Welcome to Community and Code, a podcast about the human beings behind the code or the commits or the things that we interact with in our technical web-based lives. I'm here today with Zach Hendershot. I'm Chris Reynolds, your host of this podcast experiment. Zach has been around the WordPress space, around other spaces for a while, currently at CEO of Marooni, which is an AI powered WordPress for now plugin that does the handles like sort of edit requests and that sort of thing. I'll let Zach introduce yourself and what you do and what Marooni is. And yeah, welcome to the show. Awesome. Thanks for having me. I appreciate you chatting with me a little bit. Yeah, my background is pretty diverse. I've always been a technologist and hands in the code for twenty, twenty five years at this point, starting with very early web development, running BBSs from my house for many years before that and all the way up through all types of enterprise software. I've had my hands on both enterprise and startups. I've kind of gone back and forth between the startup space and sort of large Fortune-Fifty companies and get scared as hell and then run away from that and come back to startups and then kind of go back and forth a little bit. But the underpinning for me has always been, I love to just build cool technology, especially like cutting edge stuff, leveraging new tools that are entering our tool belts and figuring out what you can do with them, right? So that's really been my background. I've played a lot of roles. I've been deep in the code for many years. and a variety of different products, but I've also been, you know, leading product strategy and, you know, leading design teams and leading sort of, you know, multi multi-platform, uh, product strategy efforts of various sorts and kind of weaving back and forth to in places that allow me to just build cool stuff. Um, it's, it's kind of been the underpinning of my career. Um, and Marini is the most recent cool stuff, I guess, certainly from my perspective. So. Yeah, you got it right. The quick summary is that we're using AI to automate the requests that you get from your clients. If you're an agency or a freelancer building a WordPress site or ultimately any website, we provide a plugin for WordPress right now that allows you to put a widget on the website for your customer to just basically go to their website, point at the things that they want to change, whether it's as simple as a spelling mistake or broader you know style change to the website they just simply describe it in natural language and we can implement a large percentage of those requests automatically for the web admin to just review preview it and apply it the hope is you know we're saving anywhere from Fifteen minutes clicking nine to ten times in WordPress for a simple spelling mistake to hours of time that may be spent going and shifting a design language or changing colors or swapping buttons out or doing broader updates across the website and making those hopefully a few seconds at a time rather than a few hours at a time. So the big pitch for us is saving huge amounts of time, hopefully. Yeah, so I've seen stuff sort of like this in various sort of stages of where it sits kind of in the stack or in the process back from things like Figma where you've got layout design things and then it allows commenting and whatever somebody can go in and click on stuff or whatever and to like actual integration in like a browser tool or something and and that you know they can actually point and click on a thing and it actually relates to the website but this is actually not only that thing but if I'm understanding correctly it it is that thing but the ai part does the change for you so that you don't have to review the comment, find the thing on the actual page or in the code or whatever, and then make the change and commit it and whatever. This is like, no, the AI does that stuff for you and then gives you a diff or a pull request or something. Yeah, you got it exactly. Yeah, it's sort of like stripping away all those other sort of manual parts of the process. The layers of translation, I guess, is what we would call it, right? And just sort of stripping all those layers of transition away and just directly making the change as requested by the customer using AI and the fact that we're deeply integrated in the CMS system. So because we do, we're tightly connected to the data, the raw data, like the JSON files, like an Elementor, website has on WordPress used a huge JSON file to configure a lot of the widgets and other sort of visual parts of the site. Like we're connected directly to the database that allows us to go and make those updates directly to change wording or style or whatever it is. And so, yeah, like it's, it's really about just point and click and describe what you want and we can just go and implement it automatically without all the other layers of it. Now there there's still, we think there's, there's lots of great, you know, applications for those other tools, obviously in different parts of the process, we're focused on, on the interesting little stat in, in kind of managing all this. And anybody who's a freelancer or, you know, runs an agency will kind of intimately feel this, but often websites spend customers, right? Like website customers or owners of websites. The average lifespan of a website is about three years. Um, and, and then before they re platform or change technology or rebrand or whatever it is, and often they spend more money in those three years, updating and evolving and changing the website than they did originally building the website. And so it's on those three years, roughly, how