No Hacks: Web Strategy for the AI Age

225: Every Website Already Has An Agent Experience And Most Are Bad With Netlify CEO Matt Biilmann

Slobodan "Sani" Manić Episode 225

Use Left/Right to seek, Home/End to jump to start or end. Hold shift to jump forward or backward.

0:00 | 45:13

The user-agent string in the HTTP header has been there since the 1990s. The web was built with software navigating it on someone's behalf. For thirty years that someone was a human. That changes now. Matt Biilmann, CEO and co-founder of Netlify, was one of the first to take seriously what it means when the "user" navigating the web is an AI agent. 

He published the foundational essay on Agent Experience in January 2025, pivoted his entire company around it, and recently shipped netlify.ai as a separate entry point built for agents. We cover the four pillars of Agent Experience, why every product already has an agent experience whether you designed one or not, content negotiation as a way to tell agents to go to a different URL than humans, why SaaS is in real trouble (with a story from inside Netlify about ripping out vendor contracts), how the data-structure assumption that has defined software for fifty years is breaking, and the one thing every website owner should start doing this week.

About the Guest

Matt Biilmann is the CEO and co-founder of Netlify, the platform that started the Jamstack movement and is leading the shift from developer experience to agent experience. His January 2025 essay on Agent Experience is the foundational text for the discipline. 

Chapters

  • 00:00 Every product has an agent experience (cold open)
  • 00:35 The architectural question
  • 01:46 Welcome Matt to No Hacks
  • 02:15 When AX became a design constraint, not a concept
  • 06:44 The January 28 2025 essay and who got it first
  • 10:22 Why netlify.ai was built as a separate website
  • 12:44 Content negotiation: telling agents to go to a different URL
  • 13:54 Qualitative data and the Axis eval framework
  • 17:12 Does AX apply to e-commerce and content websites?
  • 20:59 The cumulative media argument (TV did not kill radio)
  • 25:00 User-agent in HTTP and Al Gore-era agent commerce laws
  • 26:33 SaaS business model is dead: build-vs-buy is shifting
  • 30:44 The end of structured content as a hard constraint
  • 40:25 One thing every website owner should do now
  • 43:08 Where to find Matt online

Key Takeaways

  1. Every website already has an agent experience. Agent Experience is how AI agents currently interact with your product, whether through computer use, fetching, or working around the barriers you put up. It is not a feature you add. The only question is whether the experience is good or bad.
  2. The four pillars: Access, Context, Tools, Orchestration. Matt's framework for thinking about AX systematically. Access answers whether agents can reach your product at all. Context is the prompt-engineering equivalent for agents. Tools are the concrete capabilities you expose. Orchestration covers how agents string those tools together inside your product.
  3. Build a separate entry point for agents. netlify.ai is purpose-built for agents while netlify.com remains the human entry point. Content negotiation tells agents to go to one URL, humans see the other. The blessed-path approach beats trying to make one URL serve both.
  4. SaaS economics are shifting structurally. The build-vs-buy floor is dropping fast as AI lowers the cost of software. Traditional 90%-margin seat-based SaaS is in real trouble. Dev tool companies have upside because companies need more tools. Everyone else is going to be ripping out vendor contracts and building internally.
  5. The data-structure paradigm is breaking. Software engineering has operated on the Linus Torvalds principle that data structures matter more than code. LLMs are not built around data structures. Building software around LLMs means rethinking the assumption that drove fifty years of computer science.

Notable Quotes

"Every product has an agent experience because all of these agents, whether through computer use or through fetching your website or through working around the barriers you put up from them, have some agent experience right now. It is just a question of is it good or bad."

"There is a reason it is called a user agent in the header. It was forward-looking."

"We have been ripping out SaaS contracts. Sometimes it is heartbreaking. The rep calls to right-size the contract and the customer reacts with 'let me see if I can build it with an agent.' Then they call back and cancel instead."

"The context and the flows and your creativity are probably more important than both the data structures and the code."


What To Do Next

  • Open your website in Claude Code or ChatGPT and ask the agent to complete a real task. Watch where it stalls. That is your AX baseline.
  • Check your traffic logs for AI assistant visitors (ChatGPT-User, Claude-Web, PerplexityBot, GPTBot). The number is rising whether you measure it or not. Cloudflare reports AI assistants are now 5.5% of all internet traffic, up from 3.9% six months ago.
  • Read Matt's January 28 2025 essay on Agent Experience at biilmann.blog as the starting point. Then read the one-year retrospective for the four pillars framework.
  • If you operate a developer tool or any product with a clear automation surface, start a simple eval scenario: take a fresh agent, give it a task, score whether it succeeds. Axis from Netlify will give a proper framework when it ships open source.

Resources Mentioned

  • netlify.ai (the agent-built entry point Matt and team shipped recently)
  • netlify.com (the human entry point)
  • Matt's original Agent Experience essay, January 28 2025: biilmann.blog
  • Matt's "AI in the CLI: The Humanoid Robot of the Web" (August 2025)

  • Claude Code (the agent that flipped broad accessibility for CLI coding agents)

Connect with Matt Biilmann

  • Blog: biilmann.blog
  • LinkedIn: linkedin.com/in/mathias-biilmann-christensen-a5a3805
  • Twitter/X: @biilmann (x.com/biilmann)
  • Bluesky: bsky.app/profile/did:plc:grjr4il5dredrsuj7nosb4pq
  • Mastodon: mastodon.social/@biilmann
  • Netlify: netlify.com and netlify.ai

Connect with No Hacks

  • Website: nohacks.co
  • Subscribe to the newsletter: nohacks.co/subscribe
  • Machine-First Architecture: machinefirstarchitecture.com

No Hacks is a podcast about web performance, technical SEO, and the agentic web. Hosted by Slobodan "Sani" Manic.

[00:00:00] Matt Biilmann: I do think that that means that almost everyone that builds something that's more than just content will need to think about their agent experience for like how do those agents most efficiently access their functionalities, right? And that'll also go for like anyone running e-commerce, anyone doing almost anything interesting on, on, on the web.

