The Agentic Web Is Coming – Is WordPress Ready? | Miriam Schwab

Okay. Hi everyone. Good job still being here after four days. Yeah, you do deserve applause. We all deserve applause. Um, and thank you for coming to my talk. So, we're all thinking about obviously AI. I think 99% of the talks here were about AI and agents and the agentic web and what that means for us. Where is our industry heading overall? Where is WordPress heading and how can we be ready? So, we're going to discuss that today. So, first of all, let's just look at a few numbers. So 507% of web traffic is now bots and not human. For a long time a lot of the traffic coming to websites was not human thanks to crawlers like Google and things like that. Um but that number is increasing rapidly and a lot of it is agents. I do want to note that that agent traffic is just training pretty much. It's not necessarily interacting or any or anything like that. It's it's training on the web pages but it is increasing. Um in terms of Google searches 68% of US Google searches ended without a click. I think uh a lot of us are feeling that people are getting the information they need including us by the way. Right? I think on the one hand we complain why aren't people getting to our sites and on the other hand we too appreciate the experience where we don't have to like browse through 10 sites to get an answer. But what it means is uh people aren't necessarily clicking through to websites to visit them and they're getting enough information from Google on the Google search page. Um one so 1 billion Google AI mode users that's number is growing. That's uh Google's well AI mode. Um and that's how many people are using it and it's uh growing uh also very actively. Um, and the query growth rate is doubling in terms of AI uh queries. And of course, well actually it's less than 41% as of today, but 40.7% of the web runs on WordPress. So whatever is happening on the web impacts us and we obviously impact the web in terms of the large scale of our ecosystem. So we're going to talk about what it means to be agent ready. Um, agents are I saw someone said that uh whoever did the marketing for the word agents is brilliant because it's just a fancier way of saying bots in many regards. So agents are kind of like bots and they're running around the web and they're gathering information and that includes on our sites. So what does it mean to be ready for these agents? It comes down to kind of three categories. One is read that they should be able to easily read the content on our sites and uh access it. Query. So that's not just about reading it but also um and not just digesting it but also like getting specific information out of it and acting on it. They eventually will be able to take action on our sites and in many cases will be taking action instead of uh humans. But first, let's just take a step back and say ask this question. Why should I bother making my site agent ready if it just means that AI will be stealing my stuff? And that is a very legitimate and reasonable question. Um because with these AI agents basically training on our data and then spitting out that data uh in when people search or ask the questions and not sending people to our sites, it can seem like it's detrimental to us. So why should we cooperate? So one is relevance. The the horse is out of the gate or whatever the saying is like this is the world that we live in and the way that people will find your site if they do is going to be increasingly through AI and we want to appear there and be relevant there. And it's also not all or nothing. You don't have to just give up all your content and the tell the agents take it and good luck with it. There are different ways to create specifications that tell agents what you do allow them to do with your content and what you don't want them to do. So like they can search, but you can also tell them not to train, right? You can have a way to indicate that to the agents and the bots that are uh accessing your site. So being agent ready doesn't mean it's open season. Um it does mean that you can be the one setting the terms and it also means in this AI world that it's either that or being invisible. And I think we all have websites or build websites because we want people to find whatever it is that we're creating and publishing there. um whether it's because it's useful information or it's related to our businesses, we definitely don't want to be invisible. So, this is something you obviously all can consider and it is also legitimate to decide that you prefer to block all agents and there are ways to do that. There are even initiatives to try to create monetization around having people access your site, not a people, agents. So, like Cloudflare has something like that. Um, I'm of the opinion that I want agents to totally interact with my content and have access to it and digest it and make it part of their models so that when people want to know about me or topics that I write about that my site will be one of the resources that the LLMs uh deliver to them. So what what technically makes a site agent ready? talked about the categories of agent readiness, but how do we achieve that? What can we technically do to make our sites agent ready? So, my favorite resource for this is this site which is more than a site. It's also an MCP um which is specific the website specification and is at the address specification.website. Uh this resource was created by Yostast Devalk who is the founder of Yoast uh the SEO plugin that I think we all know