that flow and sending anybody that wanted to really do something with their website over to us. And then a few months ago we changed to a white label agreement with them where they are now like fully vertically integrating netlify and they're like white labeling us and selling it directly to their customers to give like a really integrated vertical experience. And ironically you cannot see that change in, in, in our signup numbers at all even if that was like our largest partner. Right? Like, because like the, the general growth of the space of people building web.

Martin Casado (12:31)

Like are they using AI tools? Are they using like a cursor chat GPT? Or is it just like people are like decided to make websites?

Matt Billman (12:38)

I mean I don't know for sure, right like, because I can't always see what they build it with, right? Like, but I think it would be a weird coincidence that that like just a lot more people were like I really like HTML now. Right. Like, and I think it's more likely that, that, that you'll just see that the amount of people building web apps and websites Exploding because like the barrier to entry for doing it is like massively disappearing. Like you have people today that accidentally create a web app in React, right? Like you go into chat CPT and you're like prompted the right thing and suddenly it like spins up a canvas and all this stuff and you download. It's like, okay, now where do I put it right? We're pretty much like the simplest place to put it. Right? And then we have made our tooling really great for all of these sources, right? Like, and we, we're seeing that and we can see in their Personas that come in that suddenly they go from being like developers, obviously still ton of developers. There's also a lot of Personas that like in our onboarding, like characterize themselves as like marketeers or designers or product managers or.

Martin Casado (13:47)

So I would expect if, I would expect if AI is driving this. Exactly. That would happen. Which is you go like, yeah, netlify has always been a very technical tool for developers. Like if you're like a JavaScript developer, you use like a hosting by yourself. So I would guess like probably the clearest indication that AI is doing it is exactly this like the mix of like tactical versus non technical chips.

Matt Billman (14:10)

Totally. And it really shifts.

Martin Casado (14:12)

I mean, are there any rough numbers to that or is it just. I don't, like, is it still vibes?

Matt Billman (14:16)

It's still vibes. It's still vibes, right? Like, but, but it's, I can say the vibes are very strong. Like even just, it's not even just talking to our customers, right? Like, we can, we can see this shift. We have other things that we can see things like for like our coordinates, we have this functionality called why did it fail? Where when a build, like when you open a pull request, we run a build and we push that to your site. When that fails, that's always sort of a point of like, okay, now I have to go and debug it and figure out like, why what happened? Right? Like, so we put a button in the UI that literally says, why did it fail? You click that button and we send like the logs and the code, like, to an LLM and we give you a diagnosis.

Martin Casado (15:02)

That's awesome.

Matt Billman (15:04)

Then first we put a button on saying, copy this result to an LLM, right? Like, because like, if you want, you're in that loop and you want the LLM, okay, it failed, go fix it. Right? And immediately when we put that button in, we saw that like 25% of everyone that clicked, why did it fail? Clicked copy to LLM wow, that's like also pretty big how much this is like shifting, right? Like, and now we just launched this thing called agent runners where you can like spin up like autonomous runs with cloud code or codecs and so on. Right. Like so now how we have it like this fix it button, you know.

Martin Casado (15:43)

It'S remarkable to me, I will say here's a personal anecdote which is I developed with cursor in the evenings for hobby projects. I've been using netlify for years, right. But I never remember all the commands. Like whatever was super remarkable to me is like the LLMs actually know it. So I'll be in cursor and I'll like, you know, want to be like, oh, like what are the netlify commands to like, you know, whatever, deploy this, add this, whatever. And it actually knows it. And so I would say even for the case of professional developers, they're using LLMs to debug because I mean it's just like there's just so many frameworks and many tools to know. So.

Matt Billman (16:21)

And, and you should. Right. Like I mean that, that's one of my big thesis that like the whole definition of a developer is really going to change. Right. Like where. Where the what defined a developer at its core used to be being able to write code and understand programming languages and understand the syntax and like that layer sort of below the visible substrate of things. And suddenly that part of being a developer is getting way less important and potentially going towards zero importance. Right. And then the other skill set of being a developer of clarity of thought, of understanding what users actually need you to build, of systems thinking and system design and so on, those elements are suddenly becoming like the mean qualifier for like are you actually really strong developer? That also changes the Persona, right. Like there's a lot of Personas that were like really, really great at like the puzzle level layer of like low level code and so on. But that might not be that strong in like the more humanistic layer of like taste and what to build and so on. Right. And there's a lot of people that were sort of like cut off fundamentally from development that are very strong in those layers and that start entering the field.

Martin Casado (17:44)

Interesting. I'm going to give a counterpoint to that. I don't know if this right. Or it's not, but like try this on for size. I started programming in the 90s.

Matt Billman (17:52)

Yeah.

Martin Casado (17:52)

And to program, you sat down at a computer and you knew a programming language and you wrote the programming language and you had a program.

Matt Billman (18:00)

Yeah.

Martin Casado (18:01)

And so like you were just thinking about the logic of the program.

Matt Billman (18:05)

Yeah.

Martin Casado (18:06)

And then, you know, the 2000s, especially the late 2000s, rolls around in the web and then like to be a programmer, it's no longer about writing code. It's like you have to know frameworks and there's nothing fundamental about frameworks.