do we, how do we make that really simple, really fast, really cheap to update and manage and evolve the website over time to meet the changing needs of the business? Not so much. building it for scratch, going from zero to one, we're more focused on one to a hundred right over the life cycle of the website. Yeah, that makes a lot of sense. And that's a, that's an interesting stat that sort of fits with like my anecdotal knowledge from, from agency life. But like, it's, it's nice to sort of have it have that kind of validated with actual raw numbers. Um, and I do see it like, I mean, AI is, is obviously a really, interesting component of website maintenance. One of the things that I've been looking at myself recently was with the new Drupal CMS distribution and the AI recipe and playing with that. I did some testing with that. I don't know Drupal super well, I've always been super intimidated by the menus and the admin experience. It's just a lot of buttons and menus and navigation to kind of go through as compared to what I'm used to, which is I've been in WordPress for twenty-ish years. But so using the AI recipe with Drupal allowed me to say like, hey, I want to build this type of site and I want this type of content, you know, and talk to the AI agent and the AI agent literally just says, okay, I'll build this content, like an event, right? Like, I want to put events on my website. Okay, we'll build an events content type. We'll build a view for it. It'll have, like, a date or whatever. And then it'll just do the thing for you, which is, you know, for somebody, like... And then I can probably go back and say, how did you do that? And I can learn in the process. But, like, that's a really interesting thing. And it feels like this is kind of a similar kind of take on how to do stuff or integrate AI into... the actual CMS site administrative experience. Yeah, exactly. My hot take around AI, I don't know how hot it is at the moment, but the hot take around AI is I think the chat agents and the zero to one tools are great. I think they solve lots of problems, but Using AI in deeply integrated workflows like CMS admin interfaces and streamline all the complexity that goes into sort of just updating and managing websites is where the sweet spot of AI, I think, will continue to shine. Not just in our tool, but lots of other tools that sort of automate workflows inside systems, I think is going to be a really powerful, evolving way that AI really amps up our ability to do more work with less effort, right? Because, and we measure this in the course of building Marooni, that even a simple update of a button, like changing the text or, you know, something simple about a button change on some random page. can take anywhere from nine to twelve clicks inside of WordPress, which is pretty streamlined, certainly compared to my experience using Drupal and, you know, older Joomla and other platforms that are just complicated. And they have to be complicated in a lot of ways because there's a lot of features and functionalities. But in WordPress, nine to ten to twelve clicks to just find a button and make an update, a simple update on a button, right? And so, like, All of that is relatively low complexity, low value add work that the web admin has to do each time they need to make an update. Like AI can solve that for you. Like that's a uniquely good, easy to solve problem. And it's not necessarily, you know, removing the human or replacing a human. It's just making the human faster, better and more efficient and more effective, right? And I think that is, I think, going to be the defining feature of AI is, The places where it makes the human in any capacity right now, it's a web admin, but the human just better, faster, more effective and more efficient. I think if it does that and it does that well with high accuracy, then it's going to be a win for you or whatever your workflow is. So that's what we're focused on. So let's rewind back a little bit and I want to hear about like what drew you to building a tool on AI? What interested you about AI besides, I mean, maybe it just, was it just, it's the next big thing or it's the next cool thing or everybody's talking about it or is there something, some potential or what things did you start to catch on to that said, I want to actually use this and see what we can actually build on top of it? Yeah, that's a great question. I've always been, I've always been interested and excited about the broader topic of sort of machine learning. I did some formal education in machine learning in my, in my college days, many years ago, I'm not going to disclose how many years that's information. But, you know, I did lots of learning and lots of exploration and, and, you know, I've always been intrigued by the potential and the possibility, especially in the technology space, right? Because I'm, I'm, you know, used to building the same algorithm over and over and over again over the last twenty, twenty five years. Right. Or the same set of code or implementing authentication roughly the same way, forty times in my career. And and I knew like I knew there was a better way or should be a better way. And, you know, over the last few years, as some of these tools and some of this, you know, theoretical knowledge has started to turn into practical tools. You know, I've been just very closely watching like what's possible and what's possible, what's about to be possible and what's possible today. And, you know, as we all know with ChatGPT and Claude and all these other ones, I think it's all started to click for me a little bit more like, oh, okay, this is far enough along that it can actually