PHP Alive And Kicking: Episode 4 - Ashley Hindle 🔗 **Source:** [PHP Alive And Kicking: Episode 4 - Ashley Hindle](https://www.youtube.com/watch?v=anOlogVgOlo) > [!note]+ Document Notes > tags: [php, ai, laravel, development-tools] > author: PHP Architect > source: YouTube > reference: https://www.youtube.com/watch?v=anOlogVgOlo > date created: 2024-12-29 18:30 > > # 🏷️ Title: PHP Alive And Kicking: Episode 4 - Ashley Hindle > <!-- deep-read:sections start --> ## 🧰 Action Inbox - [H1] Install Laravel Boost to improve AI pair programming quality — adds MCP tools, guidelines, and documentation API #laravel #ai - [H2] Add post-update command for automatic boost updates — run `artisan boost:update` after package changes #maintenance #boost - [H3] Develop AI usage skills as core competency — treat like databases or testing frameworks #professional-development #ai ## 🧭 Table of Contents - [📝 Summary](#-summary) - [📖 Main Content](#-main-content) - [🗒️ Notes](#️-notes) - [📚 References](#-references) - [🗣️ Paraphrase](#️-paraphrase) - [📰 Newsletter](#-newsletter) - [✅ Action Items](#-action-items) - [🧪 Hypotheses](#-hypotheses) - [📊 Methodology](#-methodology) - [❓ Questions](#-questions) - [🔮 Future Research](#-future-research) - [💡 Implications](#-implications) - [🧠 Learning Styles Assessment](#-learning-styles-assessment) - [📊 Accessibility Assessment](#-accessibility-assessment) - [🌍 DEIB Assessment](#-deib-assessment) - [✅ Logical Validity](#-logical-validity) - [🛡️ DARVO Analysis](#-darvo-analysis) - [🎯 Conclusion](#-conclusion) ### 🔗 References - [Laravel Boost Documentation](https://github.com/laravel/boost) - [PHP Architect Magazine](https://phparch.com) ## 📝 Summary - Ashley Hindle, AI Lead at Laravel, discusses his journey into PHP and the development of Laravel Boost - Laravel Boost enhances AI pair programming by providing MCP tools, guidelines, and documentation APIs - AI is positioned as a powerful tool requiring skill development rather than a job replacement threat - The episode explores pricing models, code quality standards, and the future of AI-assisted development - Bottom line: AI tools like Boost make developers more productive when used skillfully, but human expertise remains essential ## 📖 Main Content Ashley Hindle, known as "a strange bloke that does things" and AI Lead at Laravel, shares his unconventional path into PHP development starting from teenage hacking aspirations. The conversation centers on Laravel Boost, a comprehensive tool that enhances AI pair programming through three core components: MCP (Model Context Protocol) server tools, behavioral guidelines, and a documentation API with 20,000+ chunks from 11 Laravel ecosystem packages. The discussion addresses concerns about AI replacing developers, concluding that AI serves as a productivity tool requiring skill development. Hindle emphasizes that "AI isn't taking your job, but a similarly knowledgeable developer who uses AI better than you is going to take your job." The conversation explores how AI changes development workflows, pricing models, and code quality standards, with particular attention to the importance of human oversight in reviewing AI-generated code. ## 🗒️ Notes #### ⏱️ [00:03:15](https://www.youtube.com/watch?v=anOlogVgOlo?t=195) — [H1] Ashley's Origin Story: From Hacking to PHP - Key idea: Started with hacking tools in C, needed website to showcase work, discovered PHP for dynamic content - Evidence: "I built some open source tools with C to crack passwords from Windows machines back in the day... I wanted to show people how cool I am. So, I'm going to have to build a website" #### ⏱️ [00:05:42](https://www.youtube.com/watch?v=anOlogVgOlo?t=342) — [H2] Laravel AI Team Formation - Key idea: Dual role with Nuno focusing on making AI integration natural and straightforward for developers - Evidence: "It's just Nuno and I are working on all things AI at Laravel... to make working with AI with Laravel really really easy for people" #### ⏱️ [00:07:18](https://www.youtube.com/watch?v=anOlogVgOlo?t=438) — [H1] Laravel Boost Architecture - Key idea: Three core components - MCP server tools, behavioral guidelines, and documentation API - Evidence: "It comes with an MCP server which has various tools... a bunch of guidelines... and a documentation API that we built where we ingested lots of documentation" #### ⏱️ [00:09:33](https://www.youtube.com/watch?v=anOlogVgOlo?t=573) — [H2] Documentation Scope and Versioning - Key idea: 20,000+ chunks across 11 packages with version-specific documentation - Evidence: "I think we have 11 packages in the Laravel ecosystem that's chunk to over 20,000 chunks of information... Laravel 10, 11 and 12... pest three and four... filament 2, three and four" #### ⏱️ [00:12:45](https://www.youtube.com/watch?v=anOlogVgOlo?t=765) — [H3] Package Documentation Integration Challenges - Key idea: Adding third-party package documentation requires manual ingestion but roadmap includes easier integration - Evidence: "Right now you can't offer it through boost... we would have to write that... but it is on our list" #### ⏱️ [00:16:20](https://www.youtube.com/watch?v=anOlogVgOlo?t=980) — [H2] Custom Guidelines System - Key idea: Multiple ways to customize Boost behavior through local guidelines and overrides - Evidence: "You can override Boost's built-in guidelines... if your project has a/Guidelines directory with blade files in, we'll ingest those during boost install" #### ⏱️ [00:19:15](https://www.youtube.com/watch?v=anOlogVgOlo?t=1155) — [H3] Boost Update Mechanism - Key idea: New boost:update command handles version changes automatically - Evidence: "Nuno actually just released boost update... you can add artisan boost update to your post update command in composer.json" #### ⏱️ [00:22:30](https://www.YouTube.com/watch?v=anOlogVgOlo?t=1350) — [H2] Hybrid Search Implementation - Key idea: Combines vector search with traditional full-text search for optimal documentation retrieval - Evidence: "We also combine a regular full text search with typo tolerance and stemming... vector search then union the results and rerank them" #### ⏱️ [00:28:45](https://www.youtube.com/watch?v=anOlogVgOlo?t=1725) — [H1] AI Job Displacement Reality Check - Key idea: AI tools don't replace developers but skilled AI users may outcompete those without AI skills - Evidence: "AI isn't taking your job but a similarly knowledgeable developer who uses AI better than you is going to take your job" #### ⏱️ [00:32:10](https://www.youtube.com/watch?v=anOlogVgOlo?t=1930) — [H2] Vibe-Coded Application Problems - Key idea: Non-developers using AI create applications requiring professional fixes - Evidence: "There are people calling themselves vibe code fixers on LinkedIn and then charging more if the code is written by AI" #### ⏱️ [00:35:55](https://www.youtube.com/watch?v=anOlogVgOlo?t=2155) — [H3] Clean vs Readable Code Standards - Key idea: Preference for boring, readable code over sophisticated abstractions - Evidence: "I use the word boring. I like my infrastructure and my code to be boring... really really easy to read" #### ⏱️ [00:41:20](https://www.youtube.com/watch?v=anOlogVgOlo?t=2480) — [H2] Pricing Models Discussion - Key idea: Shift from hourly to value-based pricing becomes more relevant with AI productivity gains - Evidence: "There's two ways you can charge for something. You can charge for the result or you can charge for the hours spent doing it" #### ⏱️ [00:46:30](https://www.youtube.com/watch?v=anOlogVgOlo?t=2790) — [H3] AI Learning Methodology Concerns - Key idea: Risk of delegating thinking to AI versus using AI to enhance understanding - Evidence: "Some people that use AI in the wrong way and delegate their thinking to the AI and then some people that use AI the correct way" #### ⏱️ [00:50:45](https://www.youtube.com/watch?v=anOlogVgOlo?t=3045) — [H2] Future Economic Implications - Key idea: Uncertainty about economic impacts if AI eliminates significant job categories - Evidence: "What does the future look like if AI takes 20% of jobs and people still need to eat? I don't think we're going to get a universal income" ## 📚 References - [Laravel Boost GitHub](https://github.com/laravel/boost) - [PHP Architect Discord](https://discord.phparch.com) - [Dave Hicking's Laracon Talk on AI](https://laracon.us) - [Model Context Protocol (MCP)](https://modelcontextprotocol.io/) ## 🗣️ Paraphrase - Laravel Boost enhances AI development by providing tools, guidelines, and documentation access that makes AI coding more accurate and context-aware - AI represents a skill that developers must learn rather than a threat to employment, similar to learning databases or frameworks - The future involves more AI-generated code requiring human review and oversight, creating new roles like "vibe code fixers" ## 📰 Newsletter Laravel's new Boost package is revolutionizing AI-assisted PHP development by solving the context problem that plagues AI coding tools. With integrated documentation from 11+ Laravel ecosystem packages and smart behavioral guidelines, Boost helps AI agents write better, more accurate code with fewer hallucinations. Key benefits include automatic package detection, version-specific documentation, and customizable guidelines for your development environment. Whether you're using Sail, Lando, or custom setups, Boost adapts to your workflow. The discussion around AI's role in development continues to evolve, with experts emphasizing AI as a productivity multiplier rather than a replacement for skilled developers. Resources: [Learn WordPress](https://learn.wordpress.org/), [Developer Docs](https://developer.wordpress.org/), [WP-CLI](https://wp-cli.org/) ## ✅ Action Items #### DevRel - Developer advocacy: Create tutorials showing Laravel Boost integration workflows - Education: Develop workshops on effective AI pair programming techniques - Developer experience: Improve MCP tool documentation and examples - Developer marketing: Case studies of productivity gains with Boost - Community management: Facilitate discussions on AI coding best practices #### OSPO - Policy & Governance: Guidelines for AI-generated code review and acceptance - License Compliance: Review AI training data licensing implications - Security Management: Audit AI-generated code for security vulnerabilities - Software Inventory: Track AI tools and dependencies in SBOM - Developer Support: Training on responsible AI usage in development - Community Engagement: Support open source AI tooling development - Strategic Alignment: Metrics for AI productivity and code quality impact #### Community - Maintainers: Guidelines for AI-assisted contributions to open source - Contributors: Onboarding for AI-enhanced development workflows - Sponsors: ROI analysis of AI tooling investments - Ecosystem: Coordination on AI standards and best practices #### WordPress - Core: Evaluate AI tools for WordPress core development - Core Performance: AI-assisted performance optimization workflows - Core AI: Integration of AI capabilities in WordPress core - Documentation: AI-enhanced documentation generation and maintenance - CLI: AI-assisted WP-CLI command development and testing - Plugins: Guidelines for AI-generated plugin code review #### The WPCC - Programs: Fellowships focused on AI and WordPress development - Projects: AI tooling grants for community projects - Funding: Support for AI accessibility in development tools #### FAIR - Technical independence: Ensure AI tools don't create vendor lock-in - Policies: Transparency requirements for AI-generated contributions - Governance artifacts: AI usage policies and contribution guidelines ## 🧪 Hypotheses - AI pair programming tools will become as essential as IDEs within 2-3 years - Developers who don't adopt AI tools will experience decreased market competitiveness - The volume of code requiring human review will increase significantly due to AI generation - Value-based pricing models will become more prevalent as AI accelerates development timelines ### 📊 Methodology - Survey developers on AI tool adoption rates and productivity impacts over time - Measure code quality metrics comparing AI-assisted vs traditional development - Analysis of pricing model shifts in development services industry - Case studies tracking career trajectories of AI-adopting vs non-adopting developers ### ❓ Questions - How can development education adapt to include AI literacy as a core skill? - What safeguards prevent over-reliance on AI at the expense of fundamental understanding? - How will code review processes evolve to handle increased volumes of AI-generated code? ### 🔮 Future Research - Investigation into optimal human-AI collaboration patterns in software development - Economic modeling of AI impact on developer employment and compensation - Development of standards for AI-generated code quality and security assessment ### 💡 Implications - Practice: AI pair programming becomes standard workflow requiring new skills - Policy: Need for governance around AI-generated code in production systems - Security: Increased vigilance required for AI-generated code vulnerabilities - Education: Curriculum updates to include AI collaboration techniques - Funding: Budget allocation shifts toward AI tooling and training ### 🧠 Learning Styles Assessment Reading level estimate: Grade 10-12 #### 👀 Visual learners - Present: Limited visual aids in podcast format - Improve: Architecture diagrams of Boost components, workflow visualizations #### 🔊 Aural learners - Present: Full audio podcast format with natural conversation flow - Improve: Chapter markers for easy navigation to specific topics #### 🗣️ Verbal learners - Present: Extensive explanations and analogies throughout discussion - Improve: Glossary of AI and development terms used #### 🧩 Kinesthetic learners - Present: Code examples and tool installation instructions mentioned - Improve: Step-by-step Boost installation and configuration guide #### 🧠 Logical learners - Present: Clear cause-and-effect discussions about AI impact - Improve: Decision trees for when to use different AI approaches #### 🧑‍🤝‍🧑 Social learners - Present: Discord community mentioned for ongoing discussion - Improve: Structured discussion prompts for community engagement #### 🧘‍♀️ Solitary learners - Present: Individual reflection on AI tool adoption - Improve: Self-assessment rubrics for AI skill development #### 🌿 Naturalistic learners - Present: Limited environmental connections - Improve: Discussion of AI's energy and sustainability implications ### 📊 Accessibility Assessment #### Screen reader semantics - Present: Audio format naturally accessible to screen readers - Improve: Detailed show notes with time-stamped topic headers #### Keyboard navigation - Present: Standard YouTube player controls available - Improve: Custom transcript navigation with jump-to-time links #### Contrast and readability - Present: Audio content doesn't require visual contrast - Improve: High contrast transcript formatting options #### Text resizing and reflow - Present: YouTube's standard accessibility features - Improve: Responsive transcript layout supporting 200% zoom #### Media and captions - Present: Auto-generated YouTube captions available - Improve: Professional transcription with technical term accuracy #### Forms and interactions - Present: Standard YouTube interaction elements - Improve: Accessible feedback forms for episode topics #### Standards reference - YouTube player meets WCAG 2.1 AA for media controls ### 🌍 DEIB Assessment #### Cultural breadth - Present: Global remote work perspective from PHP Architect team - Improve: More diverse guest perspectives on AI adoption worldwide #### Inclusive language - Present: Casual, welcoming conversation tone - Improve: Conscious use of technical terms accessible to newcomers #### Neurodiversity support - Present: Conversational format with natural topic flow - Improve: Clear topic segments with consistent structure #### Representation - Present: Male-dominated technical discussion - Improve: More diverse voices in AI and development discussions #### Psychological safety - Present: Supportive tone about learning and career development - Improve: Explicit welcome for developers at all skill levels ### ✅ Logical Validity <!-- deep-read:sections end --> ## 📺 Video  --- # 📄 Full Transcript *Auto-generated from YouTube captions · timestamps roughly every 30 seconds · [watch on YouTube](https://www.youtube.com/watch?v=anOlogVgOlo)* **[00:00]** [Music] Hello and welcome to PHP Alive and Kicking brought to you by PHP Architect and made better by our partners at PHP Score, but more about them later. This podcast explores the new and exciting developments within PHP and what it's like to earn a living in the PHP world **[00:33]** today. I'm one of your hosts, Mike Page, better known as Mike Page Dev, and my co-host is a longtime contributor to our magazine and now team member Chris Miller. Hello, Chris. >> Hey, Mike. How you doing? >> I'm good, thanks. Uh, today's episode is a pre-recorded episode, but you can always join us in our Discord channel. Uh me and Chris will be hanging out there. In fact, we hang out there all week. There's great people in there. Uh package uh um maintainers, internal PHP **[01:04]** internal. Come on, join us, ask questions. We're all there to to help each other in the discord. Uh for that is discord.phparch.com. Uh we are on YouTube as well. Head over there, subscribe to the channel and uh like any of the videos. great would be great. That's youtube.comparch. We have for our lovely listeners a um a discount for our magazine. **[01:36]** There it is. Alive three will give you three months free off a annual subscription to the digital. I remember the digital this time, the digital edition. Uh we also run a a uh conference PHP Tech. Uh tickets are now open for next year's tech. It would be lovely to see some of you people over there. That would be brilliant. And don't forget to uh check out our swag over at uh **[02:07]** store.phparch.com. You can find some t-shirts and hats and and bottles and and and the likes. Okay, I think that is all our housekeeping ticks. >> I think so. >> Why don't we bring our new our guest in today? >> Yeah. So, today we have a guy who insists on being introduced as a strange bloke that does things. Um, I think we could also introduce him as AI lead at Laravel. We have with us Ashley Hindel. >> Hey, Ash. **[02:39]** >> Pretty great. Great to be here with you both. >> Good. Well, thank you for joining us. Thanks for having me. I'm excited to talk about AI and PHP and all those lovely things, >> which is exactly why we do this podcast, to talk about PHP and cool stuff. And I'd say AI and PHP kind of go together these days. It's it's everywhere. >> We're heading that way. Hey, >> so we always start our podcast with exactly the same question. How did you **[03:10]** get into PHP? Yeah, I got into PHP through a weird route when I was 13, 14 maybe. My actual plan was to be a really cool hacker that you would see in movies and um you know, I'd wear all black and a hoodie and I'd be evading the cops and doing social engineering and hacking things, you know, to phone lines and things. Um, and I and I built some open **[03:43]** source tools with C to crack passwords from Windows machines back in the day. Um, even though they already existed, I built another one. And then I was like, well, I need to show people how cool I am. So, I'm going to have to build a website. And so, I built a HTML website that showed how what a great hacker I am. You know, I