Viktor
00:00:00.000
you can think of it this way. The most common interface for people to interact with you is or was a browser like Chrome, and then we were all trying to optimize for Chrome. That was what we were doing, essentially. And search engines. Two things. Now your main interface is an agent, whate- whichever agent it is, there are hundreds of them, right? That's your main interface, and that's what you're optimizing for now. There were some attempts, and I will call them failed here publicly, the kind of, okay, so agents should use browsers, but I think that we are giving up on that is just pointless. So yeah, that's what you're optimizing. It's just the tool you're optimizing for or interface you're optimizing for changed
Darin
00:01:44.450
Right now, somewhere out there, in fact, on your website, you have AI agents hitting it, guaranteed. But the question is, is are they actually getting the data that they're looking for? That's what we're gonna be talking about today. On our devopsparadox.com site, we've tried to make it as agent-friendly as possible because transcripts and blah, blah, blah. Viktor, I think you're still mid-flight on updating your website, correct?
Viktor
00:02:11.695
yeah, I'm mid-flight everything. Life… I'm even in a m- mid-ages kind of. I'm, I'm mid-life everything
Darin
00:02:18.315
But let's think about it. If the site isn't ready for AI agents coming through, they're trying to grab data and they just silently go away because "Oh, I can't find it. I'm gonna, I'm gonna bounce."
Viktor
00:02:32.534
Oh, there's a worse scenario. There is, I'm going to try to check out your site and it's you're blocking me
Darin
00:02:46.686
But interestingly enough, Cloudflare has a website called Is It Agent Ready? Or Am I Agent… What is it? Let me look at my notes again. Is It Agent Ready? That will tell you if your site is agent ready or not. Now, I just ran it right before we started recording, and previously, the devopsparadox.com site had a score of about 70 out of 100. I was like, "Okay that's fine," because there's a whole section about commerce. We don't do commerce on there, so it's, that automatically knocked us down. But they've added in new things since I ran it initially, so our score went down a little bit, but that's okay. But now that we have tools, in theory, starting to show up, especially from Cloudflare, some of the biggest ones that block people welcome to, what was the status code? 429, pay me money. What are we to do? L- let's look at the history of browsing the web for just a moment. The '90s was about just getting browsers going. Do you remember Netscape and
Darin
00:03:49.606
R- so we're just trying to get a browser supported on the web, right? So browser ready. Secondly, we became SEO ready, or the way I say it, sell me all the ads ready, because that was the early 2000s, Google and everything else. AltaVista, Excite.
Viktor
00:04:10.265
sell me all the ads, but also let me be at least on the first pl- page of Google
Darin
00:04:15.355
Exactly. That was the wave. Now the third wave, and again, this is something that Cloudflare may or may not have said, so we'll attribute it to them. If not, we'll take the credit. Now we have to be agent ready because what I expect by using agents today, in fact, I was doing something today. I went to Google first to ask. I was like, "Why am I doing this in Google? Let me just crank up Claude Code and ask the question there." And five minutes later, I was done with it. Claude was done with it
Darin
00:04:46.168
Five minutes. Right. Five min- just doing… it is a long time, but I was writing a bunch of lines of code and configs, and I had to wr- So it was just figuring out what to do versus going to Google and figuring out what to do. I had Claude and we interacted, and it was done. I was fine
Viktor
00:05:02.475
Not only that, but you're increasingly more likely to hit agent or AI in Google as well. So th- there is a chance now, and that chance will be increasing over time, that when you go to google.com and you type something, you also get AI response above the search results. So you must scroll if you really want them until they disappear completely
Darin
00:05:23.165
So going back, browser ready, crawling ready, that was the second one, or ad sales ready, and now we have agents. Why does it matter now? To me, an agent coming to my website is no different than an SEO crawler from the early 2000s, except for one thing: how that data's actually being used. Previously, it was just being indexed. People would search for it and, hey, great, we end up in the top three above the fold, whatever number or however you want to say that. But now that data can actually come into something, and people actually don't have to visit my website. Hint, hint, do you need to have a great website anymore? I don't know