do real things in the real world for real people. And that's what got me most excited about, not just Marooni specifically and what we built there, but, but sort of, you know, exploring and playing with things like lovable and cursor and whatever else and start to like hit the boundaries of what's possible. So I've just been I've always been excited and interested in the possibility. I think I've just been now that it's been something that can actually be real things in the world, I've just started to get just more hands on and more excited and sort of more pushing boundaries to figure out what's possible and and certainly what's not possible. Right. Which is still a lot at the moment, I think. I am I'm trying to be as realistic as possible just because of my history and experience around this that like there's a lot of hype. I think we all know the hype is there. And yes, the gap of the hype and the reality is closing every day, but there's still a huge gap. I have a ten-year-old daughter, and I was teaching her how to use Loveable yesterday, actually. So this is a very recent memory. She wanted to build an app to match her. If you've watched the movie Clueless, and Cher has the outfit designer in her closet. And that's always been, as I've recently learned, a childhood dream of my daughter to have that thing. And so she, she tried to design it and it went really, really well for a long time, but then like, it wasn't uploading images appropriately and you know, all this other stuff. And it started to like really struggle with the, the nuance, right. And the details and the, and the long-term sort of context memory that it needs to build to do the things over a long period of time. right way now it's a huge accelerate right like the the fact that we got as far as we did in a relatively short period of time is huge but there are real limitations where real skill needs to come in and and go and build it and do it and that's kind of in a lot of ways feeding this the strategy of maroonia but but some of the other things you're trying to build is like how do you how do you find this middle ground how do you embrace the reality of what's possible and keep the human in the loop and make sure the human is a critical part of the process but but make the human better faster and stronger as I've said and and that's that's kind of in a lot of ways, guided our approach in terms of how we're deploying AI and how we're trying to use AI. But I think every day that goes by, that gap is going to get shorter and shorter and smaller and smaller. So it's going to be an interesting ride over the next couple of years, I would say. Yeah. I have an AI sort of test that I have run a couple times against different models since I first started using ChatGPT shortly after it was first launched, which also at the time happened to coincide with the first time I ever read anything by H.P. Lovecraft. So it's the Lovecraft test, I call it, which is write me a romance-themed story in the style of H.P. Lovecraft. And specifically, or write me a story in the style of H.P. Lovecraft with romance elements. And I specifically add the romance elements part because H.P. Lovecraft does not write romance. And it would be wild to consider him writing a romance novel. Right. But also H.P. Lovecraft in particular has a very particular writing style. There are particular elements that are very Lovecraftian. There's a whole ethos around Lovecraft's fiction. And just to get an idea of like what AI could come up with from sort of a creative, like not reproduced, like it can't because it doesn't exist, it can't just copy paste or like base something off of something real. It needs to sort of create something on its own. And I've run this a bunch of times and I use this as and because a lot of what I first started using AI for was for my own sort of creative pursuits, because like I run Dungeons and Dragons games. It's really good at like coming up with city names for fantasy or like NPC names or like, you know, you're in a village and there's give me twenty locations my players might run into. Right. It's good at lists of things with parameters. But I also like testing where does it start to fall down? What are the things where, okay, this doesn't really work or this doesn't sound good or whatever. The interesting thing that I found with the Lovecraft test is increasingly like I also run the test against like a scoring for like a grammar and structure and that sort of thing versus actual Lovecraft's writing and where he scores in terms of reading level and whatever. And increasingly, the reading level has gotten closer to the original, but the content and like style and the actual prose, it's okay, but it's not, it always lacks something, right? It's not a thing that I would go out and want to read. Where do you find or how do you sort of like keep a handle on like, where AI or large language models start to fall down? So this is a really hot topic for us right now. So I love that you brought this up. So testing and measuring the quality of our models that we're doing in Rooney, but broadly sort of the quality of models to meet a certain purpose, whether it's creating Lovecraftian, you know, pros or whether it's know writing code or whatever it is um it's something I deep dive it uh went pretty deep the last few weeks on um and we actually created we actually created an internal system to measure uh to run prompts and to measure the response we get and measure them both qualitatively and quantitatively and and by that I mean quantitatively In code, and even in your example, which I think is super interesting, you can measure in some quantifiable way, like