had um Le Speak on my website. you know, I had a marquee scrolling things. Um, I was, "Oh, this **[04:13]** is this is pretty cool." You know, I had images, image maps with different, you know, clicking different areas. But then to show how cool I am even more, I wanted to have like a hit counter or a guest book or something. Well, how how do I do this in HTML? Where where does that get saved? like who who owns that piece of data so that I can show it and then people kept talking about PHP and **[04:44]** naturally before that you know people were talking about Pearl and PHP and CGI bin and I had no idea what was going on and then I accidentally got PHP to work and I thought oh this is really cool and then yeah I wanted to throw into the database so then I had to learn about my SQL and and things like that you know to store the It's, oh wow, I'm a real developer. I had include header.inc.php, include footer.inc.php. Uh, you know, I had I could stop **[05:16]** repeating all of that HTML. And then from there, I was like, oh, well, I quite like writing PHP and websites, so maybe I'll leave the hacker business behind me and I'll just, you know, make websites and cool things. And so I've been doing that now for over 20 years and yeah, it's it's great. >> So effectively you got into PHP because hacking wasn't quite as cool as you thought. >> Exactly. And and harder than I thought. **[05:48]** The people in movies make it look so easy. >> Have you ever actually tried to do the stuff that's in movies, though? I mean, all it is is stand there, mash the keyboard, and then say, "I'm in." That is the entirety of hacking in movies. >> Yeah. it it didn't work when I tried it. It turns out you have to methodically test lots of different things and lots of different strings and it's quite boring just looking, you know, for different attack vectors to take advantage of. And so, yeah, PHP and **[06:18]** HTML, JavaScript, that feedback loop is much faster and much easier. So, yeah, I'll stick with that for now. >> So, I guess that sort of leads us down the path to where you are now. You're actually part of the Laravel core team and uh AI team lead. >> What does that entail? >> It's weird. Hey, it's great. Um basically it's just Nuno and I are working on all things AI at Laravel just **[06:49]** so to help people kind of pair program with AI and to build things with AI that will help people in the ecosystem. So our job really is just to make working with AI with Laravel really really easy for people like really natural, really straightforward and take away all of the complexity so that uh yeah people can just to think of AI as a tool rather than learning tons and tons of complex **[07:20]** things unless you know you're interested in that. So yeah, we've got a lot to do. So mainly my role, Nuno's role is code a lot and then speak to a lot of people, see what problems they're having and then code a lot. We're just on repeat doing that basically. >> Sounds like agency life to be fair. >> Yes. Yeah. >> So talk talk to customers, find out what they've broken, and then fix it. Yeah. **[07:51]** Sounds about right. >> Mhm. So I know one of the tools that came out is Laravel Boost. Um >> I have installed it. That is as far as I've got so far, but I have installed it. >> Oh, I've done a I've gone a little bit further. I have installed it and I have started using it to uh to try and help sort out the lack of my uh ability to write tests. >> It is very helpful there to be fair. Well, AI in general is very helpful **[08:22]** there. Mhm. Yeah. >> Yeah. So, why don't you tell us uh the the story behind Boost and what Boost is? >> Yeah, Boost is an interesting one for me personally because Boost is a package that you install into your Laravel projects that makes your AI agents output higher quality code. So coding with an AI agent, with an AI pair **[08:54]** programmer is just smoother and better with Boost. And you know, it makes less mistakes, it hallucinates less, it is faster, it writes test, it proactively formats code, you know, it it's just better. That's the goal. Um, but it and it's got three core parts. So it comes with an MCP server which has various tools that your AI agent can use like running arbitrary PHP code to debug things or querying the database checking **[09:24]** the database schema for indexes and things like that. Um and it comes with a bunch of guidelines like rules to tell the AI how to behave and it also comes with a documentation API that we built where we ingested lots of documentation from the ecosystem. We chunked it and vectorized it, created embeddings for it, put it behind Typesense and then made that available to your AI agent as well. So it gets the exact version specific documentation that your project needs and that's like so helpful so **[09:58]** valuable and I joined Laravel July 3rd I want to say something like that and um and I did some work on our MCP package first and then I started building boost a couple of weeks into my role and then we announced it at the keynote in Laracon US and then we launched it sometime in August. But before I joined Laravel, I built something similar, but only the MCP part with tools called Croft. And I **[10:31]** released Croft at Laravel Live UK earlier this year. I don't remember which month that was. Um, but we released Croft and that was great. It had similar tools to boost, but it didn't have all of the AI guidelines and all of the documentation and things like that. And then when I joined Laravel, everything felt there was more that was possible, I guess, because like I **[11:02]** couldn't personally pay for a server to ingest all of the documentation every day, then chunk and embed it and then put it behind Typesense and make sure it's always available and it's, you know, unlike a CDN, so it's faster depending on where in the world you are. But Laravel can pay for that because they want everybody's to be more productive, right? And so all of a sudden it was like, oh, we can do more. I'm not like hamstrung by just being this weird bloke who has, you know, an **[11:33]** hour an evening to do things. And so when I realized that, that's when we wrote all of these guidelines for all of these different packages. That's when we thought, let's just ingest all of the documentation and make that really easily available to the AI agent. Um yeah and so that's where we are now you know so I started with kind of my own version of a subset of boost >> but then it's became a lot better because of what's available at Laravel and the space that I have and the **[12:03]** obsession that Taylor has to increase productivity. >> Um yeah so it's in a really good place right now. >> It's almost like Taylor keeps saying we must ship it. >> It's crazy right? Yeah. and then backing that with tools and and budget and people you know to actually help which yeah is really cool to see. >> So so what sort of documentation have you got already um that you're already referencing? What what **[12:36]** >> so I think it's I think we have 11 packages in the Laravel ecosystem that's chunk to over 20,000 chunks of information uh in that documentation. >> Wow. >> And so we have things like Laravel 10, 11 and 12. All of that documentation >> we have pest three and four. We have filament 2, three and four. I think we have tailwind three and four. Um we have livewire flux and some other **[13:06]** things I forget. Um so we ingested all of their documentation and um separate it by major version at the minute. So if you use LiveWire 3, you only get Live Wire 3 um documentation. M so yeah but I really want to get to major minor at the very least >> um and ideally patch but I've looked into that and it's so much work it's so complex that >> that might be in a year or two. **[13:39]** So I have a question then. Let's say that I I don't but let's say that I have a package out there that's got a full documentation system and I know that a lot of people are using it. How would I then offer the same facility through boost? So I don't know let let's say the Laravel permissions package from >> spaty. So right now you can't offer it through boost like we've we have in we wrote ingested for specific documentation for **[14:09]** the boost back end. So we would have to write that. Um but it is on our list like we have been speaking to package developers of how could we make it so they can just add their docs really really easily >> so that that just works. You just installed the sparse permissions package. you get their guidelines for how to use that package and you get the documentation uh of exactly how to use that package for the exact version that you have and we have some ideas there **[14:40]** >> but it's it's really tricky we've not solved it yet um but we hope to solve that because I think that's really really powerful when >> just every library that you use >> your AI agent can get the exact documentation that it needs Right. >> So my my brain goes into the idea of perhaps doing local tokenization. So take the documentation locally, allow Boost to set up a tokenization server to do it locally for a specific user. So **[15:11]** just point it at the URLs and go. That sounds really easy, but I've built a tokenizer. It's not. >> Yeah, that's tricky. I mean, I I still find it fascinating the idea that you can turn the entire Laravel docs into a bunch of numbers that have meaning and it it is crazy, right? Like I find it really bizarre also that you just get an array and in my case I'm I'm using an **[15:42]** OpenAI embeddings model with 1536 vectors. So >> a word becomes 1,500 numbers in an array. That's how how make that make sense. >> But then it it's not even just a word because I assume you're using then the next layer up as well. So