Matt Billman (18:19)

Yeah.

Martin Casado (18:19)

Like, for me to understand next or react or.

Matt Billman (18:23)

Yeah.

Martin Casado (18:24)

You know, some weird cli. There's nothing fundamental. So it really is this wasted knowledge. I'm like. I'm like using space in my brain, you know, to remember some dumb design decision from like some random person. Right. And so for me, it's actually a little bit different. Which is the nice thing about AI is like all the framework stuff it knows I don't have to waste my brain on it. And I can literally just think about the programming logic.

Matt Billman (18:52)

Yep.

Martin Casado (18:53)

Which I think is actually totally what matters because you actually articulate what the program can do. So it's. It's kind of a similar thing to what you're saying, but I think maybe I'm even abstracting it more. Like, I actually think. I think code's a very effective way to articulate trade offs.

Matt Billman (19:07)

Yeah.

Martin Casado (19:07)

And to articulate what you want something to do.

Matt Billman (19:09)

Yeah.

Martin Casado (19:09)

It actually is. Right.

Matt Billman (19:10)

Like.

Martin Casado (19:10)

Like English is ambiguous. Like.

Matt Billman (19:12)

Yeah.

Martin Casado (19:13)

You know, and so it's fine to write code. I would just think all the framework nonsense is the stuff that we shouldn't like, we should like not worry about. And then of course, to your point, there are things where you don't have to describe so low level as code that you can do a high level.

Matt Billman (19:24)

And. And there's so many aspects that like we. We've both like been like hardcore. Like our brain has been changed by sitting with stuff like a Commodore 64 and like learning memory addresses and just taking for granted that like every little.

Martin Casado (19:41)

Comment I remember pokes and p. I remember, like you like poke into like independent memory.

Matt Billman (19:47)

Like you're like loading like that level.

Podcast Host (19:50)

Right.

Matt Billman (19:50)

Like, so we also forget how much of a piece barrier it is. Like just when you start, right. Like that, like that whole learning curve of like, oh, I mistyped one character and now nothing works. Right. Like, and I think we underestimate like how big of a difference it makes when all of that is sort of no longer a barrier. Right. Like where, hey, like, even if you leave out like a character here and there, like the logic still holds. Right. Like the logic is still sound. Right. Like so suddenly you just need to. To start practicing that ability to reason with. With. With logic and to understand like the system. And it's really fascinating, I think, to see some of these people. Like, I see sort of two journeys. Right. Like, I see people like us.

Martin Casado (20:38)

Yeah.

Matt Billman (20:39)

That are starting to learn how to work with AI.

Martin Casado (20:41)

Yeah.

Matt Billman (20:42)

And coming from like the world we come from. Yeah. And then I see these people that come from like the completely opposite apart and start learning to build software without.

Martin Casado (20:51)

Yeah.

Matt Billman (20:52)

That journey that we went through. Yeah. And it's so interesting to see that interplay they have with the AI then. Right. Like, because that becomes the, the, the sort of constant iteration one that's just like, why does this do this?

Martin Casado (21:04)

Yeah. It becomes basically an educational. No, no, that's exactly right. It's like. Yeah, it's like, it's like the AI not only does, but also teaches you how to do it. So you learn the fundamentals that way. No, I absolutely have seen the same thing.

Matt Billman (21:16)

Yeah.

Martin Casado (21:17)

Can I, can I move the conversation a little bit? Which is. So we've kind of been working kind of out towards the user. We've talked about like, you know, Netlify itself, you know.

Matt Billman (21:26)

Yeah.

Martin Casado (21:26)

You know, using AI and then how is a good target for AI.

Matt Billman (21:30)

Yeah.

Martin Casado (21:30)

You know, as you go towards the user, we're actually seeing the browser itself.

Matt Billman (21:36)

Yeah.

Martin Casado (21:37)

Seem to evolve. Right. And there's like new browsers, AI level browsers. Are you starting to see impact from that or is that still kind of on the come?

Matt Billman (21:48)

It's a good question. I think it's still more on to come than like something that's like very real right now. You see it in like the access patterns. Right. Like where you suddenly see like all of these sites being accessed by all kinds of different agents in all kinds of different ways that are different than human access patterns. Right. Like, and that comes from browsers and from tools like cloud code and so on. Right. I believe, like one thing that Bungie has just came up with and everybody, like us included, are copying now is this idea of like, for their documentation side use content negotiation. So when Claude code asks for a page, instead of sending the HTML, respond directly with the markdown. So it uses less tokens to like interpret it. Right. Like, yeah.

Martin Casado (22:38)

Oh, I see, I see, I see, I see. Oh, that makes a lot of sense. Right.

Matt Billman (22:41)

So we're starting to see those kind of patterns where we actually start like in practice building like content negotiation, where like, is this an agent accessing or is it a human accessing and maybe we should do different things. Right. Like, and I think those are sort of early signs of what will happen when even the browser starts being like that and starts like having AI inside that. Sometimes it's the human viewing and sometimes it's the AI in the browser.

Martin Casado (23:09)

So based on this, is your mental model that most AI is working with a human or is your mental model that these things become more divorced over time? Because the browser question to me assumes there's a human behind the browser in tandem with an AI, where like the purely agentic one, maybe there isn't a human so tightly in the loop.

