Latest / The Laravel Podcast / Laravel & AI: Inside Boost with Ashley Hindle

Transcript

Laravel & AI: Inside Boost with Ashley Hindle · The Laravel Podcast

• Matt Stauffer: Hey everybody, welcome back to Laravel Podcast Season 7. I'm your host, Matt Stauffer, and I am the CEO of Tighten, which is hard to say apparently, and in this season I'm gonna be joined every episode by a member of the Laravel team. Today I'm talking to Ashley Hindle, my friend, senior software engineer at Laravel. Ashley, can you say hi, share a little bit about who you are and what you do at Laravel? Ashley Hindle: Absolutely. Howdy. Excited to be here. I am doing all things AI at Laravel. I'm pretty new. Been here about a month and just very embedded in all things AI. Yeah. Matt Stauffer: Yep. So one of the things you know I asked at the beginning of every podcast is what was your journey to Laravel? And I'm especially interested in, I mean, I want to hear everybody's stories, but you definitely, I think you and I met for the first time and I was like, hey, you kind of have come out of the woodwork, you know, and you're like, yeah. And then I felt like a week later you're working at Laravel. So it's just very fun to kind of see, you know, kind of where you're coming from. So could you tell us a little bit about your story? Ashley Hindle: Yeah, that's a weird one. So I used to work at remote.com. I worked there for about 18 months, leading an engineering team that built a performance management system for their HR platform. And that was all Elixir and React. And I love Elixir. That was a lot of fun. And I left there in January and was trying to figure out what do I want to do with my life? Like, do I want to... Matt Stauffer: Okay. Ashley Hindle: Stick with Elixir. Do I like big companies like Remote? Do want to go back to a small startup? Because before Remote, I was a CTO at a 100 person startup. And so I like that vibe. Do I want to do sales engineering? I considered that for a while. Do I want to start a startup? I know a couple of angel investors so could get some funding, go the angel investment VC route. And I do want to be CEO of something someday. So I just took a break and just did lots of things. Lots of talking to people, lots of experiments, just what do I want? I'm an old, old man. I need to figure out how I want to spare that time, right? And so I was doing all of this, but I started posting on Twitter lots of things. So I was posting in a couple of months period just weird CLI experiments that I was doing and just weird experiments in general, just stuff... Matt Stauffer: Yeah, that was fun. Ashley Hindle: people shouldn't do really, like, you know? And then everybody, not everybody, many people are messaging me saying like, wherever you come from, you're on my feed constantly. I can't get rid of you. What's going on? And because, just because I had all of this free time, right? Like I could freely experiment and do anything I liked. So I was doing CLI experiments, lots of AI experiments and lots of MCP experiments. Matt Stauffer: Mm-hmm. Ashley Hindle: And so I submitted a talk to Laravel Live UK about MCP to teach people what it is, what's the hype around it, should you care or can you ignore it? And I was accepted. And that was super exciting. My first conference talk on stage this June. Just from doing a lot of weird stuff and sharing it. It's like you can just do things. It's like you can just do stuff and share it and stuff happens and increase your look surface. It's who knew, right? So yeah, I do the talk at Laravel Live UK. It goes pretty well. It is live coding with cursor and AI and MCP. So it didn't go 100 % smoothly. There were definitely... Matt Stauffer: Yup. Ashley Hindle: There's some issues there with AI just deciding not to play ball, but hopefully I, you know, I've vant that enough that it was okay. I got lots of great feedback. And then Peter Suhm was doing MCP experiments at Laravel at the exact same time. And he was speaking at Laravel Live UK too about how to use Cursor super effectively. And his talk was awesome. And so I was obsessed with him. I was like, hey, you're at Laravel. That's where I want to be. You doing MCP stuff? That's what I'm doing. We should talk, right? And so we chatted a lot at Livewire, at Livewire? Laravel Live UK. Yeah, and so I learned more about what he was doing. He learned about what I was building at the time called Croft. And we were hanging around after Laravel Live UK at this pub. And then the president of Laravel, Tom Crary, turns up and he's like, hey, we're going to go to this like enterprise dinner. One of you can come. Well, Peter works at Laravel, he should come. And then he was like, no, I'm joking. You can both come. And so all of a sudden I'm like accidentally at this enterprise dinner with Laravel customers, Laravel staff. Yeah. So like very fortuitous, weird turn of events. And then I actually spent most of the evening just bugging James