Viktor
00:06:03.310
I'm not sure whether actually the focus of this conversation should be agents even. ' Cause think about it, what do you do with agents, right? When they need to browse the web or stuff like that. very likely for some really up-to-date information, maybe stock, maybe weather, something like that, and that's browsing web just in case there is no service already it is familiar with, right? Which is increasingly less likely. Most of the requests, unless you're doing some really deep analysis or something like that, most of the requests are going to skip agent altogether. Not skip agent, but agent to LLM, LLM answers agent, agent shows you. So I feel it's closer to actually SEO optimization than anything else, but SEO not from the perspective of agents, but how to ensure that when somebody says Spanish paella or how do I prepare Spanish paella or any of the variations of that, model itself actually gives the answer that through is based on my data, not somebody else's data. B- And that's important for two reasons. First is that's the answer you're getting, the initial one, but second, if you do exploration it will not really Goo- literally Google or search for paella. It will search deeper based on the URLs and information it has in the model. So agent goes through the model and say, "Hey, Viktor wants to know in more detail about paella," right? Something that maybe you don't have in your training, which is unlikely, but let's say it is. And then LLM is going to respond, "You know what? Analyze those five sites." So the question is, are you the first answer? And if you're not, are you those five sites? And when you think about it, that's very, very similar to Google Search in the past Except that it's not Google index anymore, it is model training now
Viktor
00:08:06.065
it is preparing for agents, but I meant less about agents browsing the web and more agents talking to LLM and LLM answering
Darin
00:08:15.185
So there's a term for this. It's called AEO. what it stands for, I don't remember. But basically it's been the long tail. People may ask, a two or three word phrase in Google, but in what I've done over the past number of years when I would research videos to do for Jenkins, I would look for questions that were seven to 10 characters long. Really long… It's called long tail. Very long tail questions, because those are really specific, and if I can answer those people's questions, also num- it's a lot easier to get to the top of those, 'cause there aren't people asking those questions every day, but the people that asking them need that information now, Going back to your paella recipe my question would be, "Hey, how can I have a paella without dying because I'm allergic to shellfish?" That's what I need to know
Viktor
00:09:04.440
It's actually going to be much more complicated than that. I'm g- just going to follow that same example earlier, right? And then agent would actually probably answer, "Okay, here's the result, but by the way, which other ingredients do you like or dislike?" And you would do back and forth, right? You would have a conversation, maybe a few minutes. And then that whole concept will be the base of what will be retrieved for you, either directly from LLM or by doing research afterwards. So while in the past you were writing, what did you say? 10 words or whatever the amount is, now you're actually feeding the beast with hundreds or thousands or tens of thousands of text words, paragraphs, text, whatever you want. And all that is taken into the account when you get the res- response. So much bigger context than you were doing before
Darin
00:10:04.007
Search boxes were still capped at a certain amount of characters. Now we effectively have unlimited
Viktor
00:10:10.014
the next search you do, so if you go to Google Search now and say, "I want to cook paella," cool, you, you get some results, right? Say, "I'm allergic to shellfish and but I want to cook paella," it's gone. You would need to be generating that context yourself. Say, "Okay, I'm going to type again, I want to cook paella," and then continue, "And I am allergic to shellfish." So basically you would need to be copying and pasting to get that effect in the past.
Darin
00:10:43.417
every agent that's talking to a model. Let's talk about the four basic parts to make this work. And I've got my notes here. There's discoverability content accessibility, bot access control, and capabilities. Now, discoverability, robots.txt, site XML, sitemap XML, all the normal things. But we're probably not configuring those except for a handful of sites that say block everything. W- why… Okay, I want to step off this for a second. Why would you want to now block agents from your website?
Darin
00:11:22.941
Ever. Why would you ever want to completely block agents from consuming your website?