and this site is consistently maintained. What it does is it browses the like leading resources with regards to website specifications both related to AI and otherwise and is constantly updated with checklists with recommendations with statuses whether something is required or recommended or you know in test mode or whatever. And what's great about it because it's an MCP based site is you don't have to sit there and read through it. you can send your agents to read through it for you and determine what can help you improve your site. And I'm going to explain a bit about that process. So on his on this website, he has categories of the types of things that you can do to your site to improve it. And one of the categories is agent readiness. So the agent readiness category includes things like this. So, stable URL, structured data, JSON,ld, semantic, HTML, robots.ext. A lot of this will sound familiar to you in terms of it just being stuff we should have been doing all along or we have been doing. Um, but it's also very relevant to making your site agent ready. Machine readable fields, MCP server endpoints. So, like an LMS.ext. So, there are definitely things here that are new. Um, so you have 21 these are there's 21 points in the agent readiness category as of today. Um, like I said, that site is constantly iterating and updating itself based on any developments. There's 21 things in there. So in the past, if we wanted to optimize our site for something, we would go through it one by one and develop it ourselves or find plugins that could do it for us or whatever. Um the way I see it is that it becomes a it can be a kind of loop that we implement that constantly checks our site against the recommendations or requirements identifies where there's a gap and the agents can help us actually make the changes and implement them to improve our site. So this is what the loop looks like for me. First my agent and by the way when I say agent I mean Claude. Claude's my bestie and uh I probably talk more to Claude than humans in my life, which I'm sure is a problem, but that's that's how it is right now. So Claude, but you could of course use any agent. Claude reads the standard on the site and digests the checklist because it's machine readable. Great. Then Claude is connected to my WordPress site through the MCP um adapter, which is freely available as a plugin. and it looks through my site and its structure analyzes it and says you're good or here are some things that you haven't implemented on your site yet, right? And it finds the gap and then it can actually fix it. So there is a level of direction that we need to give the agents. You can't accept everything that the agent tells you point blank and just assume that it's always correct. You do need to review it, but essentially this is a loop that can be going on a regular basis to constantly iterate and improve and optimize your site. And the fixing can happen from the agent as well depending on how you set up your workflow with your WordPress site. But the way I have it set up and um you know everyone works differently is through the MCP adapter and having abilities on my site which I'll talk about um Claude can actually implement all the changes for me. So all I have to do is kind of just manage the process and I can make sure that my my site is constantly improving. So this is what the routine looks like for me in Claude. So um I'm I'm not going to assume levels of knowledge here. So I'm just I might be sharing things that you already know but just in case some people don't. So Claude uh code claude has two parts. Claude co-work and claude code. I actually only work in cloud code now even when I'm not doing coding related things. I find it to have more precision that I want. So claude code has something called routines. Cloud co-work has scheduled tasks I think it's called. And you can set up a routine to happen on a regular basis. Um, any kind of routine. It could be related to coding and it could be related to your social media posting schedule. Um, this one the it's uh I had to iterate it a bit to get it right. And every so often I'll notice it's not doing something quite right and I'll iterate it. That's also the thing with working with AI uh on your site or otherwise. It's almost never set it and forget it. There's always something in the workflows that you set up that will probably need uh fixing or improving. But anyways, this routine reads that webight specification, analyzes my site, compares it, and then gives me a report of things that I could change or fix. And it also recommends which ones should get attention based on a number of factors. Not just whether in the checklist it's recommended or required, but also at this point Claude knows my website and me very well. So my site's just a content site. I don't necessarily need to spend the time doing some very like highlevel fancy things that won't bring value. So this is like what one of the reports looks like after it ran. Uh this was from I think yesterday and it tells me I checked and here's some things that I recommend or that you shouldn't do etc. So I described this whole this loop and this routine and the reason that it can work is because since WordPress 7, we actually are pretty well set up for this type of workflow. The infrastructure that was implemented in 7.0. Um it's very functional and I would even describe it kind of as basic. I hope nobody's offended by that, but um it actually can be leveraged to do a lot of things even in its current form. And I'm sure as