Brooks because we are so similar and he's so lovely and he's such a lovely human and it's ridiculous, right? Matt Stauffer: He's such a lovely human being. My goodness. Yeah. Ashley Hindle: And he has, yeah, I really like his thought processes and his values. yeah, I got on a lot. So I just bugged him most of the night. And then Tom DM'd me like a bit later, like, We're doing AI stuff. You're doing AI stuff. You should speak to Taylor and see, like, can we merge these? Like, could you help? I spoke to Taylor. I could help. And he was like, you'll have to speak to Joe T. You know, he knows what he's doing. He does. So I spoke to Joe T. And he was like, you're all right. You should speak to the director of engineering, Andre, who was on this podcast previously. And I spoke to him. And he was like, you seem all right. So let's talk about an offer. And so, and that happened, you know, really, really quickly. And I say, fortuitously, because when I say I'm lucky, my wife gets upset with me because she says, no, you just put in a lot of work and a lot of effort, like, you know, give yourself some credit. But fortuitously feels like that's a bit of luck. Matt Stauffer: Yeah. Yeah, I love that. Yeah. Ashley Hindle: and hard work like combined. I feel like that's, yeah. Matt Stauffer: Well, that's very cool. So one of the things I wanted to make sure we got a chance to talk about before we talk about any of the work that you're actually doing at Laravel, I wanted to talk about your thoughts about AI overall. And AI is this big question. Every, I talked to my mom yesterday and she's like, is AI destroying your industry? What's going on? Everybody wants to know. And so, you know, I want to hear, you know, when folks are playing around with the latest and the greatest. Ashley Hindle: Yeah. Matt Stauffer: Sometimes those folks are maximalists, right? They're playing around with it they think AI is going to change the whole world, everything's going to be AI, and sometimes the people actually using it are the most pragmatic because they're like, I know what it can and can't do. Ashley Hindle: Yeah. Matt Stauffer: So I'm curious if you were to just kind of like say, like, what do you think the future of the programming industry is given AI's current and future state? Do you know? Do you have a sense? Or you're just sort of like, hey, I'm along for the ride. Ashley Hindle: I would say I'm closer to along for the ride. I would like to have gone off to a cabin and thought about this properly so I had a useful answer. Matt Stauffer: Yeah. Ashley Hindle: And I've heard lots of other people's answers, but I don't know how well thought through they are. But I think, what do I think? Let me figure it out right now. I think if you're not using AI, you're falling behind. I think it's a necessity in our roles moving forward. I don't think it's taking our jobs anytime soon. It is an enabler. It's an assistant. It's a pair programmer. It's a colleague. It's not a replacement. Maybe it will be in 15, 20 years. I can't see that far forward. But right now, and for the foreseeable, we should use it as an AI pair programmer. That's how I talk about it. That's how I consider it. And GPT-5 was released yesterday and I watched a live stream and I've played with it last night and this morning to see how scared should I be, how good is it, you know. And it still kind of sucks in many places, right? Matt Stauffer: Yeah, it's still a tool. Ashley Hindle: Like, this is the flagship model, spent billions of dollars on, and it's very, very good but it's not as good as us. It could work for longer hours than us, but it's not as good as us. So I think everybody should be using AI, but we don't need to be worried. Let's just take advantage of it and do cooler things, do more things because of it. Matt Stauffer: Okay, I love that. And one of my next questions I'm gonna actually delay, but one of my next questions is gonna be if somebody hears that and says, I'm not using AI right now, how do I get started? I want you to be able to introduce boost first because I want that to be allowed to be part of the answer. But I you know that one's coming because I think to me the natural question is, you know, like, okay, if everybody should be using it, how? But what I do wanna ask you right now is... Ashley Hindle: Okay. Matt Stauffer: I know that there's a lot of different contexts that people use to use AI assistants in their coding. And it's not like you're going to say what's the right one to use, but I know that there's the Claude Devotees and the Junie and PHP Storm. what is your, when you're writing actual web page code, what is your actual workflow right now? Ashley Hindle: This is a really bad time to ask because... Matt Stauffer: What was it a week ago? Ashley Hindle: That's a bad time as well actually. For the last two weeks, three weeks maybe, I've been going back and forth between Cursor and PHPStorm. Always using ClaudeSonnet4. It's the