Viktor
00:11:28.862
Yeah, and I never understood that because there is a much easier solution if you want to block whatever from accessing your site that means that you don't wanna be in public, just make it private that easy. If it's public, you need to be found
Darin
00:11:43.899
it has baffled me for years of, okay, you want to block out maybe certain crawlers, certain every- because you've seen them being not friendly, but come on, it's just a… robots.txt isn't a mandate. It's just a suggestion at best
Viktor
00:12:09.823
Yeah, and you would need some kind of user identification or something like that too. And you need to ensure that actually whomever you're blocking from cannot afford to create the user Oh, I need to pay 10 bucks a month. I- I'm not going to crawl through this thing. Absolutely not. Too expensive."
Darin
00:12:30.114
Friction. Friction. Let's talk about content accessibility. In the past, we were creating HTML pages. Forget about all the, the magic stuff. At the end of the day, it's HTML. That's what we've always had. But now, for consumption, for agents as they're crawling, they prefer Markdown, because Cloudflare's done some research. You have less tokens to parse. If you just give it a plain Markdown file or Toon or fill in the blank, whatever, basically as stripped-down text with just the content possible so it's not having to strip out all the image tags, the HTML, the header, all the stuff, then the consumption into the models is going to be better, and number one, the chances of a parsing going sideways goes way down
Darin
00:13:18.822
we've had a couple of people on that have switched writing their documents from human first to agent first. That doesn't mean they ignore the humans, but they're writing primarily for agents because if consumers of your product or project, if you're working on an open source project, are working on, in any agent, take your pick, it doesn't matter which one, and the models have been out and your projects have been around for a long enough time, the chances of that data being in the model are high. it is at this point in 2026
Darin
00:13:54.435
That's what I'm saying is sometimes you've written really poor HTML that was just unparsable or it was all JavaScript, and trying to figure out how to parse JavaScript as it's dynamically rendering is not useful. thus why we don't have any JavaScript. But what we do like is if we just give it a markdown file, life goes on. Again, you gotta get your website wired up to serve the markdown. if it's a browser, it serves HTML. If it sees that the request is coming in from an agent or it's m- making… And typically what's happening is it's making a request for a text/markdown. That's what most agents are requesting to a day. So if you see that request come in, you need to serve back markdown. If you see requests coming in for text HTML, you're serving back HTML.
Darin
00:14:41.962
It's almost no effort. And here's where there is a little bit of effort, and I will peel back the curtain a little bit, is both your site and the devopsparadox.com site are both Hugo sites, which means they are statically generated.
Darin
00:14:58.839
from Markdown to Markdown. Okay, that's a different… That gets a little weird too. But that means there's no server ru- I mean, there's a server running. There's Apache or some sort of HTML server, HTTP server running to serve the pages, but that's it. So on my side, what I have is a small listener that's capturing the request coming in, and it will do a rewrite based on… There, so it's a separate process, but it knows how to handle that. It's a little bit longer conversation, but it's simple to do. And because we don't get a ton of website traffic, it doesn't cost anything literally to do that Good, bad, or indifferent that's the way it is. So content accessibility is a big deal. If you have documentation for your products or projects, you're probably writing it in Markdown or AsciiDoc, would be one of my two guesses, especially if you're in open source. That's probably what you're writing. If you're in an enterprise, you're probably using some more content management system. Bless your heart. who knows what that's going to be?
Viktor
00:15:57.497
that content management system, excluding some outliers, that's a database essentially. And, querying a database to output Markdown is infinitely easier than any other format you're generating right now
Darin
00:16:11.427
The next item is talking about bot access. It's sort of a, a repeat of the earlier one. So have you set up your robots.txt file correctly? How do you want bots to consume your content? Again, going back to that, if you don't want bots consuming your content, then don't put your content on the internet effectively. just don't. Start shipping your website around to people on three-and-a-half-inch floppies. That will make everything better. Can you still find three-and-a-half-inch floppies? I don't think so. The fourth item was capabilities. This is the… One of the big parts that's in that isitagentready.com website is listing out all the APIs, the MCPs, everything that's available through and for your website. For the podcast website, we don't have any APIs. We don't have any MCPs. Could I think of some now? Probably, the more I start thinking about it, but that just doesn't apply to us. But if you're a SaaS more than likely, if you're a SaaS and you don't have any APIs, what are you doing? But I understand why sometimes you don't. But I've, got people, or SaaSes I've been using that have been adding MCPs like there's no tomorrow, and I was like, "Come on, guys, just get me a good CLI first. Don't spend all your time on the MCP. Just make the API a little more useful."