whether it did write something that has key traits that define a Lovecraft novel or whatever. In code, you can also measure whether the code compiles or whether the code runs and different quantitative parameters of it. But I think where things get really interesting and I think there's going to be lots of improvement and advancement of the state of the art is around the qualitative, right? Like we're not talking, this is not, these are not deterministic systems, right? These are not like put in A, B and C and get D out. This is put A, B and C and put something that looks like D or something that looks like E out, but D and E are roughly the right response. They just are different approaches to the right response or, you know, the right, the right output that you're looking for. And so the person we've been taking from a qualitative perspective, is ultimately we use another LLM model to rate whether it meets the overall defined parameters that we wanted it to meet, which you could do a hundred different ways. You could have a hundred solutions to your solution, to your approach. You could have a hundred different novels that it wrote or short stories or whatever that meet your criteria. but do it in a variety of different ways. And I think measuring what is good enough and what doesn't cross the line in terms of good enough is, I think, something that is going to be a really challenging opportunity to go and solve for somebody in the future. You can measure the quantitative things, but the qualitative things are just harder to like, where do you draw the line? What's not good enough? Especially in narrative and words, right? But in code... There are a hundred different approaches. I was just reading a LinkedIn post the other day about, what was it? It was a big set of system prompts to feed into your LLM to write good Angular code. And it was a bunch of stuff about like, you prefer to use built-in components versus creating your own components, all these different sort of nudges in the right direction to write clean Angular code. Um, but like your definition, I don't know if you write Angular, but your definition of what good Angular code is different than mine. Right? So if the LLM spits out code that looks similar to what you would write versus what I would write, is that good enough or is that not good enough? Right? Like, I think those are the problems that we're struggling with just building an AI company or like, how do we measure what's good enough in terms of the solution to implement the edit request? versus it meets the technical requirements. It did it successfully, but did it do it the right way? And I think we're trying to scratch at how to put some guardrails, and we're using LLMs and other models to try and quantify the qualitative part of the response, but it's not perfect in any stretch of the imagination. I recently saw a post that was talking about a study that Apple did on artificial intelligence, specifically like models that exist out in the wild now and everybody is talking about and all the big AI companies are racing towards AGI, generalized intelligence, where the idea that the AI can actually sort of autonomously do things on its own. And Apple was like, yeah, in the test that we ran, the study that we ran, we're actually a lot farther from that than OpenAI and Anthropic actually want you to know, because really what we found is by and large the ai is just doing pattern recognition and when you ask it a complex enough question that goes outside of the scope of what it has a base of understanding of that already exists out in the wild that it performs universally worse at those tasks because it doesn't have that framework so it's not able to sort of create doesn't have a a creation thing which means it's even farther away from the idea of agi than you know sam altman would like us to believe uh but uh and and really what it's doing is just regurgitating stuff that it found on stack overflow or whatever and what if the things that it found in stack overflow is just not good you know what are their poor approaches or inefficient code or or have security vulnerabilities right like and we see that all the time in in the responses we get from the ll models that we use right like we still have to go and baby it in some ways. We have to correct it and coerce it and push it. And yeah, I think that's the big, I think Apple is saying this, the quiet part out loud, right? Like I, and it's the right thing. Like this is not, this is not some magical set of technologies that just magically mimic the way that the human brain works and leverages key facts and just turns them into original thoughts, right? Like, We may slowly get there, but like the sooner you embrace the fact that it's just rearranging and matching patterns to requests and doing it in novel ways. Sometimes I think you, you can leverage it better and more effectively. If you know that going in, you know what you're gonna get and your expectations are set appropriately and you're gonna get. the things that you want out of it more effectively than if you just hope for magic to come out. And, and, and it doesn't ultimately, you can convince yourself it's magic, but it's not magic, right? It's, it's a regurgitated set of algorithms that if you're asking for code or like, you know, previous words or organized in a certain way and direction that seem convincing, but it's not, it's, it's about setting your expectations and getting what you want out of it. But I think, I think there's, I don't want to use the word bubble, but I think there is at least some amount of a bubble coming, right? In the sense of like, we're starting to really understand