best. But often in PHPStorm, using Claude Code, not Juni. Sometimes I'll use Juni, but often I'll use Claude Code extension within. Matt Stauffer: Okay. Ashley Hindle: PHP storm. So technically using Claude code, but with all the refactoring and PHP specific knowledge that PHP storm is amazing at, right? And that's why I went from cursor to PHP storm. Because cursor is the best idea for AI tab completion and chat, I think. But because it's so general purpose, it doesn't have all of the helpful tools that PHP Storm has, like simplifying fully qualified domain names or extracting methods and properties. Like with just a couple of keys, you can do so much in PHP Storm that you can't in Cursor. And I was missing that. So I switched to PHP Storm and Claude, but then I was missing the amazing type completion of Cursor. So I went back and I'm still not happy. I need them to combine and that'll be where I end up. But yeah, right now, that's what I'm about. I go between two and a half tools. Matt Stauffer: Yeah, so Claude Code in PHP Storm, because Claude Code is a CLI tool, so what does it look like to use Claude Code in, is it anything other than just like a little window that pops I'm completely unfamiliar with this. It's a little window that pops up that gives you just its own dedicated terminal window, or what else is it adding? Okay. Ashley Hindle: Yeah, basically, it basically just runs Claude the CLI within the PHP storm terminal. And you can also have it in Cursor and other IDEs. But the useful part is that Claude supports connecting to your IDE. So the Claude and the IDE can exchange messages. And so the IDE can say, the user is in this file right now, or the user has these tabs open right now, or the user is highlighting this code right now. And so Claude code can perform better. Matt Stauffer: Okay. Ashley Hindle: Because it has some more information from the IDE. So, yeah, super, super interesting. I don't know how they do that, but it's very cool. Matt Stauffer: Okay, I had no idea and that's always learning new things. Ashley Hindle: You gotta give it a try. It's surprisingly good. Matt Stauffer: I love that. Okay, so I think it is time for us to talk about Boost. So normally I ask everybody to assume that everyone has seen the latest, you know, Laravel announcements, whatever, but actually I think I want to hear it in your quick intro on a podcast version because I think that's gonna be a unique, even for people who've kind of seen you announce it at Laracon. Ashley Hindle: Yeah. Matt Stauffer: I think hearing you give the quick intro to podcast is going to be helpful for us to continue understanding what is Boost and what isn't Boost. So let's assume that we may or may not have watched the thing. So you got to cover everything, but just give me the intro level. What is Laravel Boost? Ashley Hindle: Laravel Boost is your AI coding starter kit. That's its tagline. It makes it really easy to get started with AI coding in Laravel projects. So you don't have to go and find random MCP servers and random cursor rules and cobble together this whole thing. I've poured my blood, sweat, and tears into it for a long, time to do all of that for you and combine everything together so that when you're using Boost, when you're AI coding, you get higher quality output, basically. That's the goal. AI works better for you. That's the aim. Matt Stauffer: That's a great elevator pitch, which gives me the opportunity to dive into a couple subjects because you would be surprised. I would say more than 50 % of listeners of this podcast have never coded with AI in Laravel before. It's a much bigger... Everybody on the internet does it, but it's a large portion of people who are not constantly on Twitter, constantly talking about those things that are not using it, instead of day to day. Ashley Hindle: Interesting. Yeah. Matt Stauffer: And a large portion of those people want to, but don't know how to get started. So one of the things I love about Boost is that it's the here's how to get started. And there's two things that you mentioned. I know there's more than just those, but those are kind of the two majority of things. And you said rules and MCP. So let's start with a simpler one. If I have never worked with AI coding assistants before, and you just told me Boost is gonna give you rules, what does that mean? Ashley Hindle: Sure. So all AI agents need steering. They need help on how you would like them to behave. And they like to be steered. They want those guidelines and rules to do better to help you. So these guidelines are things like write PEST tests, not PHP unit tests. Or if you're going to run the tests, run a single test, not the entire test suite. You've only changed one test, so just run that test. Or you can say, I don't like to declare strict types at the top of files, so never include that in new files, things like that. And so they're just advice to the AI, hey, we're pairing together. This is