I Tried Taylor Otwell's Dead Simple Dev Setup.. Now I Get It

nunomaduro| 00:13:23|May 30, 2026
Chapters12
Discusses preferred editors, AI models used for tasks, and when to switch between models for speed versus complexity.

Taylor Otwell’s team showcases a pragmatic, beginner-friendly dev setup using familiar tools, AI helpers, and Laravel Cloud features to ship quickly without sacrificing quality.

Summary

Nunomaduro chats with Taylor Otwell about his everyday dev workflow and the tools he swears by. Sublime Text remains a core editor, with Open Code and Ghosty Terminal forming a quick, lightweight trio for fast reviews and iteration. When it comes to AI, GPT 5.5 is the workhorse, with Kimmy 2.6 used for snappy, simple tasks. The conversation covers planning versus building, and how Taylor sometimes skips straight to build for straightforward features while using plan mode for the trickier work. For code reviews, Sublime Text handles the diff, but GitHub Desktop shines for diffs—an unexpectedly perfect pairing that even Taylor calls underrated. They also discuss the soon-to-be-released API starter kit, the role of sub agents to tame context windows, and the best practices guides (shaped by Pushpack) to keep Laravel code lean and idiomatic. The chat then shifts to Laravel Cloud, with autoscaling queues, price innovations like a $5/month plan, spend caps, and the promise of sandboxes for agents. Throughout, Taylor highlights the balance between handcrafted core decisions and AI-generated productivity, plus the ongoing influence of the broader PHP community on Laravel’s roadmap. Finally, the talk rounds out with practical thoughts on jobs, PRs, and a behind-the-scenes look at how leadership steers product direction across open source, Forge, and cloud services.

Key Takeaways

  • GPT 5.5 is the primary AI model used for complex tasks, with Kimmy 2.6 for faster, simple jobs.
  • The diff tool preference is Sublime Text for viewing changes, paired with GitHub Desktop for high-quality diffs.
  • Sub agents help manage context windows by organizing tools and reducing unnecessary context in conversations.
  • Best practices guides steer LLM-generated Laravel code toward idiomatic patterns like higher-order collection methods.
  • Laravel Cloud’s autoscaling queues and spend caps are designed to keep side projects affordable and predictable.
  • Even for AI-integrated workflows, Taylor values handcrafted core API design and human review of implementation.

Who Is This For?

Essential viewing for Laravel developers exploring AI-assisted workflows, cloud-based queues, and pragmatic tooling to level up how they build, review, and ship features.

Notable Quotes

"Sublime Text, Open Code, Ghosty Terminal. That's pretty much it."
Listing the core tools in the starter setup.
"Mainly GPT 5.5."
Identifying the primary AI model used for tasks.
"I would still end craft some of it, maybe a little bit less than before."
On how he’d approach building the AI SDK today.
"We tried to basically do like could we build something better than horizon in cloud because Horizon's such a popular package."
Discussing Laravel Horizon versus Cloud capabilities.
"Sub agents help with the context window a little bit as well as keeping like the conversation context shorter."
Explaining the value of sub agents.