time goes on, we're going to see it develop further and further and give us more capabilities and give non-developers or people who don't feel comfortable, you know, swimming around in claude more UI interaction with AI within WordPress. But even at in its current state with some a bit of effort and uh just learning the ropes you can do a lot with it. So what is the plumbing? The abilities API. I mentioned abilities that my site has abilities in it. Um and that those are tools that say what an LLM can have a site do or can do to a site out of the box. WordPress has like three abilities and they're very basic and boring. So, that can be frustrating in terms of our experience with AI and WordPress because if you just upgrade to 7.0, which we probably have all done, and maybe even um installed the MCP adapter, you're still going to hit a a wall in terms of what you can do. You need the abilities in the site, and I'll explain a bit about that. But once you have a decent set of abilities in your site and the MCP adapter, you really can do a lot. Um, obviously connected to your uh LLM of choice, which Claude's Claude's the you know, Claude's the guy, but you of course can use whatever you want. Um, I grade out the third one, which is the WPAI client because it's pretty simplistic. It's like um click x get y in your site like uh create an alt tag for an image or things like that. Um it's cute but it's like doesn't really give you like I would almost call it like the superpowers that the abilities API and the MCP adapter give you. So it is the third component but at the moment it's doesn't really offer that much. I mean it depends on your workflow but for this type of workflow it doesn't offer much. So what are some abilities you could have in your site? So, basically what I did, and when I say I, I mean me and Claude, I created a plugin for my site called Miriam Schwab's abilities. Great name. And that also has been iterated over time where I made sure that it had abilities built in that allowed me to do everything I want to do on my site from within Claude code. And when I found it lacking something, I discussed it with Claude and we created a new ability. At a certain point, I also wrapped the the WordPress REST API as an ability because there's a limit to the types of abilities like tech like specifically abilities you can create, but the rest API API gives you a lot more. I just don't want to be limited with what I can do with my site ever. I find that very frustrating. So, I created this plugin and it adds all these abilities uh to my site. Um I have another example of ability here. You can see Angie. So Angie is Elementor's agentic AI for WordPress. It's a separate plugin that you can install from the repo for free. And very recently and like a I think the last release of Angie, we added a lot of abilities to Angie. So you can go my approach of creating a plugin and having abilities in it. Actually, there are uh products out there um Angie being one of them where you can just install that and you already get a lot of abilities that you will be able to connect to your MCP. Um so what does the MCP do plus the abilities? It creates this kind of its own loop of an AI coding agent connecting with the MCP adapter um and making changes to WordPress core or to my WordPress site. Okay, so now another question. First we said should I really be agent ready? Isn't AI just going to take all my stuff? And now we're going to say, wait a second, we're going to let an agent loose on our site. That sounds scary. And that is true. Um, letting an agent loose on your site gives you a lot of power. But also, as we know, AI is non-deterministic, right? I can give it instructions and it will output this A and then I can give the same instructions, it will output B. It's um it's not exactly like if I do X Y will always happen and you can put a lot of guard rails in place like a lot and even then you have to keep an eye on the agents because um they can make their own decisions which aren't always in line with what you want them to do. But there are the guard rails and they get better all the time and you can tweak them. So these are some guardrails that I have in place on my site. So for example, it can't delete users. It can't delete plugins, it can install plugins, it can read information. So it has a lot of read permissions, but very careful with anything that has to do with deleting, for example, um, and things like that. So you so and abilities, the WordPress abilities, part of the configuration structure is that you can define them as uh, destructive. So if there's an ability that you give but it's like a dangerous one, part of the configuration is defining it as destructive and then that lets the agent know that it needs to be careful around it. So it is a it is like a you know a tossup of do I want all th those capabilities and I want them smoothly but I am giving up a tiny bit of control or not and also people who don't feel comfortable with that totally makes sense. Um, I find it very fun to be able to work with WordPress in such a smooth way and not be held back. I can bring all my ideas to fruition without any problem. And in my experience until now, um, Claude only once deleted something that shouldn't have deleted. Definitely always back up your stuff. That that saved me. So, yeah. So, okay. So, in terms of being agent ready and technically what that means, I'm sure you've heard a lot about markdown files. Markdown files is a way to give like uh structure and formatting to pages without it being full