what the real limitations are and what's really possible. And there's going to be a chasm that's going to form between what we think was possible a year ago and what actually is, you know, actually possible. And the tools are going to reset around it, you know, like builder AI, bankruptcy and stuff around like, No, you can't actually just magically get a fully production app. Like you had to fill in a lot of human gaps to make it seem that way. There's going to be a lot of those emerging. It's going to be an interesting time to watch as the reality and the expectations start to kind of, you know, hit their heads face on. A thing that that that comes up a lot when I'm thinking about sort of how I kind of exists and is used today and how and how companies are starting to maybe approach integrating it into their, you know, even employee base. A couple of different ideas pop up. I saw another thing recently that was talking about, you know, the idea that AI legitimately like makes you dumber because you stop Like your brain just stops firing on those cylinders, right? Like if you use AI for things to assist you in writing or even like myself in coding, then you stop practicing those things. And as you stop practicing those things, then you don't learn new things. And that alongside the idea of companies that are maybe not hiring as many junior engineers because AI can do a lot of that stuff. Like I see a danger of... the idea is the same sort of thing with like the Y two K and like COBOL developers, right? Like where the people with the experience have aged out of the system and we didn't hire enough junior developers to learn that system. So there's really nobody who knows how anything works anymore. And if like companies are choosing to use AI engineering instead of hiring junior engineers and letting their senior engineers review the code of the AI, that's fine for now. But at some point, those senior engineers are going to retire, right? And then we'll have this same sort of the situation where that we are in with COBOL developers. We have all this framework and infrastructure built on COBOL, but there aren't any COBOL developers anymore. I mean, it's a great time to become a COBOL developer. Yeah. And play that out a little further, right? Like, what if all the code that needs to train the LLMs on how to write code is all AI generated code? Like then at some point, like we enter this sort of narrow deep dive into a singularity of uselessness, right? Because it's just like training itself on its own, you know, regurgitated previous code. And so, like, who's innovating? Who's creating new algorithms and new ways of solving the authentication problem or doing it more, you know, faster and more efficiently? I think that's the real, I think you're hitting on the real risk in the next five years, maybe? I don't know. Where if nobody is firing with those cylinders, as you said, then who is coming up with unique approaches to solve problems with technology? if it's not the human. And if Apple is right, it's not really reasoning, right? It's just regurgitating. You combine those two facts together and like, What's the culmination? Not a great place to be, I would say, right? So I think how do we find the balance? How do we keep the human doing uniquely human things, which is the reasoning and the innovation and the thinking differently about an approach and leveraging AI to fill in all the boring, low complexity, low value add gaps, right? A bunch of the boilerplate stuff that doesn't need to be reinvented and marry the two together in a way that continues to advance the state of the art. I think that's the problem to go and solve. There's probably some money in it too. If anybody's out there listening, what's the problem? Does that worry you at all? Or what do you think the future of our industry of web development, but broader just technology in general, how do you see... integrating with AI. But I guess I guess I would like to hear your take on both an optimistic view of like, let doing it the right way. And perhaps like the doom alternative, like where, you know, I think we've already sort of talked about the doom is the doom is like, like, nobody knows anything, and nothing is built, nothing is creative. So what's the what's the alternative? What's the better way of using these things? And what do you see as like the path forward? You know, I am an optimist, by the way. I think we've talked a lot about some of the doom and gloom and stuff, but I'm an optimist in terms of what this technology can do for us. And I think the way that I see it from an optimistic lens is that, you know, the people who are going to leverage it and are leveraging it most effectively today are using it as an enhancer, not a replacement of the thinking. Right. And so I think. the end state that I get most excited about and that I'm chasing in my own use of AI and hopefully within Marooni as well is that if I can free up the human to focus on things that are uniquely human, right? The things that allow them to think and innovate and come up with new solutions to problems and get out of the way of the hours and hours and hours that they spend doing the stuff that doesn't contribute to that, doesn't add value. I think AI is an incredible enabler to get the human to solve harder, bigger, more complex problems than they could three years ago. And so I think that's, That's the optimistic point of view in my view is that this technology can then free up a ton of human hours globally, billions of hours that is today spent doing stupid things, frankly, and go and do them spending, hopefully, value added, human advancing, race enhancing thinking