how I work, and this is how I'd like you to work so that our code looks the same and blends seamlessly so I can easily read and understand your code. So you can easily read and understand my code, right? So yeah. That's how I like to think about it. Matt Stauffer: I love it. Yeah, we've been working on a consistent set of rules files for all our various AI tools at Tighten. And so I'm super excited to say, now we don't have to start from scratch. Let's just take the boost rules and then see if we have any modifications we're gonna make. So I'm very excited about that. Ashley Hindle: Awesome. Ashley Hindle: Yeah, and the Boost rules as well are, I call them composable guidelines. So Boost will detect the packages you have installed and only include the guidelines that your project actually requires. So if you don't use Pest, you don't get the Pest guidelines, right? If you don't use Inertia V3, you don't get the Inertia V3, V2 guidelines. So that helps. And also we automatically import any project specific guidelines you have. So if you're working on a weird esoteric project, you can add guidelines. And when you run boost install, we'll add those to the guidelines, too. So we're trying to bring everybody into the guideline space. Matt Stauffer: Love it. OK, so that was guidelines. Those rules. MCP servers is a harder elevator pitch. Because the MCP servers as a concept is relatively hard to introduce, MCP servers as it is beneficial to Boost, I think, is an easier concept to introduce. So let's operate in the assumption that you all believe that an MCP server is basically an API for AIs to call to. Ashley Hindle: Yeah. Matt Stauffer: The APIs can get something from or do something to some system using an MCP server. That's the super dumbed down version. If you're not familiar, I just gave you the really, really, really, really, really dumbed down basic understanding. So that when Ashley's talking about what they do, you can at least say, I imagine it's basically some tool, some server-based tool that basically that my AI agents can call to for doing things or asking things. Okay, so dumbed down version given what are these MCP servers in Boost going to do for us as level programmers? Ashley Hindle: Yeah. So Boost provides an MCP server so your AI agent can speak to Boost, essentially. And MCP offers lots of functionality, but the main important functionality is that the MCP server, Boost in this case, can provide tools for the AI agent to call. So the AI agent can list the tools from the server and say, what will you help me do? And then it can say, hey, okay, I want to call that specific tool. So... Boost, when you add it as an MCP server into your AI agent, will offer you 16 tools out of the box. And it allows extensions, so you can add your own tools as well if your esoteric project would benefit from a tool too. The most important or cool ones, I guess, are Tinker to allow the AI agent to execute PHP code within the Laravel application context. So it can use your eloquent models, and it can find the latest one, and it can sum and aggregate things for you and return information. That's super helpful for debugging issues or testing code snippets. Often, before I used Boost, AI agents love to create random PHP files, like test hyphen this thing hyphen dot PHP. And then it would put some rubbish code in there and then it would fail and then it would try and run them and then it wouldn't clean them up. And so Tinker just cleans that up, which is super, super helpful. My second favorite, or maybe my favorite, is a browser logs tool. So when you install Boost, we'll inject some JavaScript in, via middleware, so that your browser logs and errors and exceptions get streamed back to Laravel into your storage logs directory, into browser.log. So you get access to them and the AI agent through the browser logs tool also gets access to them. So when you say, it's broken in the browser, you don't have to go to dev tools and copy paste the right thing. It will just know to call the browser logs tool and get the most recent browser logs. And I really, I really love that because I do that a lot. I break the front end a lot and then ask AI to fix it. And then probably the best tool... Matt Stauffer: Yeah. Ashley Hindle: that's hiding a lot behind it is the search documentation tool. So this is the most powerful tool. In the backend, we have an API, the Boost API, and we've ingested and chunked and vectorized all of the documentation for the Laravel framework versions, major versions, for Inertia, Laravel Inertia Vue, Inertia React, PEST 3, PEST 4, PHP unit, others that I'm Livewire, we just added Livewire, Flux UI, all of these things, we've got the documentation for them when we put them into an API that this tool, this MCP tool search documentation can use. And when your AI agent calls it, it passes through the packages that you have installed, it will share with the API. So the API will only give you documentation