Matt Billman (23:28)

Yeah, I think, I think it's pretty obvious that we, that in the beginning of this journey it was all sort of very human in the loop, right? Like, yeah. And there will, I think there will always be a large set of human in the loop interactions, right? Like, and especially the browser, I think will generally be very like experience oriented, right? Like, and I'm going somewhere and I'm experiencing something. So maybe you have this assistant that helps you with everything, but it's still like very in the loop. But it's also very obvious that especially these building blocks around actually strangely enough, like the code agents in the cli, like Claude Code Codex, Gemini, cli, those are sort of turning out to be really strong generalist agents, right? Like, because they live in that like CLI space where they can do anything you can do in a computer pretty easily, right? Like, and they can write code, so they can even write their own tools to do new things and so on. And they are kind of like really being able to do longer and longer stretches of just like uninterrupted work where you give them some tasks and they just try to solve it, right? Like, so I think I had one.

Martin Casado (24:44)

Of these CLI tools where I'm like, I was just trying to be like, okay, I'm going to make it to a full game and what I'm going to do is I'm going to, I'm going to have a. I'm going to have a test that I'll test if the game works and then I'm gonna have it just keep iterating on it until like it, you know, passes the test. And it ran for a few hours and it cost me $10,000. For those that are watching, be careful. I was like, whoops.

Matt Billman (25:16)

Yeah, yeah. But I think, I mean, just to your question, right, Like, I think we'll see both, right? Like, I do think that we are on this cusp of seeing more and more like truly autonomous agents. Hopefully they're still always doing things because some human, as opposed to like actually having True.

Martin Casado (25:32)

True Agency the robotic takeover.

Matt Billman (25:35)

Hopefully it's still like long task but some human was like hey, go do this thing.

Martin Casado (25:39)

Yeah.

Matt Billman (25:40)

Yeah.

Martin Casado (25:40)

Do you, do you think about like, I mean you, you, me. You talked a bit about it in the content negotiation, but do you actually think about the economics and how they're shifting as a result of all of this stuff? Because it does feel like yeah. Spend is very different in the world of agents. So it really is about tokens.

Matt Billman (25:56)

It's less about like never can store repricing. Right. Like we've launched new pricing for. For the same reason. Right. Like I think everybody is starting to build credit models in some form. Right.

Martin Casado (26:06)

Like it seems to me. So I, I remember the shift from perpetual to. To. To like recurring. Like I remember that right. It feels like we're going from like a recurring to like a usage based.

Matt Billman (26:18)

Yeah.

Martin Casado (26:19)

Based on this. And is, is.

Matt Billman (26:22)

It's. It's the same I would say. I think all of us are also looking for ways to make it more outcome based rather than pure usage based. Right. Like and finding, trying to find ways because like often the usage of the agents, that's like a weird thing with usage based. Right. Like that often you, you have very little control over what like is this question I asked the agent a long.

Martin Casado (26:44)

One or a short one?

Matt Billman (26:45)

Right. Like is it going to use a lot of tokens and 20 bucks? I don't really know. It's going to be 10,000 and is this game worth 10,000 or like $10? Right. So I think a lot of us are also trying to figure out what like those elements of like value are. That sounds very tough to me and it's really tough. So I think in reality everybody's kind of like usage based right now. Right. But I think at the same time it's not impossible that over time, even if it's tough, we start getting better at figuring out elements that feels more aligned to. To. To value than. Than than just pure token usage. Which feels pretty like one out of your control and not super aligned in the sense. Like if I could do the same thing for you.

Martin Casado (27:28)

Yeah.

Matt Billman (27:29)

In like five tokens, you would probably prefer it. Yeah, it would probably pay me more to do it in really fast in five tokens than to do it really slow with like 10 million tokens. Right.

Martin Casado (27:40)

Are, are you, are you using agents and operations now?

Matt Billman (27:43)

Like we are. We are using some internally but not enough I would say. But, but we are using some like we are using some like I think especially our developer organization have, have, have built Custom agents for all kinds of stuff like linear integration and management there and things like keeping up with what's happening in next year. So we can keep doing the right framework adapter and so on. Right. Like we've built a lot of that kind of internal tooling.

Martin Casado (28:14)

Yeah, it's very, I mean for me, I'm just very interested in like, like right now, today, not science fiction, like where do agents apply? Right. Clearly they're a great productivity tool.

Matt Billman (28:23)

Yes.

Martin Casado (28:23)

Clearly they can do longer running tests.

Matt Billman (28:25)

Clearly.

Martin Casado (28:25)

Like there's like basic stuff that they're pretty good at. Like for example, if I'm like, I don't know, convert from this language to this language, like there's good. These kind of like, you know, well scoped, well defined tasks.

Matt Billman (28:35)

Yeah.

Martin Casado (28:36)

Seem less good at like you know, ongoing operations where you've got to be correct and you know, like these are kind of these complex spaces and so forth and so.

Matt Billman (28:47)

Yeah. You know, I think in the end in a lot of those, like, I mean obviously the kind of agent we like the highest, highest, highest adopters of is the coding agents ourselves as well. Right. Like where even like something like agent runners, we built like we're now building a lot of the product with those agents. Right. Like, and, and have you found that.