Viktor
00:17:31.542
I think there is a big distinction in the last thing you said who the audience is. So if your audience are developers, they use agents whose primary, without a doubt, tool is Bash, and that means I can execute any CLI, any command, anything you want, I just need to have it. That's all I need, really. But if the audience are not software engineers, then you definitely need MCP because then we're talking about different types of agents, right? we're talking about Claude Cowork, for example. It's who's going to install your CLI? Come on.
Viktor
00:18:06.442
My mom is not going to understand what you're talking about, right? That's a cooking site, let's say, right? that would be MCP then, definitely.
Darin
00:18:15.582
And that's a great example of pulling in MCPs into CoWork, Notion you name it. If your site has that kind of usability from your users, do it. There's no reason not to. In fact, that's the better solution than listening to a developer whining about, "Why don't I have a CLI?" Because you have to think about who is actually using your product
Viktor
00:18:37.862
Yeah. But, you know, I think that in a very simplistic form, those are all almost irrelevant questions, because the real important question is whether you have a good, and most companies don't, just to be clear, I stress the word good API. ' Cause if you do what's a CLI? Auto-generated something based on the scheme of that API. What is MCP? Again, auto-generated something based on the API. Now many APIs create a mess because they're just silly, but in a very simplistic form that's what it is. The real work is the API. Everything else is just can be auto-generated.
Darin
00:19:16.728
But in order to be able to get to auto-generation though I know you're talking about the CLI, but we should be able to auto-generate our documentation from our code because all the documentation is actually in the code, right?
Viktor
00:19:42.404
to be honest, I don't write docs in code. Actually more than ever, I write less docs in code than ever before And the reason for that is because I let my agents now most of the time actually rip through the code and the docs and the tests and many other things, right? And documentation, excluding manual testing from this story, documentation is probably the part where I put most attention when reviewing and I want it to be separate precisely so that I can easily review it because if it got that wrong and I, when I say wrong, I don't mean you're a missing comma over there. The documentation in a way is a representation of what you did Apart from tests. So I'm, I'm less, documentation and code guy now than I was before. It would be very hard for me to actually figure out what is what
Darin
00:20:34.926
I can appreciate that. But even-- And again, having this conversation now versus having this conversation 12 months from now, I think would be very different because I'm-- there's still hallucinations happening, but not horrible as they were in the past. But I don't know that I would want a fully… I'm gonna use the word that you don't like. I'm gonna-- Not, I would not like a fully autonomous documentation page generated for me, because when I've seen that done, it's a lot better than it was in the past. I'm, I'm not doubting that. But it j- but there's just so much content that's written
Viktor
00:21:20.811
That's one of the reasons why I said the documentation is probably where I put most of my effort now. But think of it this way, right? Isn't it easier if it does the first pass and then you have something to talk about?
Viktor
00:21:35.107
Right? So when, when I say autonomous agent, I mean you did it everything until the very end. I have a PR open, CI passes, code reviews show nothing, and so on and so forth. It's time for me to step in. that's what I mean by autonomous in this case. So, In other words, autonomous in everything but the last step, like merge, right? And in that case, actually I want to have… I'm actually insisting a lot to have a separation. Okay, I'm going to look in tests. Where are the tests? I want to look at code. Where is the code? I want to look at documentation. Where is the documentation? If it's all spaghettified I'm going to have trouble being able to focus on something, whatever that something, the out- the outcome, And that's very different from the past when pr- let's say not twen- twelve months, twenty-four months, two years ago, right? I would insist that documentation is in code for two reasons. First, because it's more likely for me to update it when I'm updating the code, and it's also that's because it's one pla- place, and because it's one place it's easier for me. Okay, so let me write this function and document it right away, it's all there. But I'm not doing that anymore, so now I'm more from the management style. If you're a manager, for your selfish reasons, would you want your team to generate it all spaghettified or separate?