[00:00:35] The web has had agents in mind since the 1990s. The user agent string in the HTP header was forward looking. It describes software navigating the web on behalf of someone. And for 30 years that someone was a human, that changes now because air assistant traffic is already 5.5% of all the traffic across CloudFlare network up from 3.9% six months ago, and it's growing faster than mobile did 15 years ago.

[00:01:05] The question that every website needs to answer now is not whether the agents will arrive and visit, it's whether your website is built for them when they finally do. My guest today is Matt Billman, CEO of Netlify, and the person who has been pushing this question for longer than probably anyone else on the web

[00:01:24] he wrote the foundational essay on agent experience in January, 2025, and then pivoted his entire company around it and now shipped netlify.ai as a separate entry point built for the agents. This is no hacks episode 225. Don't forget to subscribe at no hacks co slash subscribe,

[00:01:45] enjoy the episode.

[00:01:46] Slobodan: Matt, welcome to No Hacks. It is a huge pleasure to have you on.

[00:01:50] Matt Biilmann: Thank you. Great to be here.

[00:01:52] Slobodan: About a year and a half ago, you wrote about AX, agent experience, your essay on your personal website, your personal blog at the time. It felt a bit premature back then. It felt like the world around it wasn't ready for what you're describing.

[00:02:05] Slobodan: But I read it on that day, I paid attention. It made all the sense as a concept, but I think we- ~we've, ~we've gone past that concept phase now. It's, it's, it's a bit more.

[00:02:14] Matt Biilmann: Mm-hmm.

[00:02:15] Slobodan: ago, you and your team have shipped netlify.ai.

[00:02:18] Matt Biilmann: Yeah.

[00:02:19] Slobodan: that in a lot of detail. So walk me through the moment you realized this is no longer a concept and this is now, as you called it, a design constraint.

[00:02:28] Matt Biilmann: And so for me, that was even before I published the article, right? Like, because way back in '22 ChatGPT came out, right? Like, all of us sort of had that moment of like, "Okay, what is our longer term AI strategy gonna be like?" And at the start of '23 when, I shared our AI strategy internally, there were four pillars to it, and the fourth pillar back then was like, how do we become the preferred platform for AIs to build on? Back in, in the start of '23 when, like, you could barely get AI to write, like, a coherent func- like, function in a programming language, that, that did sound very science fiction-like, and honestly, I thought that that str- part of the strategy would be more relevant towards, like, the end of this decade, right?

[00:03:27] Matt Biilmann: Like, but, but nevertheless, it made us sort of execute in that vein, right? Like, and we did several steps along, like... We had this Netlify drop flow where you can, like, go to netlify.com/drop. You can just, like, drop in a website and you get a URL, later you can go claim an account with Netlify if you want that URL to be permanent and stay up, right?

[00:03:54] Matt Biilmann: Like, and we started adopting that to, to the AI platforms and like, when ChatGPT in '23 laun-launched their marketplace.

[00:04:06] Slobodan: Yeah.

[00:04:06] Matt Biilmann: We had, like-- ~we, ~we launched, like, a Netlify plugin that you could just @mention, and then you could deploy stuff to Netlify, and you didn't even have to sign up, right? Like, but reusing that flow would give you a link, and then if you wanted to keep it, you had to go, like, sign up or sign in ~and, ~and link ChatGPT to Netlify, right?

[00:04:23] Matt Biilmann: Like, and then we shared how we did that with the community, right? Like, and some platforms started picking up on it, like early on when the first version of Devin launched, for example.

[00:04:35] Slobodan: Mm-hmm.

[00:04:35] Matt Biilmann: Had an integration with Netlify and was publishing to Netlify, right? Like, and again, it was one of those mo- moments where when I saw Devin come out, I was like, "Yeah, this is on the one hand, definitely too early," right?

[00:04:47] Matt Biilmann: Like Devin was kind of like way behind, like way, way ahead of its time, right? Like,

[00:04:52] Slobodan: Yeah.

[00:04:52] Matt Biilmann: other hand, I was like, this model of like an autonomous agent that's just like using tools and so on, that's, that's might be too early, but it's definitely the future. And I actually like reached out to Scott immediately and talked to him and, and, and wrote a small angel check, right?

[00:05:08] Matt Biilmann: Like, and then kept sort of sharing that more, and at the end of the '24 when, when Bold and Lovable came out both of them had been experimenting with us in that flow, and Bold fully embedded it. The Lovable community was like very close around us, and we started having a bunch of other agents that integrated it. And I, and I started taking it even more seriously, right? Like now it was starting to be the first real, not these longer running coding agents yet, right? Like, but, but definitely like agents writing code and publishing to us. And I was on a call like I think in, in probably like December with one of our like distinguished engineers and talking about like he, he was shaping like the future programming interface for Netlify functions. And it was like we were talking about, okay, but these, how do we make sure that these coding agents can work well with it? And, and I, and I said the phrase on that call, like, we gotta consider like the, the agent's experience as,

[00:06:11] Slobodan: Okay.

[00:06:12] Matt Biilmann: this. And, and immediately when I said it, it was like agent experience, that's kind of something, right?

[00:06:17] Matt Biilmann: Like, and I simmered over it. then at the start of '25, when we had our company kickoff in, in January that, that year, I kind of surprised everyone by coming

[00:06:29] Slobodan: Right.

[00:06:31] Matt Biilmann: around like agent experience the future of Netlify being like the shift from developer experience to agent experience and like the whole product strategy around that.

[00:06:44] Matt Biilmann: And then like sort of a week after that, I posted my, my blog post about agent experience to the external world, right? Like, but when you ask like, when did I had a big conviction? That was like before I po- ~po-~

[00:06:57] Slobodan: Just before S-- That was January 28th, if I remember the date correctly, 2025. Also you had a one-year anniversary revisit post on your blog that, that also very, very interesting. So the idea really is with those public websites that you mentioned that you can drop, right? Slash drop.

[00:07:15] Matt Biilmann: Yeah.

[00:07:16] Slobodan: is to eliminate the agent friction.