Martin Casado (29:06)

To actually increased feature velocity?

Matt Billman (29:08)

Yeah, yeah, yeah, we can, we can see like yo. Yeah, like the, when people get good at it, it starts really increasing the velocity with which you can build stuff. And also it increases the capacity of like a really strong engineer to both go do some really hard focused work but just spin off a bunch of like the, the easy stuff along the way and get it done. Right. Like, but what I was saying there is that I still think we are more in the line of using that to build programs that does stuff when it's like automations rather than having like agents automate stuff because it becomes like weirdly unpredictable. Right. And in the end once you know what to automate, it's like simpler to just.

Martin Casado (29:51)

To actually automate it.

Matt Billman (29:52)

To actually automate it. Right.

Martin Casado (29:54)

I totally agree.

Matt Billman (29:55)

But I think the fact that it's so much cheaper to write automations changes that like you start building a lot of automations and potentially you start building a lot of apps and automations instead of like buying a lot of software.

Martin Casado (30:10)

So this is a Benjamin aside, but it just reminded me of something. So you know Max, Max Bisheman, the airflow, you know, superset guy, so like hardcore open source developer, has a company called Preset and he was telling me he's like, you know, the most amazing thing.

Matt Billman (30:23)

Yeah.

Martin Casado (30:24)

With these, with these coding agents is I've only worked on open source. Right now I'm working on a large open source. This project's been around for a long time. So all the agents know it.

Matt Billman (30:32)

Yeah.

Martin Casado (30:32)

We're doing kind of front end JavaScript stuff, which they're all really good at. And so.

Matt Billman (30:37)

Yeah.

Martin Casado (30:37)

You know, so like I feel like. And then all of the, the engineers in the company are like pretty senior.

Matt Billman (30:41)

Yeah.

Martin Casado (30:42)

So it's like, she's like. So it's like we all woke up and we were like superheroes. It was like, you know, and so like you've got these senior engineers that are using these coding models, like junior engineers on an open source code base. And you know, the velocity is 10x, right. So that makes a lot of sense. But I also compare it to, I mean I work with companies where they're doing pretty new stuff and it's not front end and it's kind of still experimental and maybe they've got more junior developers or whatever. And I feel like the velocity hasn't increased at all. So I, I would assume like, at least in the case of Netlify, like you are doing a lot of JavaScript, a lot of front end, a lot of web. I mean, I would think that it's probably pretty amenable to these models or it.

Matt Billman (31:28)

And I mean, but I, I will say that people that get good at them start using them for things across the stack. Right. Like, and, and I, I see that. Right. You can get them to where I go.

Martin Casado (31:44)

More.

Matt Billman (31:44)

More senior developers or more, more senior developers. Right. Like, I think, I think especially when the deeper you go in the stack, the more you need to really be like kind of an architect. Right. Like to get these kind of agents to do the right stuff. Right. Like, and then the higher up you go in the stack, the, the more you can now be like just jump in and do it without, like, without any development background. Yeah.

Martin Casado (32:07)

It's kind of a fun time, right?

Matt Billman (32:08)

It's really fun.

Martin Casado (32:09)

Are you coding?

Matt Billman (32:10)

I'm, I'm like. So I kind of made it like around the same time I wrote the AX article.

Martin Casado (32:15)

Yeah.

Matt Billman (32:16)

Like a little before that. I kind of started making it my personal mission that like I want to learn how to be a really good developer. Again. I would say I'm pretty strong developer. Right. I built all the back end to front end everything like for the first version. Everything like from the C to the css, everything.

Martin Casado (32:35)

Yeah, yeah.

Matt Billman (32:37)

But I sort of started saying Like, I want to make it my mission to become an equally good developer without writing any code myself.

Martin Casado (32:45)

Yeah.

Matt Billman (32:46)

And ideally, without looking at code. Wow, that's still. I still, like, you still need to do that. Right? Like, but. But I kind of, like, started to make it my mission, and in the beginning it was very frustrating, and now I think it's really fun.

Martin Casado (33:00)

Really.

Matt Billman (33:00)

And now I do a lot of pull requests, and it probably annoys the team a lot.

Martin Casado (33:05)

You're actually submitting PRs?

Matt Billman (33:07)

I'm doing that.

Martin Casado (33:08)

Wait, but you haven't looked at the code.

Matt Billman (33:10)

I mean, okay, when I do PRs to our actual code basis, I do go and look at the code because I've learned the hard way that otherwise the team does and they shout at me. So now I go and look at the code first and tell cloud code to fix it before I open the pull request. A lot of that has, like, we launched this new Agent Runner product, and a lot of that has, like, informed it because I wanted to make sure that you could actually do the whole, like, cycle of iteration before you open a pull request versus, like, the GitHub integrations where you ask, like, plot code to do something and then you get a pull request. You never knew, like, did it do something useful? Right. Like. But I do find that once you start changing that whole mentality to just work through the context engineering and the prompts and the systems thinking and guiding this model, it's pretty remarkable what you can get it to do. And there's definitely some cases where. So it's funny, there's some cases where I've built something where, realistically, if I had the time to just sit time with a code editor in a uninterrupted stretch and write the code, I would have written it faster and probably better. But as a CEO, I definitely don't have that time. And in this case, I could still build it, but just by spinning up this async run, checking in between meetings. Where did it go? What did it do? Let me give it some feedback and do another iteration and so on. So suddenly you can working that way, and it's a lot of fun.