Viktor
00:23:03.008
Oh, now there is no such excuse now anymore. There is no excuse "Oh, we are not doing testing because it occup- it, it takes too much effort," or, "We are not doing the…" No, no, No. Now everything that I can think of is being done now. The only limitation now is my ability to review it
Viktor
00:23:25.378
Yeah that's the limitations of the universe, space and time. I can have X features done on average a day, and if it would be only agent work without me testing, reviewing what's not, I could have infinite number of them. The-- my only limitation, if I wouldn't be stepping in, would be the amount of money I have for tokens. That would be the only limitation, really. And the machine, but that's also money
Darin
00:23:54.366
So let's go this way. You've allowed your agents to go off and write everything, the first pass of documentation, and y- and you've got five of these running a day. There's a 0% chance you would be able to keep up with that documentation validation.
Darin
00:24:13.452
Okay, yes. But I'm saying you're solving five features a day. I'm making up a number right now. You're solving five features a day, but you as a human cannot review all the documentation that needs to be published to the public
Darin
00:24:28.682
So what do we do? Because you could be publishing those five features at the end of the day for consumption, but there's no documentation there yet for people or agents to figure out how to use the new features
Viktor
00:24:41.672
Correct. that's the problem I'm having right now. You're nailing it, man. 5,000 features a day. That's what I really want.
Darin
00:24:50.342
So I'm gonna step back for a minute longer here on this, is we have been creating websites. There have been teams of front-end developers, back-end developers UI/UX something, something, something. We have testers, we have blah, blah, blah, blah. Now I can almost say that I'm gonna go this far. I cannot think of a website that needs to be dynamic anymore. I'm sure there are. Like banking and that kind of stuff. Yeah, I mean, yes, there's things… I could argue that amazon.com, the catalog side of it could be completely static.
Viktor
00:25:27.523
Uh, that depends how you define static. Uh, If you mean like data stored on a disk somewhere, that's just as much Markdown as a database
Darin
00:25:34.803
yes, except I'm not having to go through a database. I'm not having to pull the data from a database through an application server to get it ready to render into something. I've just got an HTML page sitting on disk that comes through a web server
Viktor
00:25:47.564
No, you're having-- You're actually having exactly what you just said, that you're not having, because you're having a markdown that goes through a server that converts it into HTML, and serves it to you
Darin
00:25:58.002
no, no. I'm saying that's already been… The markdown's already been pre-processed into the HTML flat
Darin
00:26:03.764
And there's also still the Markdown is still there, so it's… So I've got the HTML and the Markdown sitting side by side.
Darin
00:26:10.024
So based on what request was made knows what content type's being requested is what I serve back.
Viktor
00:26:16.186
Yeah, for a typical website, you know, recipes and stuff like that without login and any of those things that's the model now, yes. And I think that that's, has nothing to do with AI, just to be clear. I think that, that's how we should have been doing things for a long time, and not for WordPress. Let's put
Darin
00:26:33.663
Thank you. That's where I was getting, getting ready to say, because I don't know that you need WordPress anymore. I apologize to my daughter and everybody else that uses WordPress. But there's no reason for smaller-ish sites to medium sites to open source sites, which most open source sites I see are using fill-in-the-blank Hugo, Jekyll, Docusaurus, uh, all
Viktor
00:27:04.218
Oh, yeah. Yeah, it makes perfect sense actually when you think about it, because how is she going to find out about Jekyll or Hugo? Install it, run it, compile it, whatever you need to do. Basically the, the appeal of WordSite-like sites is I j- I just wanna sign up, give you 10 bucks a month
Viktor
00:27:24.378
and be done with it, right? So it's for non n- non-techies, let's say. It makes sense. It's still, I would argue that still WordSite should be the last one ever considered. It's the first one, and now it should be the last. That's how it works. But for techies, yeah,
Darin
00:27:40.567
but for techies, yeah. We had teams to create all this stuff. Now I've got an agent or two, or a team of agents. I'm gonna use Viktor's agent tech model. You've got four or five agents running at all times, right? Until your tokens run out. Then you're producing these features, or this, in this case, let's focus on the website. You're updating the website constantly. So in this agent deck, if you don't know what agent deck is, go watch Viktor's videos on it. You've got I'm gonna change up what your agents are here, though. You've got a copywriter, you've got a legal analyst, you've got an SEO expert, and you've got just a general IT person, right? IT agent
Darin
00:28:26.102
to help m- And then you've got the manager, the orchestrator o- over the top of those four
Viktor
00:28:30.172
I think that the potentially better way to describe it is that you h- you have three roles, in general speaking, not about my project. You have orchestrator, I know whom to delegate stuff. You have a doer, whatever doer does, it does things. And you have challengers. I'm going to challenge everything you did. That's basically, in real life, that would be me. usually my role, "No, this is bad. This is bad. Do it again." And those challengers, I think are the most important part, right?