[00:07:18] Slobodan: ~So agents at that time were not able to sign up for anything, uh, Stripe link, which is something that you are a partner of that, that just came out recently. That was not possible. ~So you wanted to just make it smooth for the agent without the human in the loop as much as possible.

[00:07:23] Matt Biilmann: Totally. And i- in general,

[00:07:25] Slobodan: I love that.

[00:07:26] Matt Biilmann: ~the, ~the thought was really that it's like the same thing ~when, ~when I launched Netlify back in the day, right? Like, it was the same. It was like looking at developers, what are they doing right now to get something live on the web, right? Like, do we as a company try to remove all the friction from their code to having a URL at the link, right?

[00:07:46] Matt Biilmann: ~And, ~and a lot of that just came down to really understanding developers and how

[00:07:50] Slobodan: Right.

[00:07:52] Matt Biilmann: right? Like, so the idea here was that, okay, now we have to start doing the same wi- with these agents as they evolve, right? Like, really understand how they work, how they struggle, what causes problems for them, right?

[00:08:05] Matt Biilmann: Like, and, that post a, a year after, I think I sort of laid out the framework I developed through that first year on, on sort of the four pillars that I

[00:08:15] Slobodan: Yep.

[00:08:16] Matt Biilmann: ~are, are, are, ~are really central, where the f- the first one is access, right? Like, can these agents actually access your tool at all? The second is, is context. I f-

[00:08:28] Slobodan: Yeah.

[00:08:32] Matt Biilmann: their world of, of seeing the world, right? Like,

[00:08:34] Slobodan: Right.

[00:08:38] Matt Biilmann: engineering but ~for, for, ~for agents, right? Tools, that's the ~fir- ~third pillar, which is really like what concrete tools do we give these agents to interact with our product? And orchestration is the fourth one, right? Like, ~what do these... What, ~what kind of orchestrations do these agents need them to work with our product and sometimes inside our product? ~so, ~so those ~was-- were, ~were kind of the pillars, right? Like, and as you said, ~when, ~when I wrote this article, I think there was a small group of people that immediately got it. Like, same reason that I had just pivoted my whole company around it, right? Like,

[00:09:17] Slobodan: Exactly.

[00:09:18] Matt Biilmann: to be about them, right? Like, there was,

[00:09:20] Slobodan: Yeah.

[00:09:22] Matt Biilmann: I think it was especially developer tool companies' founders in Silicon Valley were kind of the first that was, like, imbued enough in it and, and saw the first, like, "Hey, how do we actually approach these coding agents that were coming out?"

[00:09:39] Matt Biilmann: Right? Like, tho- those were probably the first that got it. A few people beyond that, like the, like VP of AI for, for, for Microsoft that that are sort of more out of the UX world or really got it, and so on, right? Like, but it was like, as you said, way too early for, for many people, right? Like, people still thought of these things as like... This

[00:10:00] Slobodan: At bots.

[00:10:01] Matt Biilmann: Karpathy coined the term vibe coding, right? Like, and then even after he coined that, right, like, then this long period

[00:10:07] Slobodan: Yeah.

[00:10:08] Matt Biilmann: until like the very end of last year where people still thought of like, okay, there's like vibe coding, but that's just like sort of like a fun phenomenon.

[00:10:16] Matt Biilmann: It's not like really important, right? Like, and then, then we started seeing a shift to like, okay, this, this, this is starting to

[00:10:22] Slobodan: I think what flipped the switch was CloudCode. Like, credit where credit is due, because ~that, ~that just made it accessible to so many more peop... ~If, ~if you're locked into a platform like Bolt, Lovable, you're locked in with a platform. CloudCode is totally open space. About that Netlify.ai website,

[00:10:39] Matt Biilmann: Yeah.

[00:10:40] Slobodan: why did you decide to build it as a separate service and not create a version for agents at Netlify.com?

[00:10:47] Matt Biilmann: We'll end up tying the two pretty

[00:10:49] Slobodan: Okay.

[00:10:50] Matt Biilmann: right? Like it's kind of easy for us now to put the copy, like the thing you see as a human when you get to netlify.com is really like copy netlify.ai to start deploying and paste that

[00:11:03] Slobodan: Right. ~Right. Right.~

[00:11:04] Matt Biilmann: Like that, that's something we

[00:11:05] Slobodan: Of

[00:11:09] Matt Biilmann: then for agents, ~it's, ~it's starting to get easy for us to do content negotiations on our main site and include instructions for them to go to netlify.ai to figure out how to get started, right? Like, but I still think ~my, ~my core argument was that we still, our main website ~is, ~is still to help humans understand what is Netlify, right?

[00:11:30] Matt Biilmann: Like it's humans working with agents, but it's still a website for humans. And I, and I don't think it makes sense for us to change Like, it's not like the humans have stopped caring or like the

[00:11:45] Slobodan: course, Tom. Yeah

[00:11:46] Matt Biilmann: in the driving seat anymore. but that also led to this sense that like, okay, if we're really building something specifically for, for agents, how do we give them the, the, the most compact form of like th- this is now not for humans, right?

[00:12:05] Matt Biilmann: Like, and that was where that idea of like, maybe it makes sense to just like build a different entry point that is really separately for agents, right? Like that's really focused on if you are an agent, how do you, how do you start building on, on, on Netlify, right? Like, and then obviously we build a human mode in just because like humans will still

[00:12:27] Slobodan: Right.

[00:12:27] Matt Biilmann: wanna go there and see what's this thing, right?

[00:12:29] Matt Biilmann: Like, but ~but, ~but it became like this idea that, yeah, in- instead of like mixing everything, let's have a path that you can give to an agent that would kind of like be this, this is a blessed path for you, right? Like, and

[00:12:44] Slobodan: And you said you have ~content negotiation, ~content negotiation where you literally tell the agent, ~"Just, ~just go there ~instead of, ~instead of the human site," and which just perfect.

[00:12:51] Matt Biilmann: go-- like, on our main site, we will probably start telling the agent, "Go here to get

[00:12:55] Slobodan: Mm-hmm.

[00:12:56] Matt Biilmann: Like,

[00:12:56] Slobodan: Right.