Martin Casado (34:48)

I mean, I was just, you know, I. So I. I kind of work on these home projects, and I was actually using, like, the cursor cli. And I was at dinner, I was curious. I was doing. I was like, checking this, like a VC at a dinner, like, talking about deals. I'm checking out my coding agents are doing.

Matt Billman (35:09)

I totally been, like, in the same situation, and. And it's Funny. I've noticed, like, every founder, CEO with a technical background I talk to now are doing pull requests. It's just really funny.

Martin Casado (35:24)

I think it's, I literally think it's a framework issue. Like if, if coding was simpler like it used to be.

Matt Billman (35:28)

Yeah.

Martin Casado (35:29)

Like, everybody would still be coding, but it just got, you just have to learn too many different things. And so, yeah, it's really remarkable. Do you, do you ascribe to the. That AI is going to kind of usher in the dead. What's called the dead web theory, where it's like, it's only gonna be agents talking to websites and, you know, there's no, there's no longer a UI issue.

Matt Billman (35:51)

And I get it. Like, I, I get it, but like, I don't, I, I, I, I really don't believe in it. Like, and I don't see it. I see the opposite happening now, right? Like, in some, everything is happening at the same time, right. Like, so you have a lot of slop generated. Let, like, other agents consume and you have like, like lots of LinkedIn that's just like ChatGPT and so on. Right. Like, probably people are asking ChatGPT to summarize it from it.

Martin Casado (36:14)

And so this is like, the cliche is like, to sound fancy, I use AI to create something big and then to read it, you summarize it.

Matt Billman (36:22)

But there's also just way more people having fun building crazy stuff. And there's a lot of that out there now. And there's people building, like, really cool, like, WebGL games and so on that they could like, just never managed to build before. And like, there's so much going on that I think that theory just under, like there's so many of these theories. I think that just underestimates, like, humans and like, humans.

Martin Casado (36:48)

I will say I personally have visited more new websites.

Matt Billman (36:52)

Yeah.

Martin Casado (36:52)

In the last six months and probably the five years prior, just because there's so much more cool stuff. Like, I kind of reduced everything to like a handful of sites that I check every morning in the last five years.

Matt Billman (37:02)

Yeah.

Martin Casado (37:03)

And now there's always something new and.

Matt Billman (37:04)

It'S like, yeah, I think it's totally true. I mean, I also see it as just a general moment of, like, people building way more stuff than they did before. Right. Like, and I still see and way more people building. Right. And that's again, back to that idea of, like, it becomes more accessible to be a developer. We open up, like, for so long we've taken for granted that, like, again, if you're sort of really generous, go beyond JavaScript and so on. Right. Like, let's say what would you say is like 100 million professional developers in the world? Maybe even this?

Martin Casado (37:36)

I mean we always estimate 60 roughly. It's probably more than that.

Matt Billman (37:42)

Really generous, you say like 100 million, right? Like out of like what, 8 billion people or something like that. Right. Like that's like a really, really small part of humankind and every other human can only use software that those developer builds for them. Right. Like, and that's like such a constraint on that part. Right. Like, and to me, when that constraint goes away, we're just going to have a lot like unimaginable amount of more software. Right? Like, yeah, maybe some it will be bad, but they'll also be a lot more.

Martin Casado (38:15)

The future is going to be weirder, wilder and more wonderful than we could imagine. This is an aside, but it's an adjacency which is. I really love that 3js is having a moment.

Matt Billman (38:24)

Yeah, yeah, me too, me too.

Martin Casado (38:26)

It's amazing.

Matt Billman (38:27)

It's amazing.

Martin Casado (38:27)

So for those listening, 3js is a JavaScript library for 3D and it's always just been kind of this.

Matt Billman (38:32)

Yeah, yeah.

Martin Casado (38:33)

You know, like why would you use the web for 3D? You know, like, you know, like you didn't have like a. You know, and now there's just full on like games being written, multiplayer games being written, really high quality graphics. Like, did you see that one recently? Was like.

Matt Billman (38:48)

Yeah, it was like amazing. And just like.

Martin Casado (38:55)

And it probably wasn't even using AI, but like it is, it is a time now that I do think that precisely.

Matt Billman (39:00)

And it'll get encoded into like the, the, the language models will train on it. They'll figure it out and you'll start people just like being able to just like spin something like that.

Martin Casado (39:10)

That's why I say. And you were there too because we're old. But like, I mean this does remind me of the early days of the web. It really does.

Matt Billman (39:17)

Right?

Martin Casado (39:17)

You just never know what to expect. There's this whole creative element, you know, has nothing to do with business. No B2B sass. It's like kind of random, kind of. Yeah, no, and it's a cool time to be in the industry.