Viktor
00:29:03.375
Exactly. That would be the legal, that would be reviewer, that would be whatever you call those roles, but hey I dislike everything you do, man. That's the role
Darin
00:29:14.975
So what should we do next Monday? Let's give everybody the rest of the week off as they're listening to this so they can digest it. What's the first thing that you should be doing? If you're in the middle of a website redesign right now thankfully Flash is no longer a thing, so we don't have to worry about that anymore
Viktor
00:29:34.340
It fl- I was in love with Flash back in the day. Honestly, I loved it. it allowed me to do things that couldn't be imagined before it. Now, I would not touch it with a tenth… What's the word? What's the expression? Tenth, tenth football. But I loved it, man. It's one of my first loves
Darin
00:29:55.179
We'll move on from that. If you're going through a website redesign, okay, if you gotta make it pretty for the humans, do it. But if your site is public and you are trying to sell something to the public, and you're making it hard for agents to consume or just in general crawlers to consume your content, start down that path. Like what we're doing. Publish Markdown right beside the, the HTML based on what they're asking, based on what the consumer is asking for
Viktor
00:30:24.762
you can think of it this way. The most common interface for people to interact with you is or was a browser like Chrome, and then we were all trying to optimize for Chrome. That was what we were doing, essentially. And search engines. Two things. Now your main interface is an agent, whate- whichever agent it is, there, there are hundreds of them, right? That's your main interface, and that's what you're optimizing for now. There were some attempts, and I will call them failed here publicly, the kind of, okay, so agents should use browsers, but I think that we are giving up on that is just pointless. So yeah, that's what you're optimizing. It's just the tool you're optimizing for or interface you're optimizing for changed
Darin
00:31:08.812
The other thing you want to do is make sure your sitemap XMLs are correct. But more importantly than that, there's this thing called schema.org. If you don't know what schema.org is, number one, go look at it. That's all it is, schema.org. And then make sure the pages on your site that matter have proper schema.org definitions. They're JSON, JSON LD or something. I can't remember what the thing is. It's… It looks weird. It's not pleasant to write, but it makes, especially if you've worked with Google Ads, especially for Google Commerce level stuff, you're already doing it. That's, structured data is what it's all about. But start building in that structured data. Again, it's nice to where agents consuming markdown files nice and straightforward. You say, "Hey if you want more information about my API, here's the page you need to look at. If you want more information about my pricing, this is what you want to look at." Does that make it harder for you on the backside to keep track of, okay, what did I tell the agents before of where it is? Because once you make that decision, I'm not saying it can't be changed, but you want to think hard and long before making that change, because you don't want to put agents into redirect loops like what we can do in browsers, like 302 to a 302 to a 302 to a 302 to a 302 until finally we get to, oh, this is a 404 now. That's what we don't want to do
Darin
00:32:33.672
here's something we might want to think about. If, and you've mentioned it already, if you're, especially if it's a SaaS, and you're targeting people that use Cowork or s- variations of that kind of tooling, get your MCP server out there now, even if it's just basic. N- this is another real example. The tool that I use to schedule social posts, they've published, they initially published an API. But then they published their MCP server, which I haven't used yet, but it looks pretty good, and there's no reason why I couldn't just plug that right into Claude Cowork and off I go. How do I want people using my website? How do I want people interacting with my service? If I'm still trying to push people to the website to do their work, I might be leaving a lot of money on the table
Darin
00:33:25.247
it's a beautiful day in the neighborhood. If I never have to write another HTML tag
Viktor
00:33:29.470
we, we're not done with it. It takes time, but we're definitely not done with it. If I exclude Gmail and maybe two more, three more sites, I, I don't use browser anymore. Do you?