[00:13:01] Matt Biilmann: Like, people will still expect their agent to be able to navigate our website for them,

[00:13:08] Slobodan: Of course.

[00:13:09] Matt Biilmann: they'll still expect the agent to be able to answer, like, ~what's, ~what's the, like, claim on their main website of, of, of Netlify, right?

[00:13:17] Matt Biilmann: Like, it'll be confusing for them if they ask the Netlif- like, agent then and then they come... agent come back and say, "Oh, netlify.com is just, like, a set of instructions for agents to get started," right? And they'll be like, "You're wrong." Like, I can see it, right? Like, so I don't

[00:13:30] Slobodan: That's a good point.

[00:13:31] Matt Biilmann: like, the same, the meaning of our URL that much, Like, but it becomes easy to have, like, these clear instructions for agents. Okay, but ~if you're getting s- ~if you're looking for the sign-up button, right? Like, for you, it's basically just, like, netlify.ai. And if you're a human, then, then it's like a button that says sign up.

[00:13:54] Slobodan: Makes all the sense to me. It's been ~three weeks, ~around three weeks since the launch, three maybe four weeks. ~I'm, ~I'm not sure what the exact date was.

[00:14:01] Matt Biilmann: Yeah.

[00:14:01] Slobodan: What's the thing you've learned from the traffic data ~in that, ~in that time?

[00:14:05] Matt Biilmann: ~It's still, ~it's still very early, right? Like, and I think a lot will depend on us orchestrating the site more into the wrist of, of Netlify, right? Like at first we've been less, honestly less interested in traffic data

[00:14:22] Slobodan: Mm-hmm. Mm-hmm.

[00:14:22] Matt Biilmann: in qualitative data.

[00:14:24] Slobodan: Right.

[00:14:28] Matt Biilmann: their agent there sh- that can use Netlify, right?

[00:14:31] Matt Biilmann: Like, so we've been more careful with testing that before we sort of like make it the main gateway. and we are working on, like what one project we are working on that will kind of like, that's a bit of a foundation for that. It's a project ~we're, we're, ~we're gonna launch it as, as, as an open source project.

[00:14:54] Matt Biilmann: We're still finding out the exact launch date, but like Sean Roberts, our VP of AI, is driving it and has started talking a bit about it. It's gonna be called Axis, and

[00:15:03] Slobodan: Mm-hmm.

[00:15:03] Matt Biilmann: it, it's an eval framework for AX essentially, right? Like, so it's really a framework to allow you to write scenarios

[00:15:14] Slobodan: Mm-hmm.

[00:15:15] Matt Biilmann: have different and different setups for those agents go through those scenarios, and then getting a scorecard on the other side to say ~how, ~how well they performed in that scenario, right?

[00:15:25] Matt Biilmann: Like, that'll allow you to do things like let's say For Claude code, s- like you know nothing, like take this web project and deploy it to Netlify, right?

[00:15:40] Slobodan: Mm-hmm.

[00:15:41] Matt Biilmann: or Claude code, you have visited Netlify.ai, and you've installed these skills, deploy something ~on, ~on Netlify or Claude code. You have like the MCP installed g- right? So we can actually start sort of really building the eval so ~what, what, ~what gives the right experience. Because like part of what we've always seen is that like ~it, ~it, it's easy to assume that we know what will help the models and then figure out it won't. So long way of saying it's a bit too early to give like concrete like th- this is what the traffic is telling us other than like that there's a ton of interest in this and like it, it's something that people are curious about and sending their agents to, right?

[00:16:29] Matt Biilmann: Like what I really wanna figure out is like how do we make this frictionless that we can start then really directing agents there from our main properties.

[00:16:40] Slobodan: Right. And one thing you just mentioned CloudCode install the skills. That website literally has this is how you install the skills that you need to work with Netlify, which is exactly what an agent would need. It doesn't need to read all the documentation. It can read the docu- you have docs.netlify.com as well that agents can read just as well as humans.

[00:16:59] Slobodan: But this is a starting point, and these skills are I mean, Cloud-- if you work with CloudCode or any, any other system similar, the skills are really the guardrails as much as AI can have guardrails. So

[00:17:10] Matt Biilmann: Totally.

[00:17:12] Slobodan: I wanna ask you about this. When I talk about agentic ex- agent experience and agents doing things, the main pushback is, yes For tech companies, it makes sense.

[00:17:23] Slobodan: Agents work-- Vibe coding tools work with tech companies with like Netlify. Does this, like a, a content website, an e-commerce website, does that make any sense now or in the future, or is it just a completely different category?

[00:17:37] Matt Biilmann: ~It's, ~it's a good question ~and, ~and one that I think will have a really mixed answer, right?

[00:17:43] Slobodan: Mm-hmm.

[00:17:43] Matt Biilmann: if you study sort of the history of new media over time, one of the interesting thing is that we mostly tend to be cumulative with media. Like when a new media form never really tries, like never really ends up replacing the previous form, it just adds to it, right?

[00:18:00] Matt Biilmann: Like we have podcasts now, but we also still have Radio stations and talk shows, right? Like we have TV, but, but radio obviously still, we, we still have like posters on, on, on lampposts, right? Like

[00:18:16] Slobodan: Right.

[00:18:17] Matt Biilmann: we still have paper books. We still have, right, like it's not like the typical pattern is that a new media out the old form of communication goes away. So I don't expect this to be like that much different, right? Like I think we, like we'll, we'll see a shift to new kind of things being built and the old things still being around. And I think humans between humans, like there's so much talk about like everything being on the fly generated and so on, I think like generative UIs and all of these like agents will be extremely important, and we'll see a lot of them getting built in. We're doing a lot of work at Netlify to make it a good platform to build those on when you're building on the web and so on. But I think humans also really liked, likes to have something that feels more permanent and shareable, right? Like we actually like that like a bunch of us can go to the same URL and see the same thing, because otherwise it's hard for us to, to even talk about it, right?