of too much stuff. Um, you know, it's it's very like clean. And the benefit of markdown files for agents is that it just is really easy for it to digest. It doesn't have to navigate all sorts of formatting stuff, styling, extra stuff that's around it. it just gets straight to the to the meat of the content. So, you'll see a lot of people recommending that you create a markdown version of every page or post on your site, and there are different ways to do it. Um, there are a lot of plugins already in the WordPress plug-in repo that allow you to do that. Cloudflare has a functionality that does that automatically for you if you want. And aside from it making it really easy to read, it also really reduces the size of the page source. So this is a blog post on my my personal site mirrammwab.me. I write a lot there about AI and WordPress. So if you want to learn some stuff, you can go read my post. And that's what the AHTML page looks like. That was long, right? But then I have a reference to my markdown file there. And this is the same page in markdown. Tada. So, it took away all the surrounding stuff, gave the uh gives the agent the main content on the site and also if you remember from the SEO days where there were there's crawl budgets, right? We're still in the SEO days, but Google has crawl budgets. Google is not doesn't want to crawl tons and tons of stuff that doesn't add value. So, the same here, agents also have to use resources to to train and whatever. And the less it has to train on, the faster and more efficient it can be. And the theory is that that will make it more likely to bring your content into its model and then refer it. So in that example, it was the same post, the same words, but 15 times less for the LLM to read. Another thing that I technically did to my site that I was testing out is, you know, I talked about there's three categories to agent ready, read, query, and act. So, what is ACT? A very obvious example is e-commerce. Find me shoes, buy the shoes, right? My site's not e-commerce. I was like, "Okay, how can I test making my site ready for agents to act on?" I have a contact form. I'm going to make my contact form super duper agent friendly, so they're going to love filling it out. Um, I have a post on my site about how me and Claude made it agent ready. Um, it has an endpoint API. It's also ready for WebMCP which is another specification related to agents while working in the browser. So I did both and it's still human ready. One of the advantages of going through this process was that it actually also made my form more accessible to people. Um, and that overall is something that I'm seeing with making sites agent ready, which is the more human accessible it is, the more machine readable it is, the better it is for agents. So that was just like a an added benefit to to doing this. Um, it also removed the JavaScript uh requirement from it, etc. So I added that and I've done some testing on it which I'll share with you. But okay, we've done a bunch of things to our sites. I gave two examples. markdown versions of a page made my form uh agent ready. How can I determine how well I'm doing in this space? So, there's a tool also created by Cloudflare called is it agent ready? And here is my site score um before I implemented changes. By the way, this is actually when I really started implementing the changes, my score was 51. So, but I wasn't able to undo everything I had done for this talk. So, I but I undid a lot of it to see what the score would be, but really it was 51. And after making my site agent ready according to these specifications with that loop that I showed you, I got up to 86. The reason it's not higher is because there's some crazy things in there related to like DNS records and whatever. And I was just not going there. But that's a very nice increase in terms of the scoring. But there's always but. Um that metric from my perspective is a vanity metric. And there's a few reasons for it. One is in the agent readiness checklist that I showed you, there's 21 items and only one is required. And that one that's required is stable URLs. Obviously, we should have stable URLs. So if your URLs are stable, you're agent ready. The flip side of it is that the agents aren't waiting around for all of us to create or add all this functionality to our sites and make them ready for them. If they did that, they still wouldn't be able to access most of the web. Agents, our AI, and very smart. So, they figured out their own way to fetch the content and read it in a way that works for them. They don't need all our fancy little things. And I'll when I discovered this I was like oh great so much for like that scoring and being agent ready like they actually really don't need it. So I ran um some tests through cloud code in the terminal and I noticed that it said to me uh okay I'm going to use web fetch to get the content. I was like what's web fetch? And it starts reading pages on my site and it doesn't care. Look at the URLs that it's looking at. None of them say MD at the end. It did not care that I have MD files there. And by the way, I have references to my MD files like in four different places on my site according to the specification. It's in an LMS.ext. It's in robots.ext. I think it's in a header. It's like everywhere. And Claude Code was like, I couldn't care less. I'm just going to go straight to the pages that you are interested in me checking out. Um then I tried to get it to fill in my fancy agent ready form. I was like this is perfect for you. So I