Darin
00:33:43.255
I'm just thinking. Yeah, for what we're doing for recording. No it's, pretty much limited to Yeah, that's it. I'll probably go to
Viktor
00:33:53.996
Hey and, and how long will it take until you remove those five? How long will it take when until b- I get my emails in Cowork-like agent?
Darin
00:34:08.026
Ooh. That'll happen before I get rid of websites completely, but that could happen
Viktor
00:34:13.506
Yeah, kind of like, you know, you're working, you're working, and pop, pop, notification. Whole screen is covered. Pop, notification. You got email. You want to open it now, you want to open it later, cool. Oh, the… Look, you just have a meeting in basically Your whole workspace, life, yeah. Basically, if you would make analogy with people, imagine that you're a person very high ranking person in a, company that has a secretary or maybe actually a team of people working for you, and basically what you're doing is that they're giving you reports, and you're reading or scanning or just signing and kind of, "Okay, next. Oh, w- w- Joe, what are you bringing me now? Oh, this. Oh, okay," "Go ahead. Oh, Michael, w- what's happening here? I don't… You're fine," kind of like just, just go do it. Don't even show it to me, right? That's the thing
Darin
00:35:04.036
to extend that, you've got your assistant hanging outside your door and they stay out of your way. They're just doing stuff for you until you need them, and then you summon them to you and it's like, "Hey can you t- Oh yeah, I got it for you." And off they go again, and you don't see them until get a notification saying it's done. What's interesting though is to have this, not Uber app, but Uber app that everything I'm just living in, it feels like a g- a single green screen, that I never have to leave this one application ever to do anything
Viktor
00:35:34.547
Exactly. 'Cause I'm, I'm interacting with my secretaries in this case agents. I'm not doing the actual work myself. I work with-- I interact with my team. That's what I do, right? I'm a CEO
Viktor
00:35:49.347
Something like that, right? So I, I have people doing stuff for me. I'm coordinating those people, and they-- sometimes I call them in. Sometimes they come to me and say, "Hey, Viktor, you just got an important email. I think you should look at it right now." Okay If you say so. ' Cause you know what happens if you're wrong? I'm gonna replace you
Viktor
00:36:09.657
I'm gonna improve you. I'm going to rewire your brain, man. That's what I'm gonna… You're going to get a loboto- lobotomy and we're going to start over. That's what's happening
Viktor
00:36:27.277
it's a very different thing, right? It's… if you look at it from optimization perspective only, maybe SEO, whatever point, whatever. But this is a completely different ballgame. This is we were just browsing information before. Now we are interacting with the intermediary that is doing the work for us
Darin
00:36:47.427
The other argument might be won't agents just figure out how to read HTML finally?
Viktor
00:36:52.287
Didn't we cover in one of the Thursday streams how somebody claimed with absolute certainty that HTML is the thing that should replace Markdown, and I somehow laughed at it?
Darin
00:37:12.900
You could argue that XML could be because it's more structured than HTML done correctly is compact and useful. I can count on one hand the number of sites that are that good. It's that many. By the way, if you're listening, there's no fingers being held up
Viktor
00:37:30.485
Let's say you have pure text in both cases, right? And you have italic. What is that in Markdown? Two asterisks, two char-- two characters. What's in HTML? It's more than two
Viktor
00:37:42.046
there you go. You have a title that's three characters if you count space. Hashtag, I'm giving you the headline two here, right? Few characters in HTML it's the same or more. So it's, it's not more compact. That no. And there is all the other garbage like HTML tags around and so on and so forth. It's not more compact
Darin
00:38:04.946
Who's going to win and lose in this? If you don't let's say as an open source project or as a commercial software product If you're not writing for agents first today or very soon, sure, the agents will probably figure it out as it gets pumped into the models, but is that going to be good enough for you to where… 'Cause what we want is when we're sitting there typing in our agent, fill-in-the-blank agent, it doesn't matter which one, and we get an answer back, and I'm, I'm saying, "Hey, I need to know exactly which is the best library to use with Go and show me the documentation." Those are the kinds of questions that I ask cause I don't I don't wanna take it from your knowledge within the model. I want you to show me right now what the best one is with proof
Viktor
00:38:57.861
So it, it will still come up fr-from knowledge, at least at the beginning, just to be 100% clear. Then it might extend its knowledge through live queries. It always starts with knowledge
Darin
00:39:09.408
Oh, of course. Which is what I would expect, but if I'm, if I add in the extra thing to where it's like, "Hey, make sure it's true at this moment."