that's relevant to your specific packages and the specific versions. So if you're using PEST 2, you'll get PEST 2 documentation. If you're using PEST 4, you will get PES 4 documentation. And that is the coolest thing by far. And it's like, it just looks like it takes up the same amount of space as any of the other tools, but it's hiding so much complexity and other work behind it. So I feel like that deserves like a special shout out because it's doing a lot, like we've done a lot to enable that, right? So yeah, that's the coolest one, I think. Matt Stauffer: Yeah, that's very cool. So before I dive into the questions I have, and I actually got some questions from the Tighten team members as well from this, I want to ask, is there anything that's new about Boost since Laracon that you're excited to share? Ashley Hindle: Yes, but I'm midway through it. So I'll share it, but, and I will complete it, but it might not complete for when we launch. So one thing that I've been working on that I'm very passionate about is Boost benchmarks. So I'm building an entire benchmark and evaluation system to show where Claude code or open code works better with Boost than without boost. And so I have a bunch of projects of different styles. Some are Laravel 10, some are Laravel 12. Some use Pest, some use PHP in it, some use Vue, some use React, some use Inertia, some use LiveWire, some use Flux, know, Tailwind 3, Tailwind 4. Just a whole mess of different projects that I hope people have in real life. And then I have a bunch of evaluations and benchmarks to ask the AI to perform tasks on these projects. And then I score the result of them. Did it do what it needed to do? Are we happy with the code? Things like that. With and without Boost and run it multiple times so there's no just random one-offs. And then at the end, we'll have a score and we can say if you use Boost, it is X percent better than if you don't use Boost. But also, excitingly for me, I can see where maybe it's only 1 % better in this scenario. I'm like, okay, well... Matt Stauffer: Sure. Nice. Ashley Hindle: Why? Like how can I, yeah, how can I improve that, right? And also it lets you test it across different models. So GPT-5 came out. So I want to be able to say if you use GPT-5 with Boost versus Claude Sonnet 4 with Boost, this one works better. This percent or something like that. So it's very difficult though. I started on this thinking, yeah, I know the rough pieces. I know how this will hang together. And then the more I get into it, the more... Matt Stauffer: Let's work on that. Yeah. That's cool. Ashley Hindle: I think I should have left this for somebody else. Matt Stauffer: I'm just, yeah, mean, even just trying to imagine what building a testing harness and framework for that would look like. And I'm like, yeah, that sounds like a lot of work, especially in like a week. So. Ashley Hindle: Yeah, I'm way too optimistic, but I really, I don't think it's enough to say Boost is awesome, look at it, call tools. I really, I want to know and show it is beneficial in these scenarios and where it isn't, great. Now I know where to focus my attention. And you know, all of this is going to be open source, right? So if other people see things that we don't see, I need that feedback, need that advice and those pull requests. So yeah, really looking forward to it. Matt Stauffer: All right, so I'm going to move now on to the actual questions that I had pulled up from my team. So let's go back to that question I told you was coming, which is somebody has known this whole time that they are supposed to be using AI and they just don't yet. Does Boost make that easier for them? And if somebody is trying to do that and let's say Boost has launched, which it will have by the time this releases, what are their next steps for trying AI for the first time? Ashley Hindle: Yeah. Matt Stauffer: So like literally, I'm a Laravel programmer, I've been doing it for a decade, I've never tried AI, Boost is out, what are my next steps? Ashley Hindle: Install Boost into a project and probably use Cursor in that project. That would be my advice. That's the quickest, easiest way to do it. You can do it without installing Boost, but people will give up when the AI agent's quality isn't up to the standard, right? So I think if you install Boost, you will automatically get higher quality output. So I would recommend... Just open, even if you use PHP Storm, you use VS Code, you use Neo Vim, I don't know what people are using. Just open it in Cursor and ask it, even just start with asking it to explain something. Like what's going on in this project? What's the path from this form to the database? And just let it show you that it understands the code base and it can do things. And then start super small. Ask it to write you a test. Just say, you know, highlight the code and say, write a unit test for this, write a feature test for this. Even if you could do it in the same