I Was Firmly Against AI Code… Until I Wasn’t.

I’m a very late adapter in using AI to assist with writing code, even as little as just one week ago I was writing on X/Twitter that I’m still “Firmly in the not using AI to write code camp”. I’ve been writing code with PHP now for over 20 years, and 2026 is my eighth year of working with Laravel - Read more about my journey in my ‘who, why and what blog’.

To me, writing code has always been initially a hobby, long before I got my first job as a developer, when I was still in school and college, I was writing code, even now, outside of work, I work on personal projects, so to me, writing code is, and always has been, fun!

At home after my two year old daughter has gone to bed, I have a couple of hours of me time working on personal projects, it helps me relax, helps me unwind, so from my point of view, when I see people posting online, and not just anyone, even big names in the Laravel world, talking about how they use AI for writing the vast majority of their code, I sit here and wonder why, why would I want AI to do something I love doing?

When ChatGPT took the world by storm a few years ago, I was a very late adapter, it was only 18 months or so ago I actually started using it, not to write code, but as what I describe as a sounding board, somewhere to discuss ideas, work out the best way to approach a problem, or help me look at it from another direction, but to never write code, or even directly suggest any code examples.

At times the advice it gave it was absolute garbage, it sent me down rabbit holes trying to fix an issue that wasted a tonne of time, and I ended up just giving up on ChatGPTs advice and finding the fix myself with a completely different approach, other times it has helped me spot something obvious that I just couldn’t see, so it is very hit and miss, but that’s like any tool.

But in the latter half of 2025, there seemed to be a big shift, primarily in the Laravel world, with using AI to help write Laravel Apps, Laravel hired Ashley Hindle, as the first employee in their new AI team, and Laravel OG Nuno Maduro quickly followed, moving from the Cloud team to the AI team, so this meant that Laravel the company was embracing AI and building an internal team around it, and not long later, they released Laravel Boost, a package that aims to “accelerate AI-assisted development by providing the essential context and structure that AI needs to generate high-quality, Laravel-specific code”, and Taylor Otwell himself has spent the past few months working on some top secret AI project that will be unveiled later this year.

So where did this big turn in Laravel and AI leave me, and others like me, that hadn’t embraced using AI to write code? I remember seeing the launch of Boost and thinking “Well that's probably not a package I’ll ever need”, but I still watched various videos on the package, from both Laravel’s YouTube channel, and ones on Laracasts.

I remember I kept hearing terms, that honestly, meant nothing to me, things like “AI Agent”, “MCP Server” and several others, even as recently as just before Christmas, when Laravel released videos on Laravel MCP and Laravel Boost as part of their Laravel Advent series, I watched them both together one night, and felt lost.

I remember thinking at the time that I feel exactly the same as I did way back in 2018 when I first heard about Laravel, “what's a controller?”, “What’s a model?”, “What's a service provider?”, has now become “whats an AI agent?”, “whats an MCP server?”, and in hindsight, I do wonder if that was the turning point for me.

When Laravel released those videos, I was in the first week of two weeks away from work, a combination of Christmas / New Year shutdown and annual leave, and it turned out this was a period where a lot of people were spending their free time experimenting with AI, and I think in the first week of January when everyone was returning to work, it seemed like everyone was talking about what they’d been doing with AI over the break, both on X/Twitter, and even here internally at work, all of a sudden, there had seemed to be another big shift in this new world, but here was me, still in that ‘No AI code camp’

I mentioned at the start of this blog, that tweet I made, I had only written it that first week back after Christmas, but finally, at this point, my mind was starting to open to the idea, and starting to think about ways I could use AI, and how it could fit into my workflow.

Also over Christmas and New Year, I saw Jack Ellis from Fathom Analytics tweeting about his experience with AI, and he said while it has helped him and his business a lot with a huge data migration, he sees the code generated by AI as more something that a junior developer at a business could write, and he reviews it as such, rather than trust it blindly, and that was a big thing for me to see, as I said at the start, I’ve seen lots of posts from big names in the Laravel Community pumping out project after project with AI, and that didn't sit right with me - but this concept, using it as a starting point, reviewing it like a junior developer, doing tedious tasks, that did fit my mindset and fit my work flow.

I knew from some teases online that Jeffrey Way over at Laracasts had a “getting started with AI and Laravel” course coming out at some point in January, and I remember thinking, “if nearly 8 years ago Jeffrey Way helped me get started in Laravel, he turned all those words and concepts I didn't know at the time into the thing that is now my bread and butter… if anyone can be convince me, and get me started on AI, its him…”

That was was it, this was when I had realised, my mind was starting to turn, I was starting to open up to the idea of using AI, I actually had a feature in a personal project on my todo list that I’d been putting off for a while, and I was now starting to think “I wonder if AI can help me get started with this… get the really tedious bits out of the way initially leaving me free to do the fun bits…”

I’d had a few conversations at work on the same subject, and one of my colleagues suggested “Claude Code in Action”, an introductory course to Claude Code while I was waiting for Jeffrey’s series to drop, and last week, I finally had chance to watch it, I followed along with the course, set up Claude Code on my work machine, followed along with the included example project, and honestly, some of the things in the course, and example project, did really impress me, I remember sitting there at one point in the course thinking “Did that just happen? Did AI really just make that?” and I finished that course thinking, “Ok, now I’m interested, but now I need the Laravel context, I need to work out how I can make this flow work, with what I do every day, how I can integrate this tool fully in Laravel” - and not just with Boost, but with the way I look at projects, the code styles and guidelines we follow.