Questions This Video Answers

  • How does Taylor Otwell use AI models like GPT-5.5 and Kimmy 2.6 in Laravel development?
  • What features are coming to Laravel Cloud, and how do spend caps work?
  • What is the role of sub agents in AI workflows for Laravel projects?
  • When should you use actions vs. form requests in Laravel Cloud projects?
  • What are the best practices guides in Laravel Boost and how do they influence code quality?
Taylor OtwellLaravel CloudAI in developmentGitHub DesktopSublime TextGPT-5.5Sub agentsBest practices guidesHorizon vs. cloud queuesForm requests vs. actions
Full Transcript
Which editor are you using right now? I mean, you probably don't have to ask, you know, Sublime Text. Yeah. Yeah. So, Sublime Text all ends. Yeah. Sublime Text, which is actually really nice in like the AI era to just like quickly review things very fast. So, Sublime Text, Open Code, Ghosty Terminal. That's pretty much it. Open code, Ghostly Terminal, Sublime Text. So, you mentioned Open Code. Which AI models are you using? Mainly GPT 5.5. Uh, GPT 5.5. Okay. In that dump, sometimes uh Kimmy 2.6, six, but mainly GPT 5.5. When would you choose one or the other? Uh, sometimes the Kimmy model is like really fast. If I if it's a pretty simple task and I just want to do it really quickly, I'll just do that. If it's a more complicated task, I'll do like GPT 5.5. So something like rename all these files to something really simple. And then if you want to plan or something that would be sh Yeah, exactly. Okay. Okay. You are tailor developing a feature on cloud. What is your process there? Do you plan first? Do you go straight to build mode? Would you start slowly like how would you approach it? It's a good question. I do use plan mode sometimes, maybe like 50% of the time I'll use the uh plan mode and then I'll go to build. I used to use plan mode all the time, but now like more and more I'm skipping just straight to build sometimes, especially for pretty straightforward features, but for more complicated stuff, I'll do plan mode, send a couple messages. Okay, go do that. Like view to go. Just go. Yeah. And I assume at the end you review the produced code. What tool do you use for that? Sublime Text. Oh, you mean for the diff? Yeah. Yeah. Uh, for the diff, I use GitHub Desktop, actually. Oh my god. Me, too. It's so good. Such an underrated tool on It's actually the only thing I use GitHub Desktop for. It's just like diffs because it's very beautiful there. You know, YouTube desktop is the most underrated tool in the planet right now. I do think it's pretty good. Yeah. Yeah. Yeah. That's exactly what I use. When the API starter kit will ship. Oh, that's a good question. I think it's in final review right now. We need to ask Joe Bump. Yeah. Joe Tanab is leading that stuff. Yeah. Yeah. I actually don't know the exact date, but I know it's like I know it started. Yeah. Yeah. Yeah. I think it's like weeks away, not like months away. Oh, push back is luggly on the chat. It just said very soon, which is good. One question about your general perception of your developers at Laravel, but also from the community. Uh how do you think the adoption is being in terms of using AI? Do you feel like every developer at Laravel is using AI at the minute? At Laravel, the company, I think almost everyone's using AI at the company right now. Then even in like the broader Laravel ecosystem I think it's pretty popular because of tools like boost things like that I mean are pretty popular tools. So I think AI adoption is pretty high. Do you feel any pain on open source? We get a lot Yeah. We get a lot more spam PRs. But what I've noticed is that like someone will create an issue, right? And it's almost like people have bots set up to basically watch issues and automatically send PRs. Oh my god. So what will happen is if a spam PR comes in, I close it. then another user will just open basically the same PR for the same issue you know like so there is a lot of spam PRs where they're like very easy to spot do you feel that is more contributions right now than before yes there's more pull requests in general even I mean sometimes just because a pull request was created with AI doesn't mean it's like invalid like sometimes they do get merged because they're like correct even though they came from AI even though they came from AI they can still be like correct I want to talk to you about one of the codes you have in 200 I think it was N 2025 you mentioned something which is I wrote this AI SDK, a lot of it at least was by hand. Yes. Would you still say that right now if you were to build the AI SDK like this week? I think I would build it differently. So like I started working on the AI SDK I think in like September of 2025, right? And then in December of 2025, like everything went crazy with like all the agentic development stuff, changed everything. Yeah. And you know, so most of it was written before that. Now I think I would do it a little bit differently probably like I would define all of the interfaces myself and then use the AI to write a lot of the implementation and then I would review it and like make it just to clarify you said some of the interfaces you would you still be using AI at least for the planning of those interfaces shaping it a little bit. I think I would plan it myself still. All right. So you still I think I would still basically hand plan the core API and interfaces and then all of the like boring code, let the AI generate. I think I would