amount of time the AI takes in this small test, this small example, it's very valuable to see that it is capable, right? I think lots of people don't get over the, is it capable barrier, I guess, and give up because, I could write that in a minute too, you know, that's not beneficial. But once you get past those small tests, you can start believing in it and giving it harder things. Matt Stauffer: Personally, I would say stay small. I think that the number one disappointment with AI is when people are like, do this big feature and it's terrible and you spend more time fixing the big feature than you had. And so like I would say, yes, start small and also stay as small as possible until you really have a good handle on it or maybe just forever. Give it very explicit directions. Imagine that you had just hired a new junior programmer and you wanted to onboard them into the code base by telling them, do this and it should look like this. And you could write the code. Ashley Hindle: Yeah. Matt Stauffer: You know what the code should look like. You know what a happy or unhappy output it is. Start there. Ashley Hindle: Yeah, yeah, 100%. Treat it like a pair programmer, right? You wouldn't go to a pair programmer and say, do this giant feature with no context and no information and no, maybe you should look at this file. So yeah, I agree. Start small, share context. Here's a reference file. These, when you write this test, look at this test. It does a very similar thing. Yeah, you get great results that way. Matt Stauffer: Next question is will Boost aid in authoring our own MCP servers in Laravel and you had already kind of mentioned a little bit about how Boost can pull in our own rules. Does Boost aid in making our own servers? Ashley Hindle: Not directly, but to be able to build Boost, we had to build an MCP server package. So when Boost releases, and it's released now, the MCP server package will also be released. And that makes it really, really easy to add MCP servers to your Laravel applications using a root slash ai.php file. You can register your MCP servers in there. It supports local servers via standard input output. It supports remote MCP servers and it supports more... They call them primitives in MCP land. know, Boost just has tools right now, but it does support resources and prompts and other things also. And that is largely, whatever's passed largely, almost completely down to Peter Suhm from Tailwind who contracted with Laravel for a bit. He did almost the entire MCP server package. So it's thanks to him that this is even possible really. Matt Stauffer: Nice, that's awesome. That's fantastic. The next question. Well, you already actually said this, but I just want to give you a chance to kind of explain any more if you haven't. Will packages be able to expose guidelines and docs that Boost can make use of? So you mentioned guidelines, well, so I don't even remember. You said things that packages can do, but I'm very curious. Like a third party package that's not one of the ones that you guys have done the work for us, not Livewire. It's like Ziggy. Like Tighten Ziggy packages a lot of places. Ashley Hindle: Yeah. Matt Stauffer: But let's say it's not big enough or prominent enough for you all to load your own stuff. The packages get to contribute towards it. Ashley Hindle: Yeah. That's very high on the to-do list. I said on stage that packages would be able to provide their guidelines that would get composed with the other guidelines. And that is the path we want to take. But then when I got into it, there are so many gotchas. There are so many edge cases. And I've not solved them yet to make this a nice user experience. So... Matt Stauffer: Okay. Got it. Ashley Hindle: I don't know what this looks like yet, but 100 % we want to do that because I'm just a guy, right? I can't add guidelines for every package, so I am hopeful that packages can provide those guidelines on how AI agents should work with that package. But it might be a suggestion. So we suggest that you install, this guideline to Boost. But then originally I was thinking that it would automatically be added. So we'll figure that out and we'll do like a second mini release that packages can now provide guidelines. But I don't think it's going to be there for launch. Matt Stauffer: And our last question from the team is How do you intend to keep updates to all of the knowledge tools in sync with new releases? So for example if a new patch release happens like Laravel adds an attribute is the process automated for updating the Boost tools especially this is probably especially about the docs so know about this updated feature or is there somebody gonna have a job to basically hand update the MCP tools knowledge and prompt it to go learn something new every time? Ashley Hindle: Yeah, so the documentation is re-ingested and checked every day almost late. So we're just going to automatically know about that. But any guidelines, they're