said Claude reach out to Miriam Schwab through her form and ask her about speaking opportunities and I had posted the post already about how I made my form agent ready. So even though I also had indications throughout the site according to specifications saying hey LLM this site this form is agent ready use it the way you should as an agent. The only reason it knew it could figure out that I had this agent ready form was because my blog post. It did not look at any of the other signals. And even then when I said to it okay fine you found my I I was trying to get it to find the endpoint itself. I was like okay fine you found the endpoint. thanks to that blog post fill it out like an agent and it just filled it out like through the front end. It didn't need it. So, um this is just a bit about uh Claude's web fetch tool. Um, but basically what I learned, so if claw does figure out that there are markdown files on the site or through, let's say through the cloudflare functionality that I mentioned, it will probably digest that, but it doesn't need it. Within like milliseconds, it fetches the HTML of a page and converts it to markdown itself. And it converts it to markdown in the way that it wants it to be. So it will focus the markdown to answer the query that it has been sent out to answer. Um so it just doesn't need it. Um I said did you notice that my site offers markdown files which it should have noticed. No I didn't notice that. And yeah I just the only reason I kind of noticed the the API is the API catalog. Um, so when it did fill in the form as an agent, it's kind of cool because it is a responsible agent and it indicates in the form submission, this is what I got, that it was Claude and it's an agent that filled it in and not a human. So theoretically, if we all catch up to each other, then this is a great way to know who is interacting with your site. Um, and it's because I gave it instructions in the API. If you fill it out and you're an agent, make sure you tell me you're an agent. Um, but mostly it just filled in the form itself. So even though my form endpoint was exposed in many places, the LMS just ignored it. So what did I do? I was I realized because Claude noticed my blog post very quickly. I didn't send it to my blog page. I sent it to my homepage. But right away it like was like, "Oh, you have a post." Like it was like reading around, "Oh, you have a post about uh agent ready forms. Oh, so you have an agent ready form." What I did was on my contact page just on the front end, I was I said, "Hey agents, if you're an agent, this form is agent ready. You can fill it out that way." But that's so inelegant. that Claude and probably other LLMs take the actual content on our site and not all these fancy signals that we're giving it most seriously. That's where it finds this information. That's what it's looking for. It's not looking for what's in our cute little LMS text file or our MD files. It's just reading our sites. It's obsessed with reading our sites. Um, and that's where it's getting its information from. So the Asia readiness checklist 21 things but like I said only one thing is required but there is an advantage to making your site uh agent ready which I mentioned which is that really a lot of it has to do with building it according to best practices which really also is just nothing is new under the sun that has been the guidelines for SEO friendly websites for like 20 years and here we are Then there's not really fancy. The fancy things are nice, but at the moment at least they don't make an impact. What does make an impact is building your site according to best practices. So like I mentioned, having stable URLs, duh. Um, clean semantic code, right? Proper H1 tags and proper page structure, the cleanest you can do it, etc. That makes your site agent ready. Structured data can also help. Again, it's just what's in the HTML, which anyways, Claude is going to convert to markdown. Um, and clean markup. So, essentially, bots want what humans want. They want easily readable, accessible sites that make sense, and that's helps them digest it. Agent readiness is human readiness. So, I had a com when I like realized that the LLMs don't care about all this fancy stuff. I uh had a bit of a crisis and I reached out to like some of the leading SEO people and some of the people who are like most engaged in making the web agent ready and yeah they said the pages that win in agents answers are the ones that are easy to fetch, easy to parse and easy to trust. um strategic alignment. Most of these also benefit traditional SEO and accessibility. Again, Asian readiness uh rarely conflicts with the rest of the spec. It just raises priority. So, a mistake that well I was making and I think a lot of people can make is that we treat agents differently. Um and we think that we should write for the model rather than for the human reader. And this was the same mistakes people would make with regards to SEO. And it's the same thing over again. Good content loves content as I saw and make sure that the agent can read it like uh just make sure the humans can read it so that um agents can read it well. Um do I have a slide on the No. Okay. So um those website specifications uh site that I showed you, it has a lot of other categories and they're all about making your site the best it can be. So I actually have a separate routine that runs for for me in cloud code which is always analyze my site to make sure that it is optimized and kind of achieving perfection from a best practices point of view. And