still like adjust it like even after the AI writes the code, I still move the methods around or like change things a little bit, but um it just saves a lot of typing. To recap a little bit, your answer would be you would still end craft some of it, maybe a little bit less than before. Like the very foundation would still be tailored auto handmade stuff. Talking a little bit about the SDK, you know, very popular stuff. You mentioned in your talk sub agents. Do you have like actual use cases for using sub agents? I think you mentioned some of them, but uh yeah, they can be useful both as like an organizational tool just so like you can sort of organize your tools and sub agents a little bit better. They can also help you on the context window a little bit because if an agent has like a whole lot of tools, it can really bloat the context window, especially if a lot of them aren't needed. So sub aents help with that a little bit as well as keeping like the conversation context shorter. Sub agents was actually a community contribution Nice. Uh which was cool. Just merged in a few weeks ago and then pushpack you know who's in the chat has been doing a ton of work on the AI SDK right basically like basically leading the development of that on a day-to-day basis. So one of the best examples I like to give for sub agents is the investigation story where you have a main agent and then suddenly the main agent needs to verify if is Taylor already published this specific tweet. So going to the internet and like scouting and filtering all the HTML returned from the response is actually something that can pullate a lot the context. Yeah that would that would be the use case for sub agent where let's puate the context of the sub agent and then just give me back the answer so everything is displayed. Moving a little bit to one of the latest editions uh on Lavel which is the best practices guides that were shipped with Laval Boost. How did you build those? It was you crafted them. Pushpack led the initial development of it and then I reviewed them and like some of the stuff it does is just like I think help the LLM write Laravel more like how we would write Laravel. Like for example, if it uses like a collection and it wants to go through each thing in the collection and call a method, right? the best practices skill tells it to just use like higher order method like you know collection each boom and you know whereas like before the LLM might not have used that for example right um so it just helps steer it to like more like wellcrafted code for Laravel right um and it's something we can like continue to improve and evolve like over time um because even like best practice Laravel opinions sort of like evolve over time right like a lot of people now use like the action stuff but that was like not really so common four or five years ago Um, so we it helps keep the LLM like up to date on like what's the best way to write Laravel and organize it today. Exactly. And your take on the action pattern from what I can remember at Laru was that you would you like it, you would use it. However, you would still rely on controller code for simple stuff. I mainly use actions for like more a little bit more complicated things or things I need to reuse like you know if I need to reuse the action in an API or in a controller, actions are definitely useful. Um, if it's a very simple controller method that I don't need to reuse anywhere else, like I I might not necessarily create an action class. How about things like form requests for example? Would you let's assume you we have a very simple validation in a very complex one. Would you still always rely on form requests? Would you still do both? I do a request validate a lot like request arrow validate in controller. Yeah, in the controller for simple stuff. I still I actually really like for requests for like more complicated things like I know we like when you're building a product like cloud you kind of need like requests and actions and things like that uh but for like really simple apps I think request validate is really convenient. Yeah. Yeah. On the for cloud there is form requests that are literally this size just because there is so much logic to verify before things can move forward. I want to ask you like next week you back to the office. Let's let's say that what keeps you busy during the I still start my day on GitHub like maintaining pull requests. So like even taking the train out here to the venue this morning I was on my laptop working Porter Quest but then after that it's a lot of um what is actually different about Laravel now is there's a lot of like context switching. One moment I'm talking to someone about cloud then I'm talking about forge then I'm talking about open source right and so like there's a lot of switching throughout the day or maybe one time I'm talking to someone about hiring or something like so every day is actually really different depending on like what's going on. Like lately you know we've been working on a lot of the manage Q stuff in cloud. been talking with Kieran and the infrastructure team and the app team about like getting that feature done. Um so every every week and every day is a little bit different but still the common thread of maintaining pull requests every day right I know for a fact that back in the days used to be a lot of inspired by DHH. Yeah. Do you have like fans right now people you really look up to and you know I I mean speaking about myself for