manually created and curated. So if we were to make a change to Laravel 12 guidelines and say, now our preference is that you use this attribute, not this old way, we'd have to manually go and do that. But the... Matt Stauffer: Got it, okay. Ashley Hindle: I don't know what the word is. One of the aims with Boost is to be pretty agnostic. So none of the Boost guidelines say, should do it this way. Because we don't know your project, we don't know your agency, we don't know your business. So we try and be quite general. We're not trying to impose too much. So we wouldn't go and say, you should now use this attribute unless the old way was deprecated. And then we would help in that way. But we try to stay out of those wars, you know? And we're doing a good job so far, I think. Matt Stauffer: Grateful for that. Yeah, yeah, I'm definitely imagining that. Because like we so we have built tools for code style at Tighten before including Duster, which I built recently. And that's what we use in all our projects right now. And one of the goals with Duster is to make it as easy out of the box as possible without making it so you have to agree with Tighten's goals and guidelines and know, opinions about because not everybody agrees with our code style opinions, right? So it it was an easier challenge there. Ashley Hindle: Yeah. Yeah. Matt Stauffer: because what I did with duster is I shipped a Tighten rule set out of the box with a very easy ability for you to either take the Tighten rule set and diverge from it or create your own from scratch. I don't know that guidelines systems are gonna make that really easy on you. And it sounds like you're at least thinking about that a little bit, but let's say even with all your best attempts to keep everything as agnostic as possible, you ship a preference in a guideline and I'm like, we don't agree, you know. Ashley Hindle: Yeah. Matt Stauffer: Let's say it's, because I was just talking about code style, let's say you specifically mentioned somewhere that every time you're newing up a new instance of a class, you should put the parentheses even if there are no parameters being passed. And I'm like, that is terrible. I hate those stupid parentheses because I've had to fix that in a code style tool before. Ashley Hindle: Yeah, of course. Matt Stauffer: Is it going to be built in such a way where on a project by project basis, because you mentioned it's going to consume the project specific guidelines where I can just say, disregard previous instruction about blah, blah. Ashley Hindle: Yeah. Matt Stauffer: Don't put the parentheses and are you kind of thinking through that? Ashley Hindle: Yeah, we are thinking through that. There's no clean solution, but there a couple of solutions. So one way is a project-specific guideline that says, actually, the previous guideline was wrong. You must never do this or else my entire project will set on fire and I'll be really sad. And then, you you should take that into account. Matt Stauffer: It's sort of like the exclamation important in CSS. You just, if you say it strongly enough, it's going to prioritize that one over the previous ones. Ashley Hindle: Yeah. Exactly, you z-index 999 it and that gets priority, right? We do currently have a way to exclude tools. So if you don't want the browser logs MCP tool, you can exclude that tool and it just won't get registered, right? So I am wondering, should you be able to exclude guidelines? Maybe you don't like our guidelines for Pest and you're happy to write your own. Matt Stauffer: Yes, exactly. Ashley Hindle: That's fine, right? Just exclude them and then include your own project guidelines. So there's also that. But the individual lines, like that would be one line in a hundred line markdown file, right? I've not figured that out. I don't know yet. But hopefully those two solutions are enough. Matt Stauffer: Right. Yeah. Yeah. Well, yeah, and I like what you said about if your language is gentle enough and then that gives us the ability to use more intense language. Also, that happens later. Hopefully we're able to kind of use that to unset things. But it's very interesting to see if we end up coming up with like some kind of format for instruction. Because like all this stuff is new, right? Like, you know, Boost will be six months old and half of the ecosystem will be completely transformed from there. So. Ashley Hindle: Yeah, and nobody knows what they're doing, right? We're all figuring this out together. So maybe the guidelines are worded too strongly. And people can add an issue in a PR and we'll change it, right? Like, you know, we're super open. We want to make this incredible. Like, I'm not happy with decent. I want this to be after you do Laravel New, you do Composer Require Boost. Matt Stauffer: Yeah, yeah. Ashley Hindle: Like that's my goal. I want it to be that straightforward. So yeah, hopefully we can figure all this out together. Matt Stauffer: Yeah. OK. So