you're using context aware vector. So you you have semantic meaning embedded in in vectors. So you you have you have the ability to **[16:14]** understand to a certain extent the context in which something is used. So you know that it is just an example or you know that it is here is exactly how to do this >> and yeah >> it's it's incredible to think that that can come from a bunch of numbers and it's really cool and also what I found really interesting as incredible as embeddings and retrieval augmented generation is it's not enough to have a **[16:47]** valuable documentation search API. So we also combine a regular full text search with typo tolerance and stemming and things like that and vector search then union the results and rerank them so that people get the best results because neither one is good enough itself essentially and so we have to squish them together to get the best results. And we're still fine-tuning that, like, you know, we've not nailed it, but we're getting **[17:18]** somewhere good. I mean that that on its own is incredible that you you're combining the two main routes which it not just gives you a third option but it also gives you the most powerful results possible because if you've got context awareness through the vectors and you've got typo tolerance and you've got to a certain extent because of that understanding tolerance I'm looking for something that is a bit like and if you **[17:49]** look at clause Open AI, anything like that, you can you can start being a little more vague these days of what you actually want >> because it's learned the context awareness of what people are asking about. >> Yeah, that's what because I'm very lazy human. One of the things I like about boost when you have it in your project, you can be a bit more lazy with your prompts. So before boost I had to say I'm using Laravel 12.1 and Tailwind version 3 and I've done **[18:20]** this specific configuration change that I need you to be aware of. But now because Boost kind of injects some of that information and you know we've got tools to help it. I can just say do this thing figure everything out. Uh yeah so that context has been really handy to just help me be much lazier which I've welcomed. See, this is this is the most amazing thing in development, isn't it? That most amazing tools come out because developers are lazy and don't want to do things. **[18:50]** >> Hell yeah. >> It's great. >> Why Why spend 10 minutes coding something when you can spend four hours failing to automate it? >> Exactly. >> Well, on that note, I think it's time to hear from our partner, uh, PHP Score. We'll be right back. Thank you to our partners at phpsscore.com. Every app builds up technical debt over time. It's the price we pay for shipping new features and moving fast. When we build up too much of it, though, it can **[19:21]** start to impact how we work. Team velocity suffers, bugs become more frequent and take longer to fix, and everyone starts to get a little frustrated. The key to managing debt is to measure it. A credit score can help you understand how well you're managing your financial debt. And now there's a credit score for your technical debt. Go to phpscore.com to get a free technical debt score and monitoring for all of your PHP applications today. >> Yeah, thank you. PHP score. **[19:54]** >> So, um when I found it very easy to uh to install uh boost uh like most Laravel things is just so incredibly easy to to get things up and running. Um, one of the things I noticed when if I remember rightly is it boosts uh install. >> Yeah. >> Um, it it um auto um discovers what packages you have uh within within your app uh and the versions which was really cool. Um if I was then to upgrade one of those **[20:25]** packages. So for example, the package I put it into was actually a filament 3. Um, I haven't upgraded to Filament 4 just because I'm waiting for some plugins to update. Um, once that's done, I'll be updating to Filament 4. Does that do I need is there any other anything I need to run to get Boost to recognize that I've updated or does it auto um recognize that I've updated packages within the the the application? >> That's a good question. Partially. **[20:56]** So boost won't remove any guidelines. >> Mhm. >> But it will recognize that you've upgraded to version 4 of filament. >> Okay. >> So what will happen in so Nuno actually just released boost update. So if you run you can add artisan boost update to your post update command uh in composer.json JSON and just every time you do your package management, your guidelines will get automatically updated to what they need to be based on **[21:27]** the packages and versions you have. Now, >> that's cool. >> So, in theory, actually, thinking fully through it, I think that would work magically. I I think you just run our sand boost update or our sand boost install and it will just Yeah. recreate the guidelines with your now current packages and versions. That's that's pretty cool. And so the other question I was I've just been thinking um it is more to do with your development **[21:57]** environment. So there's lots of people use all sorts of different environments to run their their applications. Is there any any plans to add documentation for any of that? Um for example in my application I'm using Lando. Um, if you don't know what Lando is, it's like a it's like a package for Docker. Um, just makes idiots like me be able to use your Docker because I don't understand all of Docker. Um, and it just makes it really easy to use Docker. **[22:28]** But one of the things was I've had to tell um tell the AI I'm currently trying out claude in it um to say instead of running PHP artisan it needs to run Lando Artisan and then whatever the command is. >> Is there any any um sort of scope of or or um ways of adding the documentations for some of those sort of environments that people use? >> Yeah, there are a few ways to do that. Um so on the road map is to support sale **[23:00]** out of the box. So it will do sale RTM rather than PHP RTSM. So we'll be solving that soon. >> But obviously we can't support every thing that exists in the world. And so I I think we'll probably only do that. Um, but you can override Boost's built-in guidelines and you can add your own boost guideline uh guidelines into the boost guidelines uh within your project. So if you if your project has a/Guidelines **[23:32]** directory with blade files in, we'll ingest those during boost install or boost update. And so if you just keep some rules in there saying always do Lando Artisan rather than PHP artisan this is absolutely critical you must do this or the whole world will perish then it should listen to that and that will just get included always into your guidelines and it's just managed through boost then so that's probably the best way but you can also override our guidelines. So if boost says in it, **[24:04]** which I think it does for tests, run PHP artisan test to run the tests, you can put a file in AI guidelines tests and when we run boost install or boost update, we'll use your guidelines rather than boosts built-in guidelines. So if our guidelines are sending the AI agent off track a little bit, >> just replace them. Just put your own in there and it it will always be correct. So they're probably the best ways right now. And then the last way, this is way **[24:36]** too many ways, maybe we should simplify it, but Nuno just released the ability for uh third party packages to provide guidelines that can go into the guidelines. So Lando in theory could have I don't know how it's built, but they could provide um guidelines to boost. So if you have that installed, they'll just automatically get included as well. So there is a possibility there of it just installing a composer package and that gives you all of the guidelines you need to make everything work well. **[25:09]** >> Interesting. Yeah. >> And I guess you've also got the next layer on top of that of the guidelines that your own AI system can use like claude.md for clawed code and all of the rest of it. And you you could go for a lot of MD files that end up telling it how to do things. Yeah, that's why I personally put things just inai guidelines and then everything is just managed by boost so I don't have to think about it and like so I >> the one that usually I have only two of my own additional rules actually. So in **[25:41]** some projects I use the action pattern. So I'll have a/guidelines.blade.php with my guidelines to use the action pattern and then I'll have a TDD uh guidelines as well. some projects I prefer to do TDD but not all of them and so I'll put that in there and then when I run boost install you know those guidelines get auto magagically added and uh have you have you got any documentation on boost out there yet for for people to sort of see these sort of **[26:13]** example maybe some code examples of how to do um do some of this >> yeah we have the readme which documents how to do all of these things. >> And we do want to move that to laravel.com >> like get it where every other piece of documentation is. Um but that's just on the list and yeah, we've not got to that yet. >> Okay, cool. Uh if you got any any more spec uh **[26:47]** there I can't talk. Any specific questions about boosts? Uh, Chris, before we >> before we move on, >> one thing that intrigues me is >> you obviously you've got the updates and that you can tell it to do the dock updates, but you also said it doesn't delete any guidelines. >> So, yeah, >> say we've got Tailwind 3 guidelines in there. We do update Tailwind 4 and now we've got Tailwind 3 and Tailwind four there. How does it know which one to do? I think **[27:19]** that's why when I was thinking through it, I