Matt Billman (39:30)

It's really cool. And I think like for me it also reminds me just like in a much. In small scale and when, when, when I started Netlify and like when was that? So I started working on it in like 2013. Right. Like and then the. We launched out of private beta in March 2015. And back then a front end developers was like they were seen as like a pseudo developer, right? Like it was like someone that took a Photoshop file, sliced it up and then handed it to a real developer that would implement it into something working, right? Like that was a front end developer. And part of like the mission back then was like, hey, if we give them like the right tool chain, these front end developers will go crazy and they will just build the whole thing, right. Like if we just separate all that backend stuff for them. And like that was like the jamstack idea, right?

Martin Casado (40:16)

I remember your early pit, I remember your early pitch where you were like best for static sites, maybe you'll see a little bit of dynamic. Seriously, like you can use it for static websites. And then like now look at it now.

Matt Billman (40:31)

Everything, right?

Martin Casado (40:31)

Everything, like everything is built using, everything.

Matt Billman (40:34)

Is built using on it, right? Like, and, and these front end developers are obviously the main developers, like no question. And that's just like the same moment I think we are with this schism where like there's still this notion of like oh these vibe coders, they'll still need like a real developer to come. Like and we will give them the tools and we all just see like this massive.

Martin Casado (40:55)

But I will, I will say like the, the hardcore developers also became full stack JavaScript developers. It wasn't like everybody just kind of.

Matt Billman (41:07)

Shifted over that and I think it'll be the same again, right? Everybody has to like take that next leap but it will be a much, much bigger pool and it'll probably also be a different shift where like today you're kind of like either a developer or not to some degree, right. Like of course there's mixtures, there's some VCs that develop software in general, right. Like people that are like developers like hired with the title developer and that's what they do, right? Like and I think that's the other shift that like I think software development will just be a skill.

Martin Casado (41:40)

Yeah, I agree.

Matt Billman (41:41)

And there will still like just like with writing there' still professional writers but like all of us also have to know how to, to write as part of our jobs. Become like that, right? Like you're like, there'll be a level of it that you're just expected to be able to do in lots of different jobs. And then there'll still be some expert developers that are like only doing that, but it'll be the minority I think.

Martin Casado (42:05)

Bring us a bit into the future. So where do you think this. So I mean I Mean just actually to try and maybe catch up the conversation a little bit. Which is, or, or, or summarize the conversation is, is last time we talked about this is a little bit speculative. You've seen actually a dramatic increase in users.

Matt Billman (42:19)

Clearly we're seeing like, we've also seen like after that so many companies start focusing on ax, right? And as I said, especially the developer tool space, right, like where I think everybody has just seen, okay, we have to go in this direction. Right? Like Bessemer had the updated version of like the developer tool loss and the first one was X. Right. Like it's like recent. It's like an example and there's so many like from Supabase to like Neon, like everybody is like super focused on like we got to work with these agents so we don't really matter, right. Like, but then also in spaces like, like the CEO of HubSpot, he's really like geared HubSpot towards like we need to just be really great for these agents to work with. Right. Like, and that's going to be the interface to much of our data and automations and everything like that. Right. Like, and you're starting to see, I think now with like chat, tbt's, shopping like integration, you'll start seeing everybody in the E commerce space be like, okay, how do we really deal with this? And so on. Right. Like, so I think it's also become a really tangible industry.

Martin Casado (43:22)

Really tangible. Changing the behavior and the CEO submitting, PR development, changing like the type of users going to the side like all of these things.

Matt Billman (43:31)

Yeah.

Martin Casado (43:31)

Like, do you have like a bit of a more science fictiony kind of prognostication for where things go or do you think now it's, can you, can you clearly see the next few years from here?

Matt Billman (43:43)

I think everybody, like I don't think anybody can right now, right. Like, I think we're definitely in this like wild space where like what, what's clear is that like that everything is changing so fast. I, I have, I have paths like I believe in, right like that the trends I believe in and things I think will definitely happen. Right. Like, and as I said, like I think, I do think we are going to this like world where software is much more of a thing that, that's not built by like this small, relatively speaking, really small part of humanity for all the rest of humanity and where software is something much more malleable and that everybody can like sort of like do do stuff on. And I think that also means that like code as we know it Today is going to be much more of a niche thing that's not really very involved in like the daily process of software which will change the infrastructure around software building a lot. Right. Like because everything is so code centric right now. Like things like GitHub and pull requests and so on. Right. Like all of that I think is going to evolve to like a level of abstraction where it's much more about understanding like the outcome of what the agents do and the iteration with those agents and like introspecting and so on. I think browsers will evolve really dramatically from this and we'll have much more like that originally concept of a user agent on the web is getting real. Right. Like we're going to have like all of these different ways of consuming the web where some looks somewhat like the browser now and some really doesn't. A lot of things that's helders like that's made it really challenging to build like very visual or UI driven experience and so on are going to be solved. Like for example all of the accessibility constraints, right. Like they, they, they're gonna not really matter because like what, what would you prefer like trying to access my E commerce through a screen reader or just through ChatGPT and voice mode. Right. Like it.

Martin Casado (45:41)

Listen, I think about this all the time, which I, it feels like the consumption layer is changing. Right. So like my son loves to chat with AI friends.

Matt Billman (45:49)

Yeah.