[00:19:21] Matt Biilmann: Like it's hard to us to actually it part of our communication, right? Like we kind of really like permalinks to things that like doesn't feel like I shared with you and you got your own version, right? Like then how, how can we talk about it? So I, I still think that like in that sense, websites and product catalogs and all of that are probably still play a strong role. the same goes for web apps, which I think is just gonna be way more important, right? Like I think the cost of software is going dramatically down means we'll build way more of it, and the web is the platform that all of that is happening on because it's kind of the only that can pub-- like you can publish on and iterate on fast enough for it to making sense with this new technology, right?

[00:20:16] Matt Biilmann: Like so I, I also think that we'll see an explosion in those, and I don't think will want no software like that just because agents can do anything, right? Like I keep thinking companies that have access to hundreds of thousands of humans don't tend to replace their software with

[00:20:37] Slobodan: ~No, absolutely. ~Yep, absolutely.

[00:20:38] Matt Biilmann: they tend to need way more software for all of humans ~to, ~to coordinate and for, ~like, ~people managing those humans to see what the hell's going on, right? Like, so, so I think that'll kind of be the same when we add, like, billions ~of, of, ~of agents to the mix. ~in, ~in that way, I think most of this will be additive rather than,

[00:20:55] Slobodan: Right.

[00:20:57] Matt Biilmann: gonna be a bunch of stuff we no

[00:20:59] Slobodan: ~I, ~I couldn't agree more. And my favorite example that Rand Fishkin shared on this podcast is

[00:21:04] Matt Biilmann: Yeah.

[00:21:06] Slobodan: became the majority, desktop traffic stayed flat. It didn't disappear. It was just a new lane of traffic. But how about those agent skills, for example, that you have? ~That, ~that you have instructions for the agent.

[00:21:17] Slobodan: Would that ever apply to an e-commerce website? Would that ever apply to any other category? I mean, Netlify is really-- It's kind of perfect for agents to work with, y-y- especially with the, the AI-assisted coding explosion that we, that we're seeing right now. Do you think this translates to all the other industries?

[00:21:38] Matt Biilmann: ~I, ~I think it will over time, right?

[00:21:39] Slobodan: Mm.

[00:21:40] Matt Biilmann: Like ~when, ~when you talk about Claude Code and so on, right? Like I, I, I had this other article on my blog,

[00:21:45] Slobodan: Yeah.

[00:21:49] Matt Biilmann: AI in the CLI, the humanoid robot on the-- of, of the web from back in August last, last year, right? Like, that was really saying that, like, I think is happening is that CLI-based coding agents, they're not like some vertical phenomenon. The, the thing that, that the f- that everybody started some faster, some slower when, when Anthropic came up with Claude Code was that having an autonomous agent that can live i-in a command line interface, having access to a whole computer can write code for itself, can write scripts and so on, that's kind of like a model of like a truly autonomous web robot or, or cyberspace robot, right?

[00:22:42] Matt Biilmann: Like, just like we talk about, like, humanoid robots are, are, are interesting because they, they can use the same interface to the world as we do, right? Like, they can potentially do everything a human can do in a world where more specialized robots needs the world to change for them to be able to use it, right?

[00:22:59] Matt Biilmann: Like, these CLI-based robots, like the first attempts were really like computer use, right? Like in taking screenshots of UIs and so on. It's just like very inefficient. I think with Claude Code, we started seeing this path of like, hey, if we give the agent access to the tools like a senior developer uses to interface with the digital world it can really start doing anything, right?

[00:23:23] Matt Biilmann: Like, and then all the frontier labs started really going there and doubling down and building these, right? Like, and it's because they're not like a vertical area. like the foundations of ph-- of, of building agents that can truly, like, do anything you can in cyberspace, right? Like, and we're seeing that with like Cloud, Claude Code Work now.

[00:23:44] Matt Biilmann: That's kind of like level up of Claude Code,

[00:23:47] Slobodan: Yes

[00:23:48] Matt Biilmann: based on the same principle, right? Like Codex is kind of going in that way. I think more and more that will lead to this world where humans will interact with everything online agents that are kind of like their agents, right? Like, and that they have memory with, and they've built their own skills with, and so on, right?

[00:24:07] Matt Biilmann: Like, and I do think that that means that almost everyone that builds something that's more than just content will need to think about their agent experience for like how do those agents most efficiently access their functionalities, right? Like, and that'll also go for like anyone running e-commerce, anyone doing almost anything interesting on, on, on the web.

[00:24:38] Slobodan: It's a very exciting future. It's, it's completely different compared to anything we've had online for the last 30 years. This, this is why I'm so excited about it, because look at all the shifts we've had. It's always been about human doing things in a different way. This is the first time that the human is absent while things are happening.

[00:24:56] Slobodan: A-a-and I cannot wait to see how this plays out.

[00:25:00] Matt Biilmann: And, and one of the things that's fascinating, right? Like, is that people had this vision for the web way back,

[00:25:09] Slobodan: Mm-hmm.

[00:25:09] Matt Biilmann: way back in the earliest days of the web, like in the '90s ~when, ~when it was built out. Like, there's a reason that it's called a user agent in the header, right? Like,

[00:25:19] Slobodan: Yes.

[00:25:19] Matt Biilmann: and, ~and,~

[00:25:20] Slobodan: Yes.

[00:25:20] Matt Biilmann: and it was like forward-looking, right?

[00:25:22] Matt Biilmann: Like I, I met one of the people that worked on like the legal works around

[00:25:27] Slobodan: Right

[00:25:31] Matt Biilmann: when Al Gore was sort of like ~driving, ~driving that work that, that opened up a lot of like being able to drive transactions and so on on the web under some legal framework. And what was interesting in that back then, they actually like we have laws from back then from the '90s of like what happens when an autonomous independent computer agent misunderstands your instructions and spend way more money than it's supposed to through an e-commerce transaction with website. it's like pretty funny that like the, the ideas actually goes all the way back there and, and people were actively like really thinking about them and, and, and now, now all of that is actually starting to happen, right? Like, and it's really interesting to, to witness.