and thoughts and whatever. I think the thing we have to fight against is what we just talked about, which is making people lazy and replacing that thought and not using that thought productively, just using it to watch TV or whatever it is. But not that that's not valuable in its own ways and different ways, but I think that's the big risk of it. But I think the net gain of our ability to, as a race, be more focused on uniquely valuable things is going to pay off dividends is my optimistic view over time as tools continue to fill in gaps of the stuff that we don't want to do and shouldn't be doing. You mentioned your daughter, and it makes me think about how AI is or is not used in sort of an educational context. One of the things that I thought very early on in my personal use of AI was that it felt like a... uh it felt like I could use it in a way that would actually teach me how to do how to write in language languages I didn't necessarily know as well right like I can say like I know like I write javascript I know javascript I might not know all the things I can ask it about a javascript related problem and they can give me an answer now that's that means that I see how it works I can implement it and I can tweak it later or whatever um I could do that with python I could do that with a number of different languages that like I do or do not know right So I see that as as an opportunity, as as a thing like, you know, before before COVID, we had the Khan Academy and like you could go in and do your own sort of individualized learning. But now I think actually even Khan Academy is integrating the stuff. But now, like I could ask chat GPT about this thing. The flip side is It's, you know, a lot of teachers and a lot of classrooms and a lot of schools see risk in integrating ChatGPT and AI in education because it's a crutch and like learning to balance that. How do you see like how, you know, using AI to build this tool for your daughter? How do you see the role of educators, the role of parents in teaching good AI habits or ethics in the same way that we need computer literacy? What is the layer of AI literacy that you feel like we should be communicating to kids? yeah I think a lot about that because of you know my daughter's now living in the age of ai for lack of a better term and and it's a tough it's a tough problem I I think that the way that I think about it is pandora is out of the box right like we're not gonna we're not gonna put these tools back away I think there's an interesting number of of people out there who are like we should heavily regulate ai we should you know roll it back we should slow down like historically speaking, good luck. I think that's a completely unrealistic point of view. And so my point of view is embrace it, leverage it as much as you can, deeply integrate it, get people to understand how to use it effectively. I think by extension, it's going to become a crutch. And I don't know if we should, frankly, I think we should embrace that in some ways. I think We have embraced, and I hate to go back to the industrial revolution and sort of draw analogies, because there are some analogies, but I think it's different in a lot of ways. But like the industrial revolution, we lost huge amounts of knowledge in the world around how to do many of those manual efforts that many of the machines automated and replaced. Do we care today that nobody knows how to do those things? You know, some people probably do, but I don't. Frankly, the fact that we have the machines and we invented the technology means that we don't have to think about that thing anymore. That that used to be a critical skill that we that we needed to have two hundred years ago. Right. I think we should look at it roughly the same way. And that might be painful for a lot of folks who spent their entire career trying to teach a specific skill to kids that kids frankly don't ever need to know anymore moving forward. And I know that's painful for a lot of people who want to appreciate that skill or teach it or spent their life learning it and cultivating the knowledge around that skill. But I think we should embrace it. We should lean into it. Those do become redundant. Those are no longer valuable. Teach kids to use ChatGPT to answer it in three and a half seconds and move on to the next set of things we should be learning that AI allows us to do more effectively or whatever it is. It could be a defeatist in the sense of just lean into it and own it. But I think in a lot of ways, that's probably the right thing to do because It doesn't make a lot of sense to spend huge amounts of time to teach kids things that are no longer something that they have to uniquely create on themselves or uniquely know themselves and can lean on ChatGPT or any number of AI tools to just do it for them. And I think it hearkens back a little bit to the calculator argument, right? You know, but I think we're seeing calculators become a critical part of learning in a way that they weren't twenty years ago. Certainly when I was when I was learning a lot of more advanced math, like calculators, even at that time were very frowned upon. But I know these days, like they're a critical part of your tool set. Right. If you if you're not using a calculator for some of these more advanced calculations, you're spending five times longer than you need to to get a quick answer and move on to the next part of the solving a problem in the world. And I think it's going to be another version of that, probably hyper scaled version of that in some ways. But that's my point of view. Lean into it, own it, embrace the things that is making redundant and use it to do