think actually it ends up in the ideal place. I I I think I was wrong. We override when we add the guidelines to the correct files, we put them in XML tags that say boost guidelines and then we actually replace that entire block with the newly composed guidelines. So actually we don't we replace the guidelines. So actually you would only end up telling **[27:50]** v4 guidelines. >> So it doesn't delete them. It just does full on replace. >> Yeah. >> So I confused myself is that you will end up in the correct place >> but we don't delete them but we just override them. But yeah that's weird. But you'll just end up in the right place. And also because we do that um just replacing of our own guidelines. If you do have a claw.md file and you have your rules above it and your rules below, it will never get in the way of any of your existing rules. So if you do want to manage it just in claw.md **[28:21]** yourself, that's fine. You know, we try to be um mindful of that. >> Okay. It really does seem like it's making it so much easier for people to develop with AI and it kind of gets AI out of the way, which is quite useful because AI is really really stupid if you don't give it good instruction. And >> yeah, >> the the fact that boost is helping to **[28:52]** strengthen that knowledge, it's kind of unique and it does make me question, okay, boost is a starting point of making this good. What's the next step? Where do we go next? Yeah, it's an interesting world because there's more that Boost can do. Like there's more on our road map of things that I would like to have time to work on. And then the AI agents themselves are constantly evolving. Claude Code 2 was released **[29:24]** two days ago. I don't like very recently. And then Claude Sonic 4.5 was released and that's you know now even better than Claude Sonic 4. Um yeah. So I think everything just slightly getting better every week or Fortnite is is a interesting place to be and yeah I'm excited like just everything getting better and not having to explain myself so much or you know AI just understanding a bit better or making **[29:55]** less mistakes um yeah is really really cool. So that kind of leads on quite nicely to um I would say elephant in the room but being it's PHP podcast that feels a bit weird but um a a subject that comes up and I've been asked the question a few times. I know people have asked it in Discord is about um is AI going is AI taking over our jobs? Man, >> that's that's an interesting discussion. **[30:25]** I'd like to dig in with you because you obviously have a lot of um experience with AI. You're very pro- AI, which I think both me me and Chris are. Um my sort of No, don't Chris is shaking his head. He's always using AI. Um no. Um so my my kind of thought on it at the moment is that AI is a tool that we can use. Um, but you need to have some sort of uh programming knowledge to be able to use **[30:56]** it because there it will spit out things that just doesn't work or isn't safe or or um for example >> doesn't exist. >> Yeah. No, there was there was an instant I used I was trying to I was trying to remember the example, but I I asked asked AI to help me. I need to push something out really really quick. And it's just like, can you just write the algorithm to do this for me? This is what I need to come out. And it did it. **[31:26]** And I then looked at it and went, "Hang on, that function doesn't even exist." And that was just that wasn't even needing to know any of the context of the project it was going into. It was more of a it just plainly made up a PHP function that doesn't exist. and I don't think it ever did exist. So I don't think it's like referenced old Stack Overflow answers or anything like that. It's it's got it from another language and got confused with the language. So in my **[31:57]** view is yes, it's good for for pointing us in the right directions, but the way you're now talking about boost, it sounds like it's getting so much more so much better understanding the context that we're asking these questions in. for maybe for example tests where I've I've been using um using it a lot in the last few days is writing some tests. I just say can you write a test for this this uh this class and it will write the test um because it's understanding the **[32:28]** context and the versions that are being used. It seems like it's that that margin error error is slowly disappearing. So >> yeah, >> it is boost taking our jobs. No. Um, an AI isn't taking our jobs, but you could allow it to take over your job. They're slightly different things. I would say there's a very good talk from Dave Hicking that he gave at **[33:00]** Laracon US about AI taking your jobs. And the too long didn't watch version is no, humans are better than AI. our experience, our taste, our decision making is better than AI and will be for a very very long time. So, it's not taking our jobs. AI and boost aren't taking your jobs. what other people say that I haven't self-reflected on how much I agree with it but maybe we can **[33:30]** talk about it is AI isn't taking your job but a similarly um knowledgeable developer who uses AI better than you is going to take your job >> and so everybody needs to be using AI and needs to be skilled with AI just as you are with databases or test-driven development or you calling APIs. It is a skill that you do need to develop. So, I'm pretty sure I **[34:01]** believe in that. But what do you reckon? Like, is that true? >> It's an interesting thought and I I probably agree with you. It's it's it's like anyone um like a Laravel, you're learning how to use Laravel, the the more disk skilled or knowledgeable person is going to be better than you. Um and I think it also kind of goes in hand in hand. What we're trying to promote in our community is helping the **[34:33]** people at the bottom, pulling them up, um teaching them that's however you do that at talks or whatever, just passing on your knowledge. Um the the interesting thought I've had and I've talked to a few people about this. I don't know if I've actually talked to Chris about it, so he can wade in as well. Is I almost feel there's going to be a influx of work out there to fix vibe coped apps because people are with no coding **[35:05]** knowledge is now thinking, "Oh, claw code. I've heard about this. Let's just go type in. I want you to create an app for me that that does this, this, and this." pushes it out there, not checking it, and they're going to go, "It doesn't work." And they're going to have to then pay a a professional developer to work out why it doesn't work. because that that's what I'm I'm kind of predicting in our in our not just in PHP but in the in the world of computer development in **[35:36]** whatever language you're writing whatever you work is it desktop mobile or web um or you're writing programs for a fridge whatever it is anyone that's trying to vibe code without the enough knowledge to know when things aren't quite right that's that's going to be bring more work into the industry and that's what I kind of get the feel for is going to happen soon. Um, from what I've seen on on social media and things like that, >> actually already started happening to a **[36:07]** certain extent. There are people calling themselves vibe code fixers >> on LinkedIn and then charging more if the code is written by AI >> because they have more to do to work out what they actually wanted. owner. I have seen sites that have been vibe coded that are beyond ridiculous. There's flaws like you can just go to slashdashboard. You don't need to log in. You can just go there >> and flaws like yeah that the passwords **[36:38]** are freely available to you for every user >> and things like this. And you can see where what they've done is they've created a prototype which is brilliant. >> And I love being able to say to Cursor, Claude, whatever, make me a prototype website that looks like this and does this. >> I don't care if the functionality is actually there. I care about the flow and the clickth through and does this feel like a website I want. And then the **[37:09]** very next thing I do is probably throw that away and work to actually do what I want. Now I know what I'm looking at >> and a AI code is improving. But is AI code improving or is people using AI improving? Maybe it's both. >> I think it's both. I've found there's a subset of people as well that value code quality less now **[37:40]** because AI can understand your badly written code and won't murder you in your sleep like a colleague would if you had terrible code that they couldn't understand. >> Yeah, I find that I'm used to that. >> Yeah, you asked for the review bike. >> I know. I think that's an interesting world. >> I mean I I am a stickler for clean code and I have literally given talks worldwide on coding using clean code **[38:11]** standards. There's a point at which clean code helps AI, but there's also a point at which clean code gets in the way of AI because it's like well I don't understand understand the depth of the abstraction here. And I'm kind of left thinking, well, maybe what is happening is rather than getting rid of coding standards, we need a coding standard that is not quite so abstracted anymore. I mean, if you look at some of the standards taught by the big names in **[38:43]** that world, the the level of abstraction would mean you'd need six classes before you can have a model. And >> yeah, >> why? Let's just have a model, shall we? Um, and controllers can be extended from abstract controllers which can be extended from the base controller which can be extended from something else. And all that ends up happening is a really complicated setup. Whereas I >> I prefer code that is clean isn't the **[39:14]** word I actually use. I use readable. >> And the core of it being that the code you write isn't for you now. It's for the next person to look at it. And that might be you tomorrow. >> It's usually me in an hour and I've already forgot everything. I I use the word boring. I like my infrastructure and my code to be boring. I don't want it to be fancy just for fancies sake. I want it to be **[39:45]** >> really really easy to read. Yeah. But yeah, not like sophisticated or or anything or advanced just >> yeah crazy boring but really easy to reason right and follow. >> One of the sales maintainers um the sales framework in JavaScript actually released the boring stack. The idea being that code should be boring because if it is boring you can understand it. >> Yeah. >> And it's the same with infrastructure as **[40:16]** well. Like I I'm not very good with serverless or Docker or anything or Kubernetes. Like I've not got into them. And maybe that is boring. >> Good with Kubernetes. >> I think I've met some people who are incredible. But and I think it's boring to them because they're so knowledgeable now like Docker and Kubernetes, you know, is boring to them. But boring for me is PHP, FBM, and Engine X >> on an Iuntu server. And I know that will work forever and won't cause me any **[40:49]** issues. You know, there'll never be a problem. It's just rock solid and boring and it's great. >> So, I'm I'm cading PHP FPM. So, >> okay. >> I'm slightly less boring. >> Yeah, >> that is true. It's but this this is one of the things that I find is brilliant in this community that people can have opinions on code. I mean you you could have completely opposing opinions. You could **[41:20]** argue about it on Twitter but you're still friends at the end of the day. I mean look at look at what happened when Nuno dared to suggest that you don't do a down in a migration. That conversation is still going on three months later and Yeah, >> I think most people are still friends. I think some people take it too far, but yeah, I'm with you. >> Yeah. >> And I mean, I totally disagreed with Nuno on that one because I like downs in my migrations because I want to be able to roll back. Whereas Nuno says you **[41:52]** should never be in the position that you need to roll back. If you need to change something, do a new migration. >> I get it. It's it's a choice. >> Yeah. And there's different risk profiles depends on your >> project, who your customers are, what business you're in, you know, you probably can't do that in a bank. But if you have built a social site like, you know, a Twitter clone, >> it's probably fine. Just roll forward, you know, and take a 5 minute, 10 minute, you know, downtime or something. **[42:23]** >> And it's not like he's written a testing framework, so has really good opinions on that. It's >> No, exactly. Nuno's too good at things. We shouldn't. >> Yeah, it's ridiculous. It's unfair. >> I'm not in the slightest bit mocking him. The guy is the guy's a genius. It's >> Yeah. >> So, in in sort of pulling us back to the the a the AI um in sort of we kind of I think we're all in agreement that AI is **[42:54]** a tool, but you need to learn know how to use it. Um, and if you you need the knowledge to be able to spot when things aren't quite right with AI because it's not perfect. It's not going to >> And this I kind of explained this to someone. Um, in fact, it was it was a bizarre uh conversation I had um with some there was an app that I was using and I'd um I posted a question uh to uh to the developers about the app and they then contacted me and we end up in our Zoom call and now I'm part of the the **[43:25]** testing team. But when they realized I was I was developer and they asked me the same thing. So just what do you think to AI? And my thoughts are we most of these AI engines are just scraping as much information from the internet they can possibly can find. And unfortunately a lot of that is out of date information. And that's where the errors creep in. um or misunderstanding uh the languages where these fictitious functions for PHP keep appearing in in **[43:57]** the AI codes. But so if you haven't got the knowledge and the understanding to be able to deal with those that is just going to build up in your code, you're going to get lots of lots of errors and then you're going to come to people like us to come and fix it. >> So why not just come to us and get it done in the first place? That's kind of where my message is. Is I mean AI for coding is a tool for developers and people with the knowledge. >> See, I go one step further. I even don't have a problem with non-developers using **[44:28]** AI to develop code. If it's a plain simple static site, sure, go for it. Put it up. Worst that's going to happen is you have to fix a picture that's out of line or something. But when it comes to developing applications, >> build it with AI vibe code away, but don't put it on production until somebody who is a developer goes, "Yeah, that's production ready." And that's a whole field that I've looked at before now of okay, we have **[45:01]** the people that go and fix Vibe code that's already gone out, but what about the guy that's going to review the code in the first place and tell you how much it's going to cost to add the extra bits. And if you've got a semideent vibecoded application, then you're you're in a good starting place and you've probably skipped six months of development there. >> Yeah, I I agree with that. Um I think the review process in AI coding land is **[45:32]** very interesting one that not enough people have explored yet because yeah people will end up with fairly bad vibecoded apps and some people will end up with great ones because you can if you have the relative knowledge you can end up with great ones like you know you don't ha they don't have to be bad just because they're built by AI. >> No no not at all. >> But the review process Yeah. is going to be critical because AI can write code faster than us. So now there is more code and it's of a much more variable **[46:04]** quality than it is like you get less from human developers but the quality is consistently high. Now you've got a lot more from a mix of human alone AI alone and human and AI co-coding with a much wider breadth of quality that still needs to be reviewed. So I find that super interesting. Where will we end up next year? Like what does that look like? >> Yeah, I'd agree. I think um yes, you can you can build complete applications with **[46:35]** with uh AI, but there's got to be that that review. You've got to have someone with quite a bit of knowledge uh to to review to make sure there's no issues. Um and test before you put it out there. Really test it. Um, so as so it's kind of boils down to it's it's a really good tool to get things started to point you in the right direction. Um, I've used it for I I know I need to I roughly know **[47:06]** how to write some function that I need to happen. So, I can kind of explain that to the the uh AI to sort of flesh it out and get it written a lot faster because it yeah, I could probably type a description of the function quicker than I can actually type the function and then AI spits out two seconds later. Um, so it's a good it is a good tool to get going, get some stuff up and running faster. Um, it's then does that then lead to the **[47:38]** expectation of clients now expecting code from us faster? >> Mhm. >> Or cheaper. Yeah. Faster and cheaper. >> Yeah, exactly. It's Is this a a new new concept? I mean, I can understand if we can turn out the code faster and we're not spending as many hours on the code, then yes, they're going to want it cheaper as well, which is is that going to be damaging and difficult for us to **[48:08]** navigate. See, this is a whole other discussion around pricing structures. There's two ways you can charge for something. You can charge for the result or you can charge for the hours spent doing it. Now, if I'm a developer who can type at 200 words a minute and I can write faster than you, does that mean I should be paid less for the same thing as you? >> No, I do. I wouldn't. This is an **[48:39]** interesting and you could you could spend hours talking about this subject now. Um, and this wasn't a subject we had in mind to talk about today. It's kind >> And this is why this podcast's fun. >> Yeah, this this naturally happened. So it is a really interesting concept and there's different ways of approaching it like you say. Um but how do you come to price? Let's say it's a website with uh some underlying uh functionalities and business logic that needs to be in **[49:10]** there. So it can't be just a static website. It is a dynamic website. How do you come to say right for that project this is how much it is? Because the way I would do it would work would kind of work down the road of like just trying to work out how many hours I'm going to put into it to try and work out what the final product would be. But also for if I was to do um a freelance piece of work like that, I might put a because I am I I don't type **[49:42]** as fast as Chris. Chris seems to be able to type at the speed of a light. Um, and and Chris is a much more uh seasoned developer than I am. Um, I I still kind of class myself as fairly sort of junior plus. Is there a junior plus? So, I'm I'm quite I'm quite I in relative to you two, I'm new to this world. I've only been developing full-time uh for five years now. So, I'm a baby compared **[50:12]** to you guys. You've you've been in the world for a lot longer than me. I would expect you guys to charge more for your hour than I would. So you we could end up coming to roughly the same price, but because I've I've lowered my hourly rate because I know it