[00:26:17] Slobodan: I'll just say finally. But yeah, y- you're right. Tho- those laws are from the '90s. Though-- I mean, user agent, all-- most of the rules that are powering the internet still are from the beginning of the internet. I'm sure there's been standards updates and all that stuff. You had a very interesting argument in a LinkedIn.

[00:26:33] Slobodan: Just a ti-- a little break from, from AI and a- agents. You said that SaaS is dead.

[00:26:38] Matt Biilmann: Mm-hmm.

[00:26:39] Slobodan: dead. I'm not going to say... You said probably dead. That 90% margin software is a service model.

[00:26:45] Matt Biilmann: model is

[00:26:46] Slobodan: Yes, yes. No, no, th- that, that's what I mean. That's what I mean. ~I'll, ~I'll be very clear. And you said it's because there were maybe 100 million developers in seven billion people.

[00:26:54] Slobodan: I think that's way too generous. I think it's way, way less than that, but let's say it's 100 million,

[00:27:00] Matt Biilmann: Yeah,

[00:27:00] Slobodan: and AI broke that scarcity.

[00:27:02] Matt Biilmann: Totally.

[00:27:06] Slobodan: What does that mean for the future of software?

[00:27:09] Matt Biilmann: Yeah, I mean, I think the core argument for that is that the build versus buy equation for, especially for companies, is really shifting fundamentally, right? Like and I think it's shifting in a way where before AI, if you look at like the largest tech companies, if you look at like or Google or Uber or Microsoft and so on, right?

[00:27:35] Matt Biilmann: Like, generally they will, spite of this size, kind of like be less prone to build-- to, to buy normal SaaS software like the sort of like non-software companies, right? Like, because they will see that across so many areas of like managing their employees or driving internal communication, all these things, they, they want to actually invest in building their own that's truly theirs and that's tailored to like them as a company. And then when you're smaller companies, on the flip side, like you cannot do that. You do not have the resources. You have to buy software that, that kind of also tells you how do I do this and, and, and that lets you operate, right? Like, and I think that equation is shifting in a way where like the lower bar for where it starts being really valuable to build your own software stack, it's just like, it's just like rapidly going downwards, right?

[00:28:38] Matt Biilmann: Like, and obviously that will put very, very real pressure and, and market pressure on the traditional SaaS companies. I think like one interesting thing that's happening is that like everybody's talking about this SaaS apocalypse, right? Like, so the stock market is kind of reacting to that by, by kind of bucketing almost all software companies together and

[00:29:04] Slobodan: Right. Yep.

[00:29:05] Matt Biilmann: they're all gonna be replaced," which I don't think ~is, is, ~is necessarily true, right?

[00:29:09] Matt Biilmann: Like I think- The software companies that are essentially dev tool companies, lot of those now have like massive upside because companies are gonna be needing more tools. But the SaaS companies, like with the traditional like seat-based model selling you a solution, are like in huge risk of either like fundamentally ruining their business model around their margin model ~or, or, or, ~or outright ~being, ~being replaced or having tons of like thousands of smaller vertical competitors like

[00:29:49] Slobodan: Right.

[00:29:50] Matt Biilmann: lean budgets and an AI like eat into their market share, right?

[00:29:54] Matt Biilmann: Like, so all of those are gonna be in trouble.

[00:29:56] Slobodan: And it's not just risk anymore,

[00:29:58] Matt Biilmann: just

[00:29:58] Slobodan: trouble. Yeah

[00:29:59] Matt Biilmann: ~we're, we're, ~we like really have front row for that, for this also just internally, right? Like, because we have forward-looking team ~and, ~and trying to do this, and we've been ripping out like SaaS contracts and just like, it's like sometimes heartbreaking, right?

[00:30:15] Matt Biilmann: Like sometimes it's like because we're growing fast and some rep calls and it's like, "Oh, I see like the, the

[00:30:20] Slobodan: Right.

[00:30:21] Matt Biilmann: up. We should like right-size your contract." And like someone just reacts to that with like, "Oh, let's see if I can just like build it with an agent," then call the rep back, and it's like, "Oh, sorry, we're gonna cancel instead."

[00:30:31] Slobodan: SaaSpocalypse or not, ~it's, ~it's an end of an era of software for sure, and, ~and this is, ~this is never coming back is what this feels like. Like th-th-this is just

[00:30:40] Matt Biilmann: no.

[00:30:40] Slobodan: the world is now different. I'm sorry.

[00:30:42] Matt Biilmann: Totally.

[00:30:44] Slobodan: Another thing you said, y-y-you wrote about the data layer being the new bottleneck, so just building stuff is not the problem.

[00:30:49] Slobodan: But

[00:30:50] Matt Biilmann: Mm-hmm.

[00:30:50] Slobodan: like you said where that data, where that app should live, which is what Netlify does. Infrastructure is, is just-- it's not at risk. That's my impression as well.

[00:31:02] Matt Biilmann: I mean, to some degree, everything is at risk, right? Like, I

[00:31:06] Slobodan: yeah.

[00:31:06] Matt Biilmann: the AI

[00:31:07] Slobodan: Of course, absolutely.

[00:31:08] Matt Biilmann: everything is at risk, right? Like, but so I think software companies have to do two things. Like I think, one, we have to work for the agent. The agent's gonna have to want to use our software, and it's gonna need to be helpful for them, right?

[00:31:22] Matt Biilmann: Like, and for us, that means that like coding agents that help humans build stuff, like we gotta be really helpful for, to them. And then I think companies need to-- software companies need to use AI to get way more ambitious themself on how deep in the stack they build and how much value they build and how like fast they build that.

[00:31:45] Slobodan: I agree with that. As I mentioned, I, I do website optimization. This is really-- I, I enjoy optimizing web, technical, CRO, all that. I follow the, the traffic statistics AI assistants and, and how that number is rising. It is currently at 5.5%, which means ChatGPT user, Claude user, not, not bots, not, not search.

[00:32:06] Matt Biilmann: Totally.