things better, faster and stronger and, you know, play a little bit of a wait and see in terms of what that opens up. Yeah, I think that I think that I mean, first of all, I think that teachers need to be paid a lot more. But but I also like part of that is, I think what I I believe that. teachers should, and people broadly, really, should be educated on what the current state of AI, and maybe it doesn't change, maybe we never get to AGI, right? But at least what the expectations are for what AI can do versus what it cannot do, so that they can more critically, you know, like, yeah, lean into it. I think that it should be something that is part of schools or learning but but I also think that it should be done in such a way that is understanding of the limitations like I think that the people that are teaching these things need to understand that like it is this thing and it's not this thing and like here are the boundaries. And when you know the boundaries, then you can better communicate that to your students or the kids or whatever. Yeah. And I'm still an advocate also. I completely agree with that. I'm also an advocate of you understanding how some of the pieces and parts, like how the sausage is made in some ways. I understand how it works and why it works and what it's trained on and all those things, because that gives you important context in terms of what you get out of these tools. Right. So I'm still an advocate that you understand like why and how, but not necessarily all the details of how you go from A to B that an LLM can just automate for you immediately, right? Like those are, I think that's some of the nuance that I think teachers are gonna have to figure out over time, all of us to figure out over time. Right, because what I've seen in a couple articles recently has been people who maybe don't necessarily have that context go into using ChatGPT or whatever, choose your LLM of choice, with the idea that this is really just a super-powered Google search. And while it can sometimes be that, It is not always that. So, like, there are, like, legal cases where, like, you know, the junior lawyers are filing, you know, doing filings, whatever, that they used AI to help them with. And it's citing cases that didn't exist and stuff like that because it's just generating stuff. And it sounds reasonable, but the judge looks at it and does whatever. Like, let me find more information on this case that you referenced. Oh, that case didn't exist. Oh, it's because you used it. And it's like, you know, the junior lawyer, like, and like this article I was reading is specifically about like how AI is used in like law generally and saying like this, the problem of like like being rushed for time and not having good like data for for some of these like depositions and whatever and filings like it's not new and it's it's not like, you know, we didn't have bad stuff before. It's just we have bad stuff now in a different way. And that way is because we're using this tool and and there isn't enough education that, you know, it's not just a really good search engine. It can be a really good search engine and certainly can be integrated into search engines to make those more robust. But also, it's essentially like playing, it's predictive text. It's the autocorrect on your phone just extended to, like, now I can write paragraphs instead of just, you know, the next word that I think that you're gonna type. Totally. I think it's both ands, right? I think it's one, you get more educated in terms of how to use it effectively and how to vet it and have the education to make sure that, you know, what it's fitting out is reasonable and makes sense and is whatever. But I think it's also these tools are going to be, I think you're going to see a huge investment over the next couple of years in LLMT. tools, significantly reducing sort of the hallucination as you sort of laid out, right? Like having layers of validation and trust built into the system, because I think that's the biggest challenge I think a lot of people have with using these tools right now is what can I believe and what do I need to ultimately just do my own research on each time, which doesn't save me a ton of time at the end of the day. Like lawyers specifically is the flagship example, but there's lots of others. I think as those tools get better and reduce hallucinations and do have some layer of verification or trust or whatever that are layered into the responses, I think we'll find a middle ground at some point. But you're right, right now it's a landmine in a lot of ways because you don't know what you're going to get and you're putting in a huge amount of work to make sure that what you are getting is valuable. I like the approach of, I don't know if you've used the deep, deep research tool, which is GPT or whatever, like, but it's heavily cited is the bottom line. And I've, I've, I've used it for lots of different sort of like market research and whatever else, but it's heavily cited. And that heavily cited content is adds a layer of trust to, to the process. You can, you can go and read the article and, and, and validate that the article that the fact that pulled out of the article is accurate or whatever, it requires huge amounts of, uh you know relatively speaking huge amounts of human effort to go and do that but I think it's a good balance you can go and check what it's saying um you don't have to and you can just copy and paste it and use it in your legal brief or whatever but at least it gives you the set of tools to build some trust to make sure that you know you're using this content responsibly and you don't have to do it but it gives you the