Running Clock - Laravel's AI Initiative with Ashley Hindle
Chapters11
Introduces Ashley Hindel and frames the discussion around common misunderstandings of AI in developers workflow.
Ashley Hindle argues AI is a learnable skill for developers, not a magic button, and shares how Laravel Boost shapes practical AI workflows.
Summary
In this quick Deep Dive, Laracasts hosts Ashley Hindle, creator of Laravel Boost, to unpack the reality of bringing AI into daily development. Hindle cautions against the myth that AI is skill-free and emphasizes practice, much like mastering Git, new frameworks, or TDD. He acknowledges AI will replace some code reviews over time, though human intuition will always matter for critical judgments. Hindle prefers manual setup for new projects, citing unreliable AI defaults for packages, CI, and test suites. He shares practical tips from using Boost, such as leveraging the HTTP client effectively and discovering a wealth of hidden methods in collection and array helpers. The discussion dives into how AI speeds decision-making by quickly laying out options, with a caveat that Laravel-specific testing patterns remain challenging for AI. Hindle characters the idea of “vibe coding” as generally dangerous, recommends Claude Sonic 4 for most tasks and GPT-5 for frontend work, and hints at forthcoming enhancements for Laravel Boost like richer guidelines and broader ecosystem tooling. The episode closes with goals for more powerful MCP tools and improved documentation search, while keeping the playful banter intact.
Key Takeaways
- AI is a skill developers must practice, not a magic box that spits perfect results.
- Code reviews will be partially automated in the future (e.g., OpenAI's codeex does first-pass reviews), but human taste remains crucial for important decisions.
- Manually setting up new projects (packages, CI, tests) yields more reliable foundations than letting AI do the initial setup.
- Using Laravel Boost highlights effective use of the HTTP client and reveals many lesser-known collection/array helper methods you didn’t know existed.
- AI can reduce decision paralysis by presenting multiple options quickly, allowing you to pick and proceed without exhaustive deliberation.
- Laravel-specific testing patterns (feature vs unit, database refresh, QFake, Carbon freeze) still trip up AI today, indicating a skill gap rather than a universal flaw.
- Vibe coding (out-of-band code generation) is generally risky; Hindle recommends avoiding it except for tiny, non-critical scripts.
Who Is This For?
Developers curious about integrating AI into Laravel workflows, especially those considering Laravel Boost. It’s especially useful for those who want realistic expectations about AI’s capabilities, limits, and best practices in testing, project setup, and tool choice.
Notable Quotes
"The biggest misconception is thinking that AI is skillless and you just type in the box and get what you want."
—Hindle debunks the idea that AI is a magic button and reinforces the need to practice like any other development skill.
"Code reviews are sacred and critically important, but AI will still replace many code reviews at some point."
—A balanced view on automation’s future impact on reviewing code.
"I like to manually make sure the foundation is what I want and allow AI to work from there."
—Prefers reliable manual project setup over trusting AI for initial scaffolding.
"Claude Sonic 4 right now for the vast majority of things and GPT5 when I'm working on the front end."
—Shows tool preferences across tasks to optimize AI performance.
"Vibe coding, not looking at the code, just pushing it, 99.5% of the time it's wrong."
—Cautions against using quick-and-dirty coding practices.
Questions This Video Answers
- How realistic is it to rely on OpenAI codeex for first-pass PR reviews in real projects?
- What are best practices for initializing a new Laravel project without AI, and when should AI assist?
- Which Laravel Boost features most improve AI-assisted workflows?
- Can Claude Sonic 4 and GPT-5 genuinely improve frontend and backend tasks in Laravel projects?
Laravel BoostAI in developmentOpenAI codeexLaravel testing patternsHTTP client (Laravel)Vibe codingClaude Sonic 4GPT-5Frontend toolingMCP tools
Full Transcript
Today I've got Ashley Hindel, the creator of Laravel Boost. This should be a lot of fun. Ashley, the clock starts now. [music] [cheering] What's the biggest misconception developers have when they first try AI in their workflow? Probably the biggest misconception is thinking that AI is skillless and you just type in the box and get what you want. But but that's not true. This is a skill that you develop that we need to embrace and practice and learn just like we did Git and new frameworks and TDD. I I think that's the correct answer. That makes sense.
Do you think AI will ever replace code reviews or is this something sacred? Yes and yes, but not all code reviews. So code reviews are sacred and they're critically important, but AI will still replace many code reviews at some point. And OpenAI has just announced that their codeex tool is reviewing the vast majority of their PRs as a first pass. And so human intuition and taste will always be better here than AI, but often AI is good enough. So I think it will, but code reviews are still critically important and sacred and we shouldn't let it take over so easily.
Uh, what's a task that you still prefer doing without AI? Probably new project setup. So I was starting lots of different projects and I don't trust AI to set up packages and composer and CI or test suites or libraries. [music] So I like to manually make sure it's the foundation is what I want and allow AI to work from there. I have only bad experiences letting AI do that initial setup. Good tip. Uh what's one thing that you have learned about Laravel by using boost? Yeah, probably the biggest thing recently is how to use the HTTP client effectively.
And then randomly I'll learn about a new method in the collection or string or array helpers class that I didn't know was there cuz there's like a hundred of them that do things you don't know that you need. Um, so that's always cool when it's like did you know this exists and I didn't. And so yeah, probably those This thrilling episode of Running Clock is proudly brought to you by Lariccast. Yes, sir. [music] Yes, ma'am. It's the Rutinest Tutinis Laravel screencast this side of the information highway. And back to the show. All right, we're halfway through the time.
Uh, what is an underrated way AI makes you faster? For me personally, probably decision making. I sometimes I get into decision paralysis of like there are so many ways to do things. What the hell do I do? And so it's nice for AI to present all the options to me and then I can just say okay let's do option three and move on pretty quickly. And so and because I know AI can do refactoring if it turns out I don't like that and I don't have to go through the pain of refactoring is much easier to just decide and move forward.
So probably that nice. Uh what is one Laravel specific task AI still struggles with? In my experience, testing or Laravel specific testing patterns like feature versus unit, uh, refresh database, Q fake, carbon freeze, it just seems to ignore all of the nicities we have when I try to build tests. Um, but maybe that's a skill problem. Maybe I just need to be better. Um, what is your personal stance on Vibe coding? Vibe, don't do it. [laughter] Um, proper vibe coding, not looking at the code, just pushing it. 99.5% of the time it's wrong. For some of my side hacky projects, I will do it if it's never going to go anywhere and I'm only going to use it two or three [music] times.
Um, but yeah, don't do it. Don't do it. We have 45 seconds left. What is your preferred large language model if any? Claude Sonic 4 right now for the vast majority of things and GPT5 when I'm working on the front end. Oo, nice tip. What does the future of Laravel boost look like? That's a tough one. More guidelines, broader ecosystem support. um more powerful MCP tools, better documentation search. There's more that I want to do, but I can't share those things just yet. And that's all the time we got. Uh thank you so much. You have done amazing.
Thank you. Even if I'm talking, you should just stop it at 5 minutes. Like if I'm in the middle of an answer, it'd be great. Just black screen. Yeah.
More from Laracasts
Get daily recaps from
Laracasts
AI-powered summaries delivered to your inbox. Save hours every week while staying fully informed.