Martin Casado (45:50)

And when my, my son's 15, when my son becomes 18.

Matt Billman (45:54)

Yeah.

Martin Casado (45:55)

To get a bank account, like is he gonna go to some shitty sass.

Matt Billman (45:59)

Yeah, yeah.

Martin Casado (46:00)

Bank website or is it gonna come through this?

Matt Billman (46:03)

Yeah.

Martin Casado (46:03)

Which also kind of begs the question is like, you know, like the web was never super conversational based, it was more UI based. And so does that shift is like, is the consumption layer going to include AI going forward?

Matt Billman (46:18)

Like yeah, I think it's, I think the consumption layer is definitely going to include AI and a lot of things are going to happen inside like agents that people even have. Like I think people will also. That's another reason I think AX is important versus like company that think they will build. Like you'll go to their website and they will build but it's probably the.

Martin Casado (46:39)

Consumption layer is going to become AI and then the AI consumes the traditional.

Matt Billman (46:43)

Web precisely because I don't think you like, I think you will like already now you see people like have history with their ChatGPT or Claude and they really have a preference. Right. Like, and they don't want to go to like the banks and talk to that Right. Like, they just want to interact in that way. So I think a lot of that will be really driven by like the agents you interact with. Then I think at the same time, I think the application layer, it's going to be a whole new part of the web. Right. Like where a lot of interaction and seeing what's happening and that whole concept of being able to build something and put it in a URL and everybody can use it and interact with it and can be collaborative and so on, I think that's still such a strong underpinning of the web that now that everybody can build more of it, we also just at the same time we're going to see less of the traditional bank website, but we're going to see much more, like, interesting stuff.

Martin Casado (47:39)

Yeah, yeah. I think it's so remarkable, the affinity between users and their. To your point, like their AIs. Yeah, like, and it's not just memory, like a lot of people, like, well, you have the chat history of memory. It's not that. Like, did you see like the outcry in the move from GPT 4 to 5? They're like, Ah, yeah, 4 was so much better that it turns out that just meant it was more friendly.

Matt Billman (47:56)

Yeah, yeah.

Martin Casado (47:58)

You just got used to this. I think it's pretty clear that there's going to be a pretty serious attachment and they're going to want to like, actually filter the web through that. So, you know, something that occurs to me is that there's two types of approach here. So some of the tools that I use, it's like they kind of do the work for me, which maybe as an old developer, I don't quite like because I like to have the creativity, I like to be very involved. And one thing that I actually really appreciated, I mean philosophically about kind of what netlify is doing, and I just love your thoughts about this, is like, I get to use whatever tools I want and like, I get to be the developer. And like, how do you view this as it plays out? Like, is this actually a philosophical distinction that's real or am I making that up?

Matt Billman (48:40)

No, no, I think it's real. And I think there's two really different mindsets on it. Right. Like, there is that mindset of like building tools that claim to be like, we are now your developer and like, you don't need developers anymore because, like, the AI is your developer, outsourced developer, and you outsource it. Our approach has really been like, you are now a developer. Right. Like, whether it's to our existing developer base, or whether it's to all of these new people that come in and start this journey. Right. Like, it's really like, okay, if you want to go in this journey, you're now a developer and you have to learn things and you have to understand what happened. Like, how software is built. Right. Like, code is like a little part of it. Right. But if you want to talk to a database, you have to just like, some element of like, what's happening in the world. And you have to know that to be able to collaborate with these AI agents. And, and it's not like you, you're not going to get success if your expectation is that you just go prompt them, build me XYZ app and they will do perfect. Right. Like, you're going to be successful if you're willing to learn how to develop software in a world where these agents are able to multiply your abilities to write code, to do all kinds of things, stuff, but still have to take the direction from someone. Right. Like, and I think that's what sort of underpins our whole approach to this and our whole philosophy, this idea that we are, like, helping billions of people become developers and take part in this practice of building software. And not that it's like, everybody's just a consumer now. Right.

Martin Casado (50:23)

This is an aside.

Matt Billman (50:25)

Yeah.

Martin Casado (50:26)

But I wonder if you have the same experience that I do, which is I find that, like, using AI as someone who's been developing for a long time is like an exercise in abdicating opinion. Because these, these models do dazzling things. But it's not always the dazzling thing that I would expect it to be. Right. It's almost like I've got to, like, let go of expectations because I, I can articulate, I want this and it'll do that, but I would have done it quite differently. And so, you know, A, do you have the same experience of, like, you kind of have to like, let go and then B, do you think this is just something because we've been developing and, like, that will just not be an issue as a new set of developers shows up.

Matt Billman (51:08)

I mean, maybe that's why all founder CEOs is now programming with agents, because it's like, we've already gone through that in a way, hiring these developers and you ask them, like, do this thing and they come back with this thing. That's like, pretty cool. Cool. But it's like, I want to start, like, why didn't that way.

Martin Casado (51:25)

I love that. You're almost like reclaiming power. That's so funny.