[00:32:08] Slobodan: AI agents people send to do stuff for them. That number is 5.5% now. I think it was 3.9% of all internet traffic across Cloudflare network six months ago. This is growing faster than mobile traffic was growing 15 years ago at a, at a rate that is faster. This is happening and

[00:32:26] Matt Biilmann: Yeah. ~Yeah. Yeah. Yeah.~

[00:32:26] Slobodan: ~it's, ~it's impossible to ignore.

[00:32:27] Slobodan: I know most people are still not fully aware of what's coming,

[00:32:31] Matt Biilmann: Yeah.

[00:32:31] Slobodan: but ~this will cr-- ~this will not kill all the other traffic. Like you said, this is a new, new way for people to consume information, and it will just be on top of everything else.

[00:32:40] Matt Biilmann: Yeah.

[00:32:40] Slobodan: So this is ~this is, ~this is really, really interesting. One thing that you wrote about is the, the-- there's no need for structured content, API interchange format, schemas, all that stuff.

[00:32:50] Slobodan: You didn't say it's completely unnecessary, but you said that we have that because developers needed structure to build on top of content.

[00:32:56] Matt Biilmann: Mm-hmm.

[00:32:57] Slobodan: said ~that ~that constraint is going away because LLMs can read unstructured input and combine it ~with, ~with the templates directly. So can you explain that?

[00:33:04] Slobodan: I'm not sure I fully agree, but I want to hear it from you.

[00:33:07] Matt Biilmann: Yeah. Yeah, yeah. I mean, I think it's the broader sense that like, And again, this, this I think is another prediction that we'll see

[00:33:13] Slobodan: Mm-hmm.

[00:33:14] Matt Biilmann: now,

[00:33:14] Slobodan: Okay.

[00:33:15] Matt Biilmann: and then one and a half year from now, maybe not so early. And

[00:33:19] Slobodan: Right.

[00:33:24] Matt Biilmann: software engineers always operated from the notion of like the, the foundation of great software is the data structures, right?

[00:33:35] Matt Biilmann: Like there's the famous quote from Linus Torvalds, the creator of Linux, that good data structures are more important than good code, right? Like and I've always like really embedded that, right? Like what is the input structures? What are the output structures? Because in the end, like computers have been machines that works on data structures and do transformations of them, right?

[00:33:56] Matt Biilmann: Like in any normal computer program is a series of transformations of data structures, essentially, right? Like

[00:34:02] Slobodan: Yep.

[00:34:03] Matt Biilmann: be encoded in ca- type systems and so on, right? Like now the really fascinating things with LLMs is that they completely break that, right? Like they are not inherently around data structures.

[00:34:17] Matt Biilmann: They don't internally-- They're not internally built around any concept of data structures, just like we are not as humans, right? Like our natural way of interacting with the

[00:34:27] Slobodan: Right.

[00:34:29] Matt Biilmann: Like rigid structures, but through language and symbols and and so on. And that's actually the same for the LLMs, right?

[00:34:36] Matt Biilmann: Like, their inherent way of interacting with the world is not through structured data. And that means that, like, building systems around LLM, we need to sort of rethink how we build software when it's not about transforming structured data, but around, like, actually working through logic and reasoning with, with, with, with models, right?

[00:35:01] Matt Biilmann: Like, and small examples of, of, of, of that that I can already give, right? Like things like I, I changed, like, the publishing engine for my own blog a w-a while ago, Like, where typically you might use a, a, a content management system. What that content management typically do when, when you publish blog posts is that it allows some human to translate some unstructured

[00:35:32] Slobodan: Right.

[00:35:32] Matt Biilmann: into structured content so that you can build web properties around it, right?

[00:35:36] Slobodan: Yeah.

[00:35:37] Matt Biilmann: and typically the original content is, like, unstructured, right? Like, it's typically like a team working in a Google Doc or a Notion Doc or something like that, right? Like, that's just most like, at, at most like loosely structured, right? Like,

[00:35:48] Slobodan: Absolutely. Yep.

[00:35:50] Matt Biilmann: and, and before LLMs, you had to do that conversion, right? and then when you had done that conversion as a developer, you had to build the whole website with the structure that was there, right? Like, you couldn't just later say, like, "Give me some different structure," right? Like, that's the structure you have, right? Like, now, for example, in my, in my blog, right, like internally in the software, of course, there's data structures, right?

[00:36:18] Matt Biilmann: Like it's an Astro blog. It has like of articles. Each of those is marked down with some front matter and so on, right? Like, my publishing mechanism is now just telling an, an agent runner from Netlify, And, and before we had agent runners, I had a little admin panel with an LLM, but now for a different reason, I use agent runners. Just tell it take the, take the AI and ambition doc from Notion and publish it as an article.

[00:36:48] Slobodan: Yep.

[00:36:48] Matt Biilmann: And, and doesn't need me to give any extra structure or anything. It can figure out how to read that article, like carry over the, the body text correctly format all the images and, and insert them and put those in public.

[00:37:03] Matt Biilmann: It can figure out coming out with like an SEO description that's fine and so on, right? Like and, and it didn't actually need, didn't need any human to first structure the content, right? Like... And then, then I asked it to do something else that was like more for fun and then more an experiment, right?

[00:37:20] Matt Biilmann: Like then I also thought like- this, the index of my blog post was always just like, okay, what do I have to work with? Like, I have, like, the title and the summary and,

[00:37:30] Slobodan: Yeah.

[00:37:31] Matt Biilmann: list of articles, right? Like, so then I to- like, told, an agent runner, ~make, ~make a skill that where you pretend that you're like a, a magazine sort of like,

[00:37:46] Slobodan: Mm-hmm.

[00:37:48] Matt Biilmann: page editor from a print magazine, and every blog post is gonna be like a new edition, and you just gotta decide, like, what should the front, like, front page be like based on this edition.

[00:37:58] Matt Biilmann: And you

[00:37:59] Slobodan: Right.

[00:37:59] Matt Biilmann: tool to generate, like, illustrations and so on, right? Like, and, and, and it's kind of fun, right? Like, now whenever I publish a new blog post, like the index of, of my blog is completely gonna change, right? Like the, the AI's gonna come up with like some image and how to lay it out and how to pick, a theme and like it, will take the content and it will create new summaries, right?