yeah, and I love that. You're not just kind of like, I'm the person telling you what to do. It's like, I'm throwing something out there, and let's figure it all out. I've been operating the assumption the whole time, which I think is correct. That's why it's an assumption. But I want to verify with you, especially because I do think that MCP servers are a little hard to wrap your head around at the beginning of understanding them, that... Ashley Hindle: Yeah. Matt Stauffer: Boost's MCP servers are exclusively for the use of developers working on this application, and none of them have the intention of being exposed to the general public once the application goes live. Ashley Hindle: 100 % correct. Yeah, when you add Boost, you do Composer Require Laravel Boost, hyphen, hyphen, dev, and then you run Artisan Boost install, and that will add boost into your various IDs and agents for you automatically. And yeah, it never touches production. Matt Stauffer: Okay, if someone were to want to build an MCP server that they're exposing the general public, for example, somebody authenticates with the tool and you know, in their AI and now their AI can ask questions of like the state of their, you know, their, their account in your tool. Do you think that the, the underlying MCP server tool would be something they can do that with, or is that tool also intended primarily for local development? Ashley Hindle: Yeah, Boost is all local. So then you would add the MCP server package that we have, and then you would use that. Matt Stauffer: That's what I was, yeah. So that you could build something on that, that that that package is not intended only for local, right? That's just for everything. Ashley Hindle: no, that works for everything. We are hoping that people use the remote element of it so it's available via HTTP URL. And then you can add those now really easily into Claude AI, into ChatGPT, into all of these tools. They support those remote MCP URLs really well now. Matt Stauffer: Yeah, very cool. Okay. I am out of my questions, but is there any aspect of Boost, AI, your journey, or what you're excited about right now that we did not get a chance to cover today? Ashley Hindle: I don't think so. What comes to mind is how awesome it is to work with such awesome people. Like, I've not done this alone. I've... Peter Suhm did a ton of work. Somebody I used to work with, Katherine, Binary Kitten, she's helped with the ingestion of docs and tidying up my mess. Pushback has joined Laravel recently, he's already ingested docs for so many packages and improved that code. Yeah, so when I left remote, I had this period of trying to figure out life. And that was pretty lonely, right? Just randomly sharing videos, hoping people saw them. And so now it feels like a really special place to work with incredible people, right? That's pretty special. And I've realized I need that in my day to day. So yeah, I'm grateful for that. Matt Stauffer: That's amazing. What a wonderful thing to end our episode on, I'm glad you're here, my friend. I'm glad you're here. Ashley Hindle: Thank you. I'm glad you're here, Aaron Francis, taking over all of our little idioms. Matt Stauffer: Cool, well thank you so much for hanging out. Thank you for your work here. If anybody is trying out Boost and they want to either follow along with development or they're like hey that Ashley guy is really great or if they want to kind of give you some feedback or collaborate with you how do they get a hold of you? How do they keep up with what you're doing? Ashley Hindle: Yeah, probably best on Twitter, Ashley Hindle. Straightforward, I'm spending way too much time there. And when Boost is released, I'll be eagle-eyeing all of the issues and all of the PR so people can use that. And actually Boost, one of the tools is to report feedback. So if you ask your AI agent, hey, tell Boost this thing sucks, that will come directly to me. I'll cry for four seconds and then I'll fix it. So... Matt Stauffer: Nice. Ashley Hindle: There's another way. Matt Stauffer: It's just for some reason, this is the stupidest, I don't know why my brain is here. I was like, I need to add, need to PR another tool in Boost that just says like, send Ashley Hindle like a little, you you're doing great emoji that we can all do, so. Ashley Hindle: Now I'll approve that immediately. Yeah, that'd be great. Matt Stauffer: Fantastic. little hugs, what do they call it? Hug ops, a little hug ops emoji. Awesome. Ashley Hindle: Yeah, I love it. Matt Stauffer: Well, thank you so much for hanging out and for the rest of you, thank you for joining us. Ashley, I was going to say we're all looking forward to this, but they already have access to Boost. So y'all, I'm looking forward to it because it's coming out in a couple of days. The rest of y'all make sure you go take a look at Boost and give Ashley some great feedback. S Ashley Hindle: Thank you so much. Matt Stauffer: Okay, see y'all later.