example I when I I still do but like back in the days I was a huge L fan boy. I still am, but I used to look a lot to people like you, people like Jeffrey Wade, Adam Web. Do you still have those kind of like people you really like, okay, I'm going to watch this Twitter all the time? I still keep up with like a lot of developers on Twitter. Um, because I think there's a lot of people building cool stuff in PHP, but also just like other languages. They're building cool stuff and I still take like inspiration from them as like, oh, this is a cool idea or right, you know, this would be really cool in Laravel. M um so I basically just am always watching like you know Twitter or just like online where people are hanging out and building stuff and just taking ideas. We have a lot of new stuff coming for Laval Cloud. Do you want to mention one or two things that you think are really cool and people should literally know from day one? Yeah, I mean there's a bunch of big stuff coming which I tweeted um last week or week before. So like out I guess yesterday US time uh today for us was uh manage cues. Uh that was a really big feature we built for Laravel cloud where basically you have an autoscaling queue scales to zero. We basically tried to take like what is the dream Q system we've ever wanted to build for Laravel and build it in also has like all the graphs and everything built in basically artisan. Yeah. We tried to basically do like could we build something better than horizon in cloud because Horizon's such a popular package. Then um I tweeted we're also coming out with like new pricing. So, we're going to have a $5 a month plan on Laravel Cloud, which combined with our new scale to zero stuff, which wakes up in like milliseconds, you know, means you can run like lots of side projects on cloud and still stay within your $5 budget. We're also shipping spend caps. So, like one of the big I guess like biggest pieces of feedback on cloud is people are like, I'm scared I'm going to get like a bill that's like way too high. And so, you can define like I only want to spend $10, you know, like no more. And so we're shipping that on cloud in I think two weeks maybe next week. What happens if I have my event website or my website the cap you know what happens at so you can choose you can either one will notify you at like 50% 80% 90% that helps and then once you hit it you can choose either to just like turn everything off so like you don't spend any more money or you can choose to keep going and just get like the notifications that you're like approaching your spend limit. So it's up to you for like side projects you don't care about. Maybe you're just like, "Okay, turn them off." Like, uh, but if it's something in production, of course, like you probably don't want to turn it off, so you could keep it running. One question from the chat real quick. Someone is asking if there is plans on Laravel Cloud to have sandboxes for agents. We actually have this on our road map. I'm not sure when it will ship. We even like experimented with building it. You know, sandboxes being this like really popular thing right now for running like untrusted code or agent generated code. So, I don't know. I I I think it's possible we will have that. We have other like AI related plans we would like to do as well. Uh probably in the second half of the year you know so Lacon US is coming is in Boston is coming. There's anything you can share or tease to people of stuff you or the cloud team or you know anyone at Laval will be announcing and people cannot miss. Oh gosh, I'm sure we'll be discovering a lot of cool stuff on cloud, open source, everything we've been working on, right? Um, you know, I usually use Laracon these days as like a recap and a preview like recap. Okay, here's a bunch of cool stuff that we have missed last couple months and also here's some new stuff that like no one has ever seen before, right? So, it's it'll usually be a mix of that those things, but um and it could be in cloud, Forge, open source, you know, across the entire like company basically. Yeah. Plus then much more resources now we have at Latal. So there will be news about literally pretty much every single piece of product we have at Laval which is good. There is also past stuff coming. Uh I've been working on past five mean some of my time at Laval which is awesome and there is past five coming as well. Um and you you don't want to miss it by the way. It's a cool stuff. That's the great tailor. Yes it is. What is your workout routine? Yeah that's a good question. What is your workout routine? That's funny. So, I've been going to the gym pretty consistently for like the last maybe six or seven months. Okay. And I've been doing uh bench press, pull-ups, squats, dips, push-ups, planks. Yeah, that stuff mainly. Also, play a little basketball, but not too much, right? But keep it pretty simple, but just try to be like consistent. And try not to eat too much. But here in Japan, that's actually pretty hard. Impossible, if I won't say. Appreciate you coming, man. Thank you so much for having me. It's awesome. I'm going to unmike you. So, this interview will be subject to post-prouction and be published potentially as a YouTube video. And yeah, thanks for coming, man. See you at lunch already? No, I'm about to. All right, have fun. Thank you. That was awesome,

Get daily recaps from
nunomaduro

AI-powered summaries delivered to your inbox. Save hours every week while staying fully informed.