will take me longer to do something. Does that make sense? >> I I get what you're getting at. And you're looking at the pricing per hour. You're calculating it. Then you get to an end value. Let's say that you say your rate is£10 an hour and it's going to take you 100 hours. That's 10 grand. **[50:44]** >> Mhm. I could say £50 an hour and take 20 hours and it's still 10 grand kind of thing. >> Or I could go what is the return to the company on the work that is done. >> So if I go and I do a website for the local corner shop, the return is not going to be in the millions. >> However, if I go to Laravel and I say I'm going I'm I've been contracted to rewrite the website. Well, then that **[51:14]** return is going to be much bigger. The work may be very similar, but the return for the customer is higher. So, I'll always price on return value. >> That's an interesting approach. I hadn't thought of that one. >> So, take taking that one step further, let's say I'm going to design a logo. >> I could have a company called, I don't know, JD, and the logo is going to be J& D. I mean, it's not going to take me long to design this thing probably, **[51:45]** but if it's JD, the little corner shop that sells alcohol, it's probably going to be worth a hundred. If it's JD Sports, then it's going to be worth a heck of a lot more. >> And it's it's the value to the company. Do do you think that McDonald's paid 50 quid for their M? >> No, probably not. I don't know if they I don't know if they just got one of their kids to draw **[52:15]** the M with a crayon to be honest. I don't know how it came about. But >> it it's been through revisions. The McDonald's logo has and they've paid millions each time it's been revised. I'm still trying to find the revision, but that there has apparently been changes to this logo. And for me, it's it's about the value to the person that I'm proposing to. So that that's a whole other discussion on pricing, but that that's the way I do **[52:46]** it. >> Yeah, that's interesting. It makes me feel glad that I am employed full-time to right not I'm not self-employed and having to worry about all those sorts of things. Um >> it's much simpler to be full-time employed. There's quite a few famous books on value versus time pricing and how to pitch yourself to companies and how to, you know, explain it and stop them from going down the per hour path, **[53:16]** >> which probably those books and those trainings are going to be even more relevant now because people will now not necessarily go down the per hour path. Well, it's only going to take you 10 hours and there's rebuttals for that. They'll say, "Well, I could, you know, I'll build it with AI or You know that's going to be an interesting argument which for some projects you can just build it with AI and then you know obviously some projects you shouldn't and it will cost you double now to fix the AI's job and then you know do it **[53:46]** better but um yeah I think just have a full-time role and don't don't get into those headaches. They're a nightmare. >> I mean I don't I don't feel like I am employed. It's it's it's one of those strange thing. We both obviously we as you know we PHP Architect is an Americanbased um company so we're both fully remote and we are the >> the closest >> yeah we yeah it's a little bit difficult **[54:17]** to get over there. Um I mean we've got we've got developers all over the world now within PHP architect. Um, and the team's grown with Chris joining us what, a couple of months ago now. How long have you been with us? Not not long. Um, it's it doesn't it doesn't feel like we're we're employed. We are we're we're just here doing I mean it's probably I don't have to deal with accounts. So I **[54:47]** don't have to um deal with the well I do have to deal with some sort of taxes because that's that's because I have a do a few little things on the side that I I used to do that I kind of keep in the background for um due to my previous life as a a theater technician. Um but so I do have to do a tax return every year. So I can't I haven't got rid of that yet. But um we were talking about it. We did a pre-recording with uh someone else um last week or the week **[55:19]** before talking about being a a freelance and there's so many extra hats you have to wear. So although I don't feel like I'm going to the office to work um like a lot of other people do, I don't have all those other things over my head that I've got to worry about to do with being freelance. Um, and I think some of that, to be honest, is actually down to I absolutely love what I do. I'm I'm one of the very lucky people that has found **[55:49]** a career. Yes, I'm later on in life when I found it. So, all that anyone that's listening or watching or anything, don't be stressed about trying to work out what to do in your 20ies. Don't worry about that. You It will come. It'll happen. Um, I've now found exactly what I really love doing. And I don't feel like I'm working when I'm sitting here just trying to fix things. I really enjoy getting those awkward tickets come through and I'm looking at them going, "Oh, this is going to be a nightmare." But then going, I really then that that burning desire to fix it and go, "Yes, I **[56:21]** can fix it. I've done it. I've fixed it. It's working." That's uh that's that's that's great and I can't thank the people that given me the opportunity enough for for all that. But we're going off track again. This is >> very lucky people. Hey, >> I did want to touch on that part of >> maybe not being late in life, but learning to code with AI as a knowledge partner/pair programmer >> because **[56:51]** when I learned to code and maybe now, I don't know, >> you had to have that burning desire to solve the issue and you had to be willing >> to smash your face against your desk a hundred times in a and then you solve it and it that's the the best feeling ever like oh okay that makes sense yeah of course like of yeah naturally >> but now with AI debugging and troubleshooting often I will just say to AI this is **[57:24]** happening figure it out and tell me tell me why it broke and then I'll go do something else be because I've done that face mashing and I you know I do value it but I've got other things to do as well. So as a new developer coming into this world, AI being a natural thing you you should do, a skill that you must learn. >> Mhm. >> There are going to be some people that use AI in the wrong way >> and delegate their thinking to the AI **[57:56]** and then some people that use AI the correct way of working backwards and forwards with it to understand things better and and progress their learning. and then maybe people in the middle. But I wonder what that ratio is or how we can help people do it the right way and not fall into the like delegating your brain trap. Um yeah, I worry about that future. >> Yeah, that sort of feels like the agentic AI. not not agentic as in the **[58:28]** true way of using agents in AI, but this whole concept of well, I've got an agent that's my secretary and I've got an agent that does me emails and I've got an agent that books me calendar and I've got an agent that washes the dishes and it's that kind of future. At the moment, we don't actually have an AI that is capable of doing those things and we're trying to coax a bigger AI into being lots of little ones. So, you know, you take Claude, you make yourself an agent on top of Claude, and all **[58:58]** you're doing is giving it a prompt, and it opens it up to prompt injection, which then opens it up to being abused. And look at the stories only a few months ago about car companies who were using AIS for customer service on the website and then people getting the AI to agree to sell them a car for a pound. And it >> that shouldn't be possible. And I don't think we're quite at the point yet where we've got enough safeguards to put in place on that. And **[59:29]** people are trying to allow AI to take jobs that cost money. But from a from a business manager's point of view, I can cut the bottom line by $20,000 by not employing that person. Great. But that person needs that $20,000 to be able to eat. M >> yeah they're two very difficult things. So prompt injection currently unsolved **[60:00]** currently unsolvable but I hope they do solve it because it's an incredibly yeah dangerous attack vector. And then yeah what does the future look like if AI takes 20% of jobs and people still need to eat? I I don't think we're going to get a universal income. Um that everybody, you know, just gets a set amount of money or anything like that. So yeah, I don't know that any of us have like an economic sociology background to have like something useful **[60:33]** to say there. I don't. >> But I find it really intriguing. >> Yeah. I mean, I I can't I honestly can't not see AI taking completely overtaking a developer's role in my lifetime and potentially the also the lifetime of maybe some of the the 20-year-olds. I think there's still going to be issues. I mean, there's is it Network Chuck that's got a whole a video about hacking AI. Um he he released a little while **[61:03]** ago. He managed to do some weird stuff with AI. I haven't watched it myself and I can't remember what the caption was. It's something that I I forced AI to I mean obviously he's doing things all likeheartedly if you've ever seen his his videos. Um so yeah there there's there are there are things that people have to really seriously think about and I do it does worry me that some of these corporate types they don't they see dollar signs to be able to save money with AI. **[61:33]** they'll hear people talking about yeah you can build you can get um claude to build a whole whole application for you um and then