Viktor
00:39:18.038
Yeah, exactly. Exactly. Then it would, it probably still select what it knows, and then it will go to get up-to-date information about things it thinks it knows.
Darin
00:39:26.930
so the next step to that is if you're still only producing JavaScript-driven sites, it's game over for you, in my opinion. It just, It just is. I mean, humans don't even like them, much less agents
Viktor
00:39:48.237
Yeah, the same way. Exactly ' Darin: Cause that's how we figured out how to do things Yeah. I mean, Actually I would right-click view source, copy paste, and then start changing it to understand what it really does
Darin
00:40:02.825
I don't know, man. I don't know where this is going to land. There's so much to do and so little time to do it in. But that used to be the real argument, but the argument is now, okay, we've got too much stuff being done for us and we don't have enough time to process it ourselves
Viktor
00:40:19.302
Yeah. And we are not actually very good as people just as agents at long-term memory. You probably already forgot how your life was a year ago, and you're not aware that actually two hours in and you're already done for the day. If everything would stay, everything else would be the same, the scope
Darin
00:40:45.264
Yeah, 'cause I can get things done a lot faster. The problem with being done in two hours, though, is I'm also as tired as I was after eight hours in two hours, but I still have six more hours to go
Viktor
00:40:50.484
No, I mean, if your scope would be the same, you have six more hours to rest from that tiredness
Viktor
00:40:57.586
I never, ever, even when I was young and I had nothing better to do, I never played more games than now Really, I work for an hour or two and then off we go. Now I'm playing Wuchang for an hour for my brain to clear. The moment I kill more than 1,000 of creatures, I'm ready. My brain is ready
Viktor
00:41:19.875
Oh, 1,000. I k- I killed that. Yeah, exactly, in five seconds. Yeah. I've been getting those limits left and right. Sometimes the l- limits of how much account my subscription allows very often I… Recently I see a lot "Hey, I left you work for the night," and then, then I go and check and if nothing is done, what's happening? "Oh you, you just hit our internal limit," "Man, you're going crazy. Stop, Stop that. You, You have enough tokens, but we are not gonna let you continue."
Darin
00:41:48.413
If you haven't gotten your site ready for agents yet, now is the time. I guess that's the bottom line, is if you want to show up, just like how you wanted to show up in the search engines, and you still want to show up in the search engines, uh, you've got to be showing up in the models as well. And what does it take to get there? Check out the Cloudflare site, isitagentready.com. Pop in your website, look at it. It will give you a score. You'll be very sad. I can already tell you that, just because practically nobody's doing it correctly. And the, and this is, again, Cloudflare's angle of it. Does Cloudflare have a, have some, uh, skin in this game? Yes. Are you kidding?
Darin
00:42:29.033
They blocked the agents, and now they're willing to make sure that the agents are passable
Darin
00:42:45.777
maybe the first step is to go ahead and make sure that you're able to actually publish multiple versions of your content at one time. If you can't, that's going to be your first thing. It's like, okay, can I publish Markdown at the same time? And then you start figuring out how to wire in your robots.txt and all the other things that you're going to be learning about, your LLMs.txt, your LLMs-full.txt. Yes, there's a whole lot of new things you can do as a web developer that just boils down to, I don't have to lay anything out. Hashtag space, here's my title, and start writing. That's all we want. So what do you think? Head over to the Slack workspace, go over to the podcast channel, and leave your comments there.