tools to make it easy to do it you know because I think more and more of that as we go Yeah, for sure. And I definitely, in my own sort of prompting, and even in custom prompts, I've kind of gone towards just as an additional note, like, please cite sources whenever I'm asking it questions about things that have an answer. Like, please just tell me where you got this information because I don't believe you by default, basically. Yeah. Switching gears a little bit, going back sort of more into the WordPress space where your plugin currently lives, what do you see? WordPress is kind of in a bit of a, I would almost articulate it as like a transitional phase. There's a lot of like change happening around both the community, maybe the core software itself, governance, that sort of thing. What is your take about the state of the project as broadly, and where do you see WordPress, which currently powers forty-three percent of the internet, where do you see WordPress going in the next five or so years? Yeah, it's an interesting time, as we all know, following along in the WordPress space around this fight between centralization and decentralization and sort of the, you know, the core ethos of open source. And what does that mean? And, you know, Matt's perspective is different from other people's perspectives on this. And it's a swirling world right now. I think this push towards decentralization with FAIR and everything else is is healthy is, is the bottom line for me. I think it's a healthy thing to have options. I think it's a healthy thing to create new paths to, um, you know, for, for plugins, to, to, uh, to install different like plugin repositories and different ways of sort of like, you know, creating, creating the right environment for the right sets of tools and creating more of a, you know, distributed community in a lot of ways. I think that's all healthy stuff, but I think the challenge. that has to be considered. And I think, you know, on the path towards answering where WordPress is in five years, I think is how do you leverage the decentralization to, you know, not create sort of vendor lock-in, but create the right amount of flexibility for, you know, people to do what they want on the platform, have some, you know, leverage the things that are great about open source, but drive adoption, drive confidence in large corporations and other folks who need to use a platform, have dependability in terms of like, you know, security and other things like that. I think that's the biggest risk with things like FAIR is how do you how do you create consistency? How do you create safety and security by being deeply decentralized? I've seen lots of projects in my twenty, twenty five years building software, especially in the open source space that sort of start to diminish in terms of use because there's too much variability, too many plugin repositories and too many stores. And Linux distributions are a good example of that for people who have used Linux over many years. There's been some centralization over time, but there are dozens of different Linux distributions that have different pros and cons. And that's good for the niche, for the people who need a specific type of server running on a specific type of CPU architecture, who have very unique needs of high IO or whatever it is. But in terms of a one size fits all, That's what large organizations who are going to drive real adoption want to need. They don't want to pick out of one of seven hundred different distributions. They want to know how to use Linux and how to deploy it. There's a reason why Linux isn't the de facto OS that everybody uses. There's a reason why Mac OS became sort of the thing that people used even over Windows is because it's easy and it just works. And it just works. And I think that you summarize, I think the biggest concern I have with you know, fair and sort of this decentralization push is like, then we're dangerously at a cusp of it just not working because this, you know, fork of WordPress or this fork of, you know, the plugin and this repository over here is broken here, but not over here. And like, there's no like unified path for how you just get a website built and deployed for a customer. Now, you may build your own way in which to do that, but that may not work for the next person who comes in and takes it over or whatever it is. And so I think that's the biggest risk that I fear in that five year timeline. But if it's well organized and the Linux Foundation being a part of that is, I think, good in some ways because it provides some amount of oversight that allows allows the minimization of some of those risks. But I think that's the biggest thing that everybody should be worried about a little bit is fragmentation in the WordPress space is now upon us. And there are pros to that. But I think there are some real risks to adoption and growth of the WordPress ecosystem that we all should be somewhat concerned about. Thanks. Thanks, Zach, for coming on. Before we go, do you have links or places where people can find you online? Things to share for folks that are looking for you? Yeah, it's quite easy. You can find me at maruni.io, M-I-R-U-N-I.io. You can find me on LinkedIn as well. I talk a lot about AI and agencies and how to leverage AI and some of the new ways that AI is changing and evolving and simplifying how you run your business, both as an agency and freelancer. So always happy to engage and talk about AI with anybody who wants to reach out. So you can find me on LinkedIn as well. Awesome. Well, thanks again for coming on. And thanks, everyone, for listening to the show. And we'll see you next time. Awesome. Thank you.