they'll do it and then there are issues there so it's sort of the the message is just be careful with AI is not it's not perfect there you need to have people around you that know how to use AI but also know how to um troubleshoot problems to code review so they have to have that that sort of um **[62:04]** the programming knowledge. That's the words I'm looking for. Programming knowledge. >> We're sort of bordering on an interesting world at the moment and I want to see where this one goes. We are in the position that we can have AI write the code. We are also in a position that we can have AI review the code. Yes, >> I was I was going to mention that at one point thinking hang on because it's something that that has been discussed recently between a few of us about the some of these packages you can now use **[62:35]** to do a >> almost a pre code review before a human does the code review. And I'm sort of kind of confused about the concept that I could go to Claude and say, "Build me this." Claude can push it to Git for me. Claude can then review it for me on Git before Claude tells the pipelines to put it on the website. I could literally go from prompt to website. Okay. **[63:08]** >> You can go from linear issue >> to website. >> Yeah. >> You can go from user feedback automatically ingested, creating stories, coding, reviewing, pushing and testing in production without human input. Absolutely. I don't know how well it would go, but I know some people are doing that. That sounds actually buy a domain and do it and just publicize it as this thing is entirely AI. **[63:39]** >> All I'm going to do is prevent illegal stuff from happening. >> Do what you want. >> Sounds like a fun experiment. >> Yeah. >> If people can upvote like the most upvoted feature request is added by AI automatically each day. I think that'd be cool. >> It'll be Yeah. Yeah, it' be an interesting experiment to see how good the code is and then we can we can go back a year later and look at the code or something and go, okay, let's not do this ever again. >> But it would also be interesting to look **[64:09]** at how weird the feature requests get. I mean, don't forget we had a public vote that resulted in boaty mcbboat face. >> So, you can get really good things from public votes. >> Yes. >> Yeah. And if we if we then invite the whole of the PHP world to come and vote for stuff, we're going to get some weird stuff. We know some weird people in PHP. >> Yeah. >> I mean, I can see a feature being allow impersonation of other users and then **[64:39]** you can be anybody like Yeah. Very problematic. >> Yeah. Yeah. >> I mean, so long as you never attach it to anything of real value. >> Mhm. And I guess this is where you would have to keep an eye on what's happening, but only touch it if you absolutely have to intervene. You know, somebody goes, "Please make a feature that hacks NASA." Um, let let's not do that. >> Yeah. Could go go a bit wild. Um, or that you **[65:10]** I suppose what you could do is then just have a a a group of people requesting things that have all agreed they're not going to do something stupid like try and hack NASA with it >> and and just come up with some weird and crazy sort of features that are completely legal. Um, and just see what happens and see see what what it turns out. We'll have to revisit. I think that would be >> I mean you can have an AI decide whether a feature is acceptable **[65:42]** also before people can vote on it. So if you say install a Bitcoin miner >> then it will just say no actually no that's okay. Let's not do that. >> Are we bordering on just giving an AI a computer to live in at this point? >> Yeah. >> Do what you want. We'll see what happens. I'm I'm almost tempted to go and buy a cheap Mac Mini and put it in the corner and just put an AI on it and say, "This is yours. Have fun." >> Come up with some experiments now. **[66:12]** >> I'm just hoping they fail because if they don't, we're all in trouble. >> But >> wouldn't it be better to know? >> No, I think >> your ignorance is bliss. I would rather know that it is capable of taking my job so that I can modify what my job looks like because my job will very fast become kill the AI >> or just AI monitor. >> Just sit there and monitor the AI. >> Did you see the joke posts about OpenAI **[66:44]** and their um their button pusher? That is literally what they reported it as. The job was to sit there by the server with your finger on the power button just in case. >> That's a good job if you can get it. >> I think it was like $50,000 as well, so it wasn't exactly a small amount of money. >> Yeah, just do something else whilst you're doing it. >> Yeah, definitely. >> But AI can't do our jobs. It can help us with our jobs. So yeah, I think it will **[67:16]** fail. So like so as an experiment with a hypothesis that probably not worth running, but as a a fun experiment of what would happen, I think that's that's pretty cool. >> I'm going to do it. I'm going to find the server that I can put >> Well, I think we'll have to do it and we'll do a show >> and maybe we'll get Ashley back on and we'll review the code in >> If we're going to do that, we've got we've got to bring Nuno in as well. Oh yeah, review the code. >> Nuno needs to reply to our messages. **[67:46]** >> Yes, he does. >> We've been trying We've been trying to get him for a while. >> I think the review will be This is bad. Don't put it in production. So, I think we can shortcut that. >> So, I'm going to save that clip. Ready? Ready for when we do this and we'll just have Come. This is bad. >> No. Solve. >> Yeah. The review show is just that. >> Yeah. just this is bad. So on that on that lovely note um we are coming to time. So uh actually thank you very much **[68:18]** for spending some time with us and chatting AI and some other stuff that we just went off on. Um but that's what we're about. We're we're just trying to highlight it's cool stuff and talking about what it's like working in PHP and uh and hopefully encourage some new people over to our lovely ecosystem and our uh our do you say family or do you say community? We'll say community. Family feels a bit **[68:49]** too >> like that one. >> Too strong. Yes. Like Discord. I'm getting there. I'm getting there. These are my prompts. He puts these up to show remind me what I've got to talk talk about. So yeah, Discord, head over to uh Discord. Uh we are all hanging out there all week long. Loads of cool conversations going on there. You'll also get a uh warning about 30 minutes before any of our podcasts go live by Chris. He's our official reminder person alarm to remind people. **[69:20]** It's true. He he reminds certain people. He tags them. and he goes, "Right, we're going live." Um, >> ask me for the podcast. It is 1 hour and 37 minutes >> or when he remembers to do the reminder. >> But yeah, head over, come and come and chat. Come and hang out with us. Uh, we talk, we don't just talk PHP stuff. We talk all sorts of things. Uh, there are people from all over the all over the world, all over around the uh the community. Um, also don't forget head over to our website pitchbarch.com **[69:54]** uh to subscribe to the magazine uh where you can keep up to date with all cool things in within the web development and PHP. We've got a code on the screen there alive three alive is all uppercase and no spaces for anyone listening to the audio version of this. I always forget to remember this. People might just be listening not watching. Um, you'll get three three free months of the digital version when you sign up for a year subscription. Uh, there's some **[70:25]** brilliant articles and, uh, uh, information there. Helps you keep up to date and learn new things. Uh, don't forget also to head over to our YouTube, uh, like and subscribe and all that all that goodness. Um, PHP Tech 26 tickets are on sale. Come, come on, come and see us. Um, that that's in Chicago next year in May, towards the end of May. Love to see people over there. Um, and don't forget to head over to our and I'm going to **[70:57]** share the screen if I can do that quickly. I'm not going to do it quickly. That's ridiculous. Um, it's swag store. So, that's store.php.com php.com where you can pick up some uh t-shirts, some hats, hoodies, and all sorts. Uh my favorite t-shirt, which I still need to order, is this one here, the ink blocked elephant, which was actually the cover of uh of one of our magazines when we talked a lot about mental health within **[71:27]** our community. I think that is everything. So, um, just a a reminder that although we were doing every Fortnite, I'm going to say the word Fortnite. I know it confuses some people, but that's every two weeks. >> It's already been said once in this podcast, so it's fine. >> Has it? >> Yeah, Ashley said it. >> Oh, okay. Fair enough. I I must have missed that one. Sorry. You must have said I I Yeah. Okay. uh every Fortnite, but we are actually now releasing every **[71:58]** week because we've got so many of these pre-recorded. We'll have a pre-recorded one week and then we'll have a live one the following week. Fingers crossed as long as we can get people to join us live. But, uh, apart from that, thank you again, Ashley, for taking the time out of your day to come and talk to us. Um, >> thank you for having me. It was fun to chat about AI and roughly adjacent things. Yeah, we should we should get you back on in sort of a maybe six to 12 months time. We'll see how Boost is **[72:29]** getting on and see if how our experiment goes on. If we do do >> I'm excited about the experiment. We'll see. >> That'll be interesting. Well, anyway, thank you very much uh for your time and thank you very much for everyone watching and listening. Uh we will see