I run that routine and I fix things on a regular basis. Um which now I think also has a major impact on whether the site is actually agent ready. But does that mean that we should ignore all these things? No. These are things that are in progress being worked on um being tested. We have some way to go before uh the big models and let's say the browsers uh all agree on certain specifications that will become required or very very highly recommended. But at the moment it's all pretty theoretical. But it's worth it to keep an eye on it so that you know what's going on. And some of it is just like you might as well. it doesn't hurt. So that if one fine day they all flip a switch and say now it really matters, you could be ready. Um so some resources for that. So is it agent ready? The nice thing about the is it agent ready site is that it uh will tell you what's missing and give you prompts that you can feed back into your LLM of choice and then improve it. Bing web master tools has um pretty good analysis of AI and agent traffic to your site. Uh I just discovered that. And page speed insights. So Google's page speed insights now has a category related to how agent ready your site is. It feels like they put it in there because they had to because everyone talking about agents. It only has like three things that matter. So and they're pretty easy to achieve and then you get a three out of three. But um I'm sure that that will be expanded as we go along as well. So let's talk about is WordPress ready? Yes, WordPress is actually ready now to be agent ready thanks to those core the core plumbing that allows us to consistently iterate and improve our sites to make it better for the web, better for people and then as a result better for agents. But our ecosystem isn't ready. So, I mentioned abilities, which are the tools on a site that are what enable the LLM to do stuff on your site via the MCP adapter. And like I said, out of the box, there's like two or three and they're very boring. So, my site is ridiculous. It has 81 abilities on it. Um, only six are from core and 75 are from plugins. A lot of those plugins are mine. I have been creating plugins for my site and I make sure that all of them have abilities. Why? because I can and I might as well. Um, but if you look at it from an ecosystem perspective of the abilities on my own site, and this is just this is not representative all sites, but I think if you would turn on um the abilities, I can't remember what it's called. It's in the WordPress menu. If you've turned on the AI plugin, you can turn this on and it will tell you how many abilities on your site. You will probably see that it doesn't have a lot of abilities and a lot of them are not from plugins if if they are there at all. So I have like 40 abilities that I created. I have Angie installed. So it's another 30. But of the other plugins I have on my site, so I mentioned some are mine, but some are widely used popular plugins. Um there's only like two or okay, eight two other abilities um that are from Yoast, right? Yos has two and then like a bunch from the other ones, but it's very minimal. For some reason, our ecosystem isn't exactly jumping on the bandwagon of making their products agent ready. Um, but it is very doable and then people can manage their plug the plugins as well through their LLMs, which is what I do. So, I mentioned that I built a bunch of plugins and you don't have to build this from scratch. I already put in a lot of work. So, if you go to the to mywab.me/plugins, you'll see the plugins that I've developed. the two that matter. Some of them are just for like fun or actually I have one that um does ongoing uh speed optimize uh speed analysis and then um gives you ways to feed it back into your LM. But that's well it's better for humans so it's probably better for agents too. Anyways, the two that uh are related to this talk is what there's one um called make my site agent ready and that that's you know I showed you the score of that my site uh had a score before uh adding all this of 71. Um the the features that took it up to 86 are in that plugin. Like basically I turned off the plugin, ran it, then I turned it back on and ran it again. So make my site Asian ready is in there and then there's the other one that's called MS for mammwab abilities or something like that and that has a lot of abilities in it. I will just give a caveat that I at the moment are the only user of my plugins. So I don't take responsibility for them. Maybe install them first in like a staging or local site and see what they do for you. And also if you have feedback please give it to me. I would be very happy for other people to to use them if it benefits them. Um and then the other direction is don't use Miriam's you know un not widely tested plugins. Uh you can use Angie um which like I said has a lot of abilities uh already in place. So where are we in terms of our industry our ecosystem and our timeline. So WordPress is 23 years old, which is pretty old for software. But I think AI brings really fresh, exciting opportunities to our ecosystem that like I showed you, we're not taking advantage of and they really give us a lot of capabilities. So, we're at a green field stage again where you can be first movers in the AI space by creating AI functionality in your existing products or creating AI innovation for the WordPress ecosystem. There's a lot to do here. I mean, just me and Claude, we created a bunch of plugins just for my own use and that was just, you know, pretty simple. And