At the time, I was on a client project, that project already had Laravel Boost installed, and I had a quiet afternoon, I’d finished the feature I was working on, so I took the opportunity to create a new branch, and do some playing with Boost in a real project, and I was pleasantly surprised with the results, as part of the feature I did before, I wrote a series of tests for it, this was a project we’d inherited so it was missing a comprehensive test suite, so I asked Claude, using the tests I’d written for this route and controller as an example, write tests for this other controller, covering authentication, authorisation, validation etc, just some basic smoke tests,

The results came back within a couple of minutes, and they were surprisingly good, so I continued with asking it to write more tests, and I carried on with this feature early this morning too with more tests, and they continued to impress me, the only issue I had was at one point it was trying to make changes to an old database migration because one of the tests it had created was failing, and it decided that was the best course of action to fix it, so I had to interrupt and tell it to never change existing migrations, if the data is incorrect fix it in the test or in the factory, not in the database schema, and after that it was fine.

After the success I’d had at work last week, I set up Claude code on my personal machine with the intention of spending a couple of hours experimenting on one of my own projects, setting up Laravel Boost in that project, and writing some custom guidelines for Boost to use, and over the weekend, it did help me get started on that big feature I’d been putting off, so far I’ve only been asking it to do little things, help with some migrations and refactoring a series of models, factories, and tests to match this new migration and this brand new data structure, I’ve not asked it to do anything ‘big’ in this feature yet, I still default to writing the ‘fun’ code without even thinking about it, but it's a start, especially from someone who resisted AI, and it has helped, and it’s got the tedious starting point I’d been putting off, done and out of way.

So in just a week or so, I’ve gone from being totally against using AI to write code, to starting to have my mind opened, learning about it more and actually seeing with my own eyes what I’ve been missing.

I’ve been able to think about and see where it can help me, actually using AI, and even wanting to experiment with it more, and thinking about how it can help with both client projects at work, and any personal projects at home

Do I trust it completely? No, not at all, when I was using Claude to help write tests on that client project last week, I was reviewing the git diff after it had finished, and in a lot of cases I went in and changed a little bits, more to fit my (And Jump24’s) coding standards, adding some line breaks between different blocks of code, making sure it uses Model::query() and not just using Model::where(), I even changed the test names once or twice to something more suited - but they’re nothing major, and all things that would more than likely get solved by AI by using custom guidelines alongside Laravel Boost.

If I’ve made a total 180 degree shift in such a short space of time, where will I be in another week or so? Or even a month? AI is constantly changing and getting more powerful. I saw that change happen over the Christmas and New Year break, and that was my turning point, when the bell dropped, and my mindset shifted.

If I hadn’t made that shift, where would that have left me in a few months time? Who knows..? But if you’re like me, I’ve been writing PHP since I was 14 years old, I’ve seen it come a long way, I’ve been writing Laravel for nearly 8 years and discovering that framework, and embracing it completely changed my career, and even how I write code… is me embracing AI now in 2026 going to have the same result as me embracing Laravel did in 2018? I don’t know if it will be quite as drastic, but honestly I think it can only help improve things… So if you’re like I was a couple of weeks ago, still in two minds with AI, or even still against AI code… take it from someone who wore those shoes recently… give it a try, watch a course, experiment, either in a project you’re an expert in, or just in a laravel playground, you might be surprised with the results.

As recently as yesterday Claude Code helped me fix an obscure bug that had just occurred in a personal project. I got the bug report in through Laravel Nightwatch, and at first glance I couldn’t see the problem, it was a simple action class, that had a test class backing it up, I told Claude the file, the test, the error and the line number, and less than two minutes later, it had it fixed.

Within two minutes of me sending the instructions to Claude Code it had identified the issue, and explained to me what it was, fixed it, and even wrote an additional test in the test class that match the structure of the other tests in that file and prove the test worked. If that bug popped up last week, it would have taken me a lot longer than two minutes to work out what the actual obscure issue was, and fix it. I obviously checked the code it wrote before I blindly trusted it, I reverted the changes in the action so I could run the test and see it fail, which it did, reapplied the changes, ran the test, and all green.

Less than five minutes after I got the email from Laravel Nightwatch about the issue, a fix was getting deployed to production, and that wouldn’t have been possible without AI assistance.

Two AI courses have dropped on Laracasts over the past few days around using AI with Laravel, he’s made both courses free to watch for everyone, and both are on my must watch list, Taylor’s big AI project is due to be released as a public beta after Laracon India, which is just a couple of weeks away… so I suspect more and more big things are coming surrounding Laravel and AI.

Laracon EU is just a few weeks away too, and I wouldn’t be surprised if there’s a few talks there around AI, and that Taylor’s Laravel keynote at Laracon EU could also focus quite heavily on AI, I’ll be attending this year, it will be my first Laracon, and a few other members of the Jump24 team are there too, we’ll have a stand in the venue, so come and say hi if you’re attending and see what we can do for you!