My AI Guidelines for Laravel/Filament: March 2026 Update

Hello guys, some people ask me do I add any custom guidelines for Laravel projects on top of default Laravel boost guidelines and skills and I used to do that all the time. I have my own AI guidelines which is on Laravel daily here. This is the set of guidelines for PHP and Laravel and also I have a separate filament set of guidelines on filament examples. One of the tutorial is this. But in 2026 something's changed. So I was using that guidelines in 2025 by copying them on top of Laravel boost. So in the documentation of boost you can see adding custom guidelines. So I just copy markdown files into/g guidelines and then they are generated in addition to cloud MD or agents MD in the same file. And to be honest it's still working but two things have changed. First the models themselves got much better. So not sure if you noticed kind of the revolution over the Christmas break when a lot more people tried OPUS in action. Also OPUS 4.6 and GPT 5.3 codecs and even Sonnet and even GLM models got much better without any external guidelines on top. And then second thing I kind of looked again more closely at the guidelines. I zoomed that in and I stopped caring that much about some of the stuff here. So for example, generate enums always in the folder app enums. Yes, this is my personal preference, but does it really matter that much? I mean, I do still care about code quality, but many of those things are just personal preference and the guidelines which don't really affect the quality of production code. They affect the maintainability for the future maybe. But I decided to revisit those guidelines and retest them with the newest models with the thought that maybe those models internally got much better at doing most of the things that I prompt for themselves without my help. So maybe I don't even need those extra guidelines. And here's what I tried. I vip coded two projects specifically vibe coded intentionally without even looking at the code. So one of the project is a small filament project with three rows for bakery like recipe and production. So nothing really fancy and I did not use my guidelines and this was the prompt the plan for the project. So user roles recipe management daily production plan. So quite a small project all done in one prompt in six minutes by cloud code us but the code and I decided to ask cloud code to evaluate that. So this was the prompt. The project was produced by AI agent. Compare the code against two guideline files and tell me which guidelines were missed potentially. And this is the result. This is the table with seven points and a few of them critical. So for example, things I told you like I don't care that much about is this example unnecessary temporary URL variable. I mean, yeah, it could be optimized in the code and this is the result. But in a way, it's a false positive here and the URL variable is easier to read outside than to have it in line inside with all the match structure. So things like that from my guidelines have low impact or sometimes even our false positives. But some of the things are critical. So for example, without the guidelines, cloud code did not generate smoke test for filament. This is in my filament guidelines. The first rule you must in caps generate filament smoke tests and partially it can be insured by claude MD and laral boost guidelines because tests are mentioned a lot of times in the claude MD but there's no enforcement. So there's a skill for pass testing but in my case I was enforcing the test as part of my custom guidelines on top. Another high impact thing is custom blade with tailwind but no custom theme. So this page this is a custom page with some CSS on top with Tailwind and for that to work in filament in latest versions you need to have custom theme and this was again part of my filament guidelines. If you create custom blade with Tailwind you need to create a custom theme for filament specifically. I actually ended up creating a separate package fillet check which runs after AI generated code in my experiments usually detecting the deprecated code and one of the rules is actually in the pro version specifically if custom theme is needed which looks something like this and flag that the custom theme is needed in addition on top after cloud code is done with the task but this is another example of guidelines that should be present also user enum missing filament interfaces with high severity of impact because if you use enums it's beneficial to use them deeper in all the project instead of just string values in other places. Let's take a look at details that it found. So yeah to show rows everywhere it should have has label has color and has icon. This is kind of a standard thing in filament now. So this project was about filament and this kind of proved to me that at least a few of those filament rules should be happening on top of Laravel boost default claude MD. And then my second experiment was much bigger project also with filament but with Laravel front end on top to make a reservation or order in the restaurant. The project description was from Upwork with 12 phases. Pretty complex one. So Claude code didn't one shot it. was 12 shots, one for each phase in total like one hour of work. And I decided to run the same prompt this prompt. So the project was produced by AI agent compare the code against those two guidelines. And this is again the table of things that were missed from the guidelines with severity high or medium or minor. So again, no filament smoke tests. Another proof that I do need to have that enforcement in the guidelines. And the same thing enum missing has label has color. So filament guidelines are definitely here to stay at least some of them in the future. But if we took a look at PHP Laravel guidelines some of the things are really minor. For example, I prefer to use helpers o helper instead of o facade. But it's just a personal preference. I don't care that much about it. And even cloud code identifies that as minor. Similar thing query create. I like to use eloquent query create only in select statements but on the other hand it doesn't really affect anything to be worse in fact it's more chance for autocomplete if I want to write code on top so again minor impact but this one again missing test for multiple features so same for filament same for laravel so I need to enforce the test or be more specific in what test to generate and probably I will shoot a separate video on that after more experiments. But yeah, if you don't enforce test in cloud MD generated or other guidelines file, the default Laravel boost guidelines claude MD and agents MD seem to be not strong enough enforcing the tests and even within the test there are a few issues identified hard-coded enum strings and missing some input. Let's take a look at examples. So yeah, same thing with enum actually that we discussed a bit earlier. Instead of this that should be enum if that enum already exists why not reuse it in all project including tests and this is also the test but the test don't ensure that the old input is present after validation errors also I think this is pretty important so what I'm trying to say here is some of the guidelines are important some are not that much anymore so for example no tests or flaky tests or incomplete tests this is really important or anything that would cause something to not work at all like custom theme missing for filament case or something like that. But I guess I would shorten the guidelines to avoid probably those minor things or for example obvious boilerplate comments. I don't even read those comments myself anymore that much. If those are AI generated, the code should be clear enough. This is kind of the classical approach. So, I'll probably spend some time with more experiments and shorten my guidelines to the points that really are crucial and important, but wouldn't overwhelm the LLM and AI agent with unnecessary noise. In fact, speaking about the noise, I'm not sure if you've seen on my other channel AI coding daily, I had this experiment. I deleted CloudMD and try to test if LLM works correctly after that. So I will link that video in the description below if you want to check it out. And also I will put the links to the current guidelines version for Laravel and filament that I showed in this video. Maybe you can take something from there or maybe you can advise me on what should or should not be in those custom guidelines. We can discuss all of that as usual in the comments below. That's it for this time and see you guys in other