there's a lot that we can do here that will not only keep WordPress relevant, but also potentially help it lead in this new AI era. WordPress has a lot of advantages in terms of being open source. The LLMs know WordPress really well because it's open source. It's been learning the code um and the infrastructure, you know, these past years and knows it really well thanks to all of us creating all of this code and all the content around it. Um, it has a really good CMS structure that just works well. The plugins, you know, I I base a lot of adding functionality through plugins and it works well, the themes, etc. Um, so there's a lot to work with here. So, embrace your abilities and yeah, do stuff with AI, get your hands dirty. It's really fun. um obviously back up test on staging and local, but uh if you allow yourself, there's a lot that you can do and there's a lot that you can bring to the ecosystem for all of us to benefit from as well. And hopefully, you know, we can see really cool things emerge from from WordPress. So, thank you very much. We have a few minutes if anyone wants to ask questions. There's microphones. >> Hey, Miriam. Okay. >> Uh I also use uh is your site agent ready? So the the thing that you said about that being a vanity metric is like >> like I'm now like having to rethink everything. Um but >> I mean that's just my that's my opinion and like what I learned through this. >> No, I I I agree because I I've seen the same thing with the with the Cloudflare markdown files and stuff and it it Yeah, Claude doesn't care. And so like that's I mean I the only thing I can think of for that is like I think it makes a difference I guess. Um but no one of the things that you were saying about like having content on the page for the purpose of having Claude read it made me think of screen reader text and like the practice of like you know wrapping screen reader text in like a span or something so it's hidden but like it's still there on the page. And I'm wondering like like it's just an idea that occurred to me like well we could do that for AI I guess too like I don't know thoughts. >> No it's actually interesting. Um if if I am correct that the front-end text like matters that much which is what I saw then maybe we start figuring out ways to make sure that there's like these important messages for agents that they have like a home on the front end. The reason I'm not saying on uh like in the source code is because one thing that I saw is that when the agents convert the pages to markdown, they skip everything. Like they they strip out headers. They strip everything out. So like you think you're putting stuff in there that they'll care about, but they actually just skip over it to get right to the content. So it might have to stay in the main content in some way. That could be actually interesting and cool where it's like >> sidebar. Hey, >> at least a little bit better than uh than having it on your on your page where it's like, you know, >> yeah, >> agents read this section, you know. >> Yeah. Yeah. >> Yeah. >> It's interesting. I wonder if we'll see that, but it makes sense. Thank you for your thoughts. >> Hi. >> I had a comment more than a question. and I'd like to hear what you think about it. Um, spoken to a lot of people where everybody's talking about AI. Everybody's saying where is it going to be in the next five years. Uh, we've been to multiple different sessions where they're telling us how we need to layer AI into our pipelines and our layouts with all these specific things. And you just kind of cut the head off of all that, which actually I really like. >> Okay. >> And you made this amazing point. what's good for humans is good for robots. How do you see this progression playing out in the future? Like obviously it's going to be a little bit of both, but >> yeah, for sure. >> Do you think it this thing is going to keep rolling in the direction of what is good for robots is good for humans and vice versa. So my theory at this moment of time and of course as we know in this AI era things shift like can change a lot from day to day is that first of all it's going to take a long time for all the big organizations to agree on stuff that we all should kind of get behind for making things agent ready. So because of that and like I said agents aren't waiting around for us all to agree that we should have this kind of functionality on our site. They just work with what they have, which is our messy, colorful, weird internet, and they make the best of it. So, I there's not ex I don't think there's exactly an urgency to make all of that required. Also, if it becomes required, that puts a lot of websites um in a weak stance because are disadvantaged because a lot of sites aren't run by people who are reading AI news every day, right? And are going to be like, oh, like I have a bakery. Hey, Jim, you know, after you make the bagels, we need to add LM. Like, it's not going to happen, right? Um, and that would mean if if something becomes required like that, then a lot of the internet will just not be agent readable. So that also seems crazy. I think what we're going to see is better and better AI tooling that integrates with sites that help us make them better. The web MCP angle is like an example of that, although that's also going to take a while to get approved or anything like that. Um, or even for Chrome to have the So Web MCP is like uh if you're in the browser, uh, you can have an agent in there. um doing stuff, but it's a Google specification and Chrome itself still