Matt Billman (51:29)

Kind of already got used to that, right? Like, so now I can just do it in a closed loop with an agent that don't actually get angry at you and you're like, no, change it to my. So, but, but there's definitely something to it, right? Like, that you have to work at that level of, like, it's not exact. Like, it's not exact, right? Like, but you have to be really good at articulating what is it that matters, right? Like, and what is it you actually care about and what is it that the aging can just make up? Because, like, that's fine, like, as long as it works, right? Like, and, and it feels very different than, like, it's a very different feel than the traditional sort of part of really, like, centering yourself, like, figuring out exactly in your head, like, how should this software be put together and so on. And you need to sort of like, out of that step, have like, just a really clear idea of, like, what is the flow that really matters? What is, like, the part of this software really care about? And then what is the part that, like, fill in the blanks?

Martin Casado (52:37)

So one of the consistent things in this conversation we've had is the system should not be the developer for you, like the user, you know, sysabilities should be the developers. But now that's not just the subset of professional developers, it can be anyone.

Matt Billman (52:52)

Yeah.

Martin Casado (52:53)

So maybe let's just end on this for those that are listening to this, who are not professional developers that want to enter this, like, what would you recommend to them as far as, like, where places to start, like, how, how to get going, maybe things to read.

Matt Billman (53:05)

Yeah, yeah. I mean, so ironically, I think it's harder for me to guide that journey than someone from that space, right? Like, but people I see going on that journey, right? Like, I see them start with like, tools like Bold or Rebel it or whatever, and then I see them jump very quickly from like, did you say. Yeah, yeah, we have like, for example, like, we just had, like, that's great, by the way. I love, I love it. Right? And we just had this, this, this launch event where one of our challenges, like, silly challenge, right, like, but we wanted an event and all of our team was building the product. So it was like, who's going to build the event site for, like. So I told the team this guy that's like a customer success manager, that's his background. And like, you go in this LinkedIn, right? Like, no sort of technical background or anything, but he had been like, super active in the Bolt community. And he's just become really good at prompting Bolt, right? Like, and he was tweeting at me with like a redesign of netlify he made in bold and it was like really cool. Like, let's just ask him to make the event page.

Martin Casado (54:14)

No.

Matt Billman (54:14)

And he just prompted the event page entirely in bold.

Martin Casado (54:18)

That's so cool.

Matt Billman (54:19)

And the event is over now, but the page is probably still@netlify.com deploy right? And I think it's like a really like, I think it was a very high performing event page for us. It's actually one of the highest performing event pitch in last four years, right like, and, and he has just like, he has really learned a skill set, right? Like, it's not. I could not go alamed. I can't go prompt Bolt to make that event pitch. Like, but he's really built that skill set and he's just built it by being like really curious and then looking at all the things he likes, right? Like here's like a side. I think it's really cool. It has that pattern. I'm going to tilt bolts, like look at this site and make. Replicate this pattern and then combine it with this other pattern, right? Like, and I think a lot of the people that start now starts with that kind of sort of remixing, right? Like, and just like, how can I just like remix something? And then the people, like, then some of them start just going deeper, right? Like they start using plot code and cursor and play around and they go directly to netlify and they start like learning how more of the software work and they interrogate the models constantly. Not like they don't just take the outcome for granted, right? Like they keep interrogating like, why did you do that? And like how did you build that? And try to actually like use curiosity to understand. And those people are really learning fast.

Martin Casado (55:41)

I do think this is a really important point to make that everybody should know is. So I work with many of these AI companies and they have, you know, whether it's in a creative domain or a co domain or a technical domain, they have users that don't come from those domains. And, and they still have to work hard, right? So it's not like labor is being removed. Like it's still the, the best people work the hardest work the hot. The best people still spend the most amount of time, right? It's just like the, the, the product is different in this time. And so for those of you that are interested, I do think, you know, and to use these AI tools just realize that, you know, like, it's, it is managed magic in that it gives you result, but it is not magic in that you still have to put in the time and the effort to learn it. But they're great learning tools. They produce great outputs.

Matt Billman (56:26)

And so it's funny because, like, I, in, in the past, before all of this AI, when people asked what to do to be a good developer, I said, like, just have a really high tolerance for frustration.

Martin Casado (56:39)

That's about right. Because, like, you're just constantly going to.

Matt Billman (56:42)

Do things in the beginning that this, like, it doesn't work and it breaks and so on. Right. Like, and I think.

Martin Casado (56:47)

I'm not, I'm not sure if it's any different.

Matt Billman (56:49)

Exactly. That's my point.

Martin Casado (56:50)

Right.

Matt Billman (56:50)

Like, I think it's kind of the same. Right? Like, it's like, that's part of. It's probably still the same. You still got to be just like, okay, it didn't work, but I'm going to try again. Right? Like, and it broke and I'm going to try again and like, I'm going to figure out how it works. And you got to just have that kind of like, tenacity of like, just keep going.

Martin Casado (57:06)

Amazing. On that note, I think this is a great place to end. And thank you so much for showing up. That was a lot of fun.

Matt Billman (57:11)

That's really fun. Thank you.

Podcast Host (57:22)

Thanks for listening. If you enjoyed the episode, let us know by leaving a review@ratethispodcast.com a16z we've got more great conversations coming your way. See you next time. As a reminder, 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. Please note that A16Z and its affiliates may also maintain investments in the companies discussed in this podcast. For more details, including a link to our investments, please see a16z.com disclosures.