[00:38:27] Matt Biilmann: Like it, it'll like fields that isn't there in the markdown and so on, right? Like, and again, it's a small example, right? Like it's just my blog and I'm just playing around. I think it's a good example of, of what's gonna be possible term if we start thinking that we are not restricted to working with structured input and structured output, but, but that we can actually just have LLMs make these decisions on the fly and then use

[00:38:54] Slobodan: ~That, that proc-- I agree with that. ~I agree, I agree with you now that you explained exactly what, what the process is you, you had in mind. I still think I mean, your front matter is your structured data in some way. That, that, that's really,

[00:39:05] Matt Biilmann: but

[00:39:06] Slobodan: So

[00:39:06] Matt Biilmann: gone from being

[00:39:07] Slobodan: yeah.

[00:39:07] Matt Biilmann: work with to just some

[00:39:08] Slobodan: Yeah.

[00:39:10] Matt Biilmann: like step ~for, ~for an LLM, right? Like...

[00:39:12] Slobodan: that's really good.

[00:39:13] Matt Biilmann: And I'm just saying like imagine ~you're, ~you're, you extrapolate this from these simple examples, right? Like into larger and larger workflows, and maybe you start connecting like your systems so like your agent is fully connected to where like to all these different sources ~of, ~of context and, and

[00:39:29] Slobodan: Yep.

[00:39:30] Matt Biilmann: content and so on, right?

[00:39:31] Matt Biilmann: Like, and now you extract... instruct them to do a bunch of stuff with that, and those instructions are not structured and the origin's not structured, right? Like, but they can create this on the fly. I think that'll really ~a lot of the Like, a, a, ~a lot of the paradigms around how we think of,

[00:39:47] Slobodan: Mm-hmm.

[00:39:48] Matt Biilmann: of software, right?

[00:39:49] Matt Biilmann: Again, it'll sort of finally change that, like, truth from linear stuff like, okay, the data structure is more important than the code, right? Like, now the context and the flows and your creativity is probably more important than both the data structures and the code.

[00:40:06] Slobodan: Iqbal, I'll just add, the handover between steps does not require data structure, and that, that's really the, the biggest shift. You don't need to have data in a certain format when you pass it off to the next agent and the next agent. And that's kind of great,

[00:40:20] Matt Biilmann: Yeah?

[00:40:20] Slobodan: really. If you start with ~what you, ~what you want to start with and you end the process with exact...

[00:40:25] Slobodan: like a website, a webpage, whatever you wanna work on, who cares about ~those, ~those steps in between ~if, ~if it works? ~I, I, ~I completely agree with that completely. Closing question. Since you ha- You've been pushing this agent experience things for-- thing for, for longer than anyone or almost anyone else on, in the world.

[00:40:46] Slobodan: For anyone listening, no matter what industry their website is in,

[00:40:49] Matt Biilmann: Yeah.

[00:40:50] Slobodan: what is the most important thing they should start thinking about and considering moving forward?

[00:40:55] Matt Biilmann: I think the most important thing is to try to understand, like, are the agents that your users will be familiar with and using,

[00:41:07] Slobodan: Mm-hmm.

[00:41:08] Matt Biilmann: agents already today interfacing ~with your, with, ~with your products, right? Like, because they almost certainly are, right? Like,

[00:41:14] Slobodan: Yep.

[00:41:15] Matt Biilmann: right now, in some areas, it might be a small niche of your users, and in some areas, it might be a lot of your users.

[00:41:21] Matt Biilmann: But regardless, there's bound to already be of people that are working with agents that they care about are not your agents. It's not your chatbot on your website. It's agents that, that they care about and that they have a relation to. about how-- like, what agents are those users using, how is your experience there, right?

[00:41:46] Matt Biilmann: Like, because it's not like... Like I've said many times, AX is not something you add to a product, right? Like, it's not like some new feature or something. Every product has an agent experience because all of these agents, whether it's, like, through computer use or through fetching your website or through working around the barriers you put up from them, have some agent experience right now.

[00:42:08] Matt Biilmann: It's just a question of, is it good or bad? So start by figuring out where you're actually at. ~is it good or bad? ~And what does it take to start, understanding that experience, building a discipline of understanding, like, the agents that use it? Ideally start building evals around it and then improving it. Because I think it's impossible for anyone to imagine that if you have, like, a small amount of people using your product through agents today years from now, that's still gonna be a small amount of people, right? Like...

[00:42:40] Slobodan: Yeah. ~Uh, this is great advice. This is excellent advice. ~Matt I loved, loved this conversation. I'm, I, I'm fully on board. I've been on board since January 28th, 2025. I, I think-- I see that this is the future. I don't think it's the future. I see this happening. I see my analytics. Since I started writing articles on, on the NoHacks website that are targeting agents,

[00:43:00] Matt Biilmann: Yeah.

[00:43:00] Slobodan: my agent user assistant traffic to human traffic is ten to one to 20 to one, depending on the day.

[00:43:06] Slobodan: And, and I, I can see it going up.

[00:43:08] Matt Biilmann: Yeah.

[00:43:08] Slobodan: What is ~the best people for, uh, ~the best way for people to get in touch with you and follow your work?

[00:43:12] Matt Biilmann: Twitter, LinkedIn should be pretty easy to, to find. Blue Sky as well, if that's your thing. In any of those platforms typically where, like, ~feel, ~feel free to, to hit me up, and I'll, I'll, I'll be happy to engage.

[00:43:26] Slobodan: And the links will be in the description. Also, check out netlify.com and netlify.ai. There's a lot to learn from how they are approaching this agentic shift. And to everyone listening, I hope you enjoyed the episode. Please consider rating, reviewing and sharing it, and I will talk to you next week. 

[00:43:39] Oh

[00:44:07] no. Hatch,

[00:44:27] no hatch.

[00:44:32] Go back.


Podcasts we love

Check out these other fine podcasts recommended by us, not an algorithm.

From A to B Artwork

From A to B

Shiva Manjunath
Product for Product Management Artwork

Product for Product Management

Matt Green & Moshe Mikanovsky