doesn't have a like a default flag on for it. So, I don't know. So, I think the direction of human ready um keeping an eye on things, but like human ready, that's going to be the main thing. Uh but using AI from the other side, which is to consistently make your site better and better and built according to best practices. I think that what might we might see sooner is um interactive agentic web for commerce uh because that's where the money is and so that generally motivates people to do that and people are definitely going to be doing more shopping like through agents. So I think maybe we'll see it there um but even that I think will take a while. So yeah just in my opinion keep building for for humans in the best way possible. Any other Oh yeah. Oh, okay. No. Yes. >> Hi, Miriam. >> Hi. >> Um, how do you handle skills that build on other skills? So, like I have a general development standard skill and then a a specialized skill that should follow those standards. So, should one reference the other or should I let Claude discover and use both? >> That's a good question. Oh, by the way, if you go to my my GitHub, which is where my plugins sit, you'll see I have two skills there that I keep maintained. One is how to build a WordPress website skill and one is how to build a plug-in skill and then I keep them updated as I learn stuff uh along the way. Um from the in my experience it's good to have focused skills for each thing. Um and Claude figures out what to look at. So let's say I I have more let's say I'm I'm writing a blog post about plugins. I don't know this isn't a good example. Claude will know to look at my Miriam's voice skill and my plug-in skill theoretically and you know that kind of thing. So um the way I work is I keep them separate and I like try to use wording to and it's not that hard to indicate to Claude like the skills that it should look at. Maybe people are working in this cascading way. I Claude's not so great with keeping track of stuff as it you know uh builds on top of each other. So I I could see it becoming hard for it to remember all the rules in one go if they're all stacked on top of each other. But basically I just know how I work. I it could I don't know. Um sorry I'm not I can't be so helpful about that. >> I knew it was a tough question. I didn't know the answer myself. That's why I asked it. Thanks. >> Hey, >> thank you once again for the presentation. Um, and sorry if I missed this if you already addressed this, but I wanted to get your opinion on the value of the LLMS files. Um, I've seen, you know, different comments online. Even Google's made comments on the value of them. So, I wanted to see in your experience if you feel like they're really valuable or if they're just a nice to have. >> So, I I'll tell you why I think they're nice to have. So, I did a lot of testing on those as well, and I could not get Claude to see that I had the file. I put it in like four different places that you're supposed to put it in. And then I put it in the footer of my site because twice it was not pretty. And because Claude strips everything out when it's reading a page, it just didn't see it because it wasn't like important content and it didn't look at the four places that it was referenced. So it couldn't care less that I had it. It's like you might as well you can do build it easily with you know an LLM these days. So, okay, put it there and make sure I have a reference from the regular places. But I'm telling you, I I was like practically yelling at it, "Hint, hint, the site has LMS.ext file." And it still wouldn't find it. It was ridiculous. So, that's just from my testing what I saw, but maybe people saw something else. >> Appreciate it. Thank you. >> No problem. So for years, Google has said, you know, if you use Yoast, you can >> Okay, that's >> you can put in what you think the SER ought to be, right? >> You can put in what? Sorry. >> What the what the meta tech meta description? You can you can put whatever you want in the metad description. >> Yeah. >> And Google will take that under advisement. >> Exactly. >> And it's probably the same thing with uh things like the uh LLM file. Uh the other thing that uh I ran into years ago when we were supposed to be generating AMP files for Google, right? >> Yeah. Yeah. Yeah. >> Is that they don't always update as quick. And so I would imagine that Claude and other agents are probably going to use the HTML as the ground truth because that typically is is going to be up to date whereas other things like MMD files may not be in sync. So >> that's a very good point. if something goes wrong like uh with your automated process creating markdown files and you know it's not updated to the latest version of your post you made changes it's possible that the LMS will start to notice that and then just be like forget it I'm going to access HTML myself and you know generate my own markdown file that works for me so yeah it's very possible whenever things are like add like a level of complexity like the app file that you mentioned um that nobody can maintain and you know it's unclear how much value they add just it just doesn't become part of the the overall workflow So yeah, but we everything's in flux and who knows what will be. This is just what I know today, >> right? Basically, you know, it's an old website if it still has AMP. >> Yeah. Yeah. No, no, totally. Yeah. Okay. That's all the questions we can take. But thank you very much uh for being here and um I'm happy to answer any questions where wherever you find me. Thanks.