Marco Deleu

Working with Napoli on this has been a great pleasure. I loved the experience of making something extremely simple for developers and learn Typescript along the way. Turns out 7777 is so much better than I expected and it was also more challenging than I expected 👇

@IanLandsman I’m out of enterprise, but from what I’ve seen it’s kind of the opposite: everything takes forever and now John Doe from sales is implementing a prototype of what he wants with Claude and saying they can’t wait 3 months for it “its too easy, I can do it”

Bit of a hot take, but I think AI in the corporate world won't mean moving faster. It'll mean: similar speed, but no excuses for poor quality.

@rawkode You’re using kubernetes

You’re horizontally scaling a Kubernetes Deployment. The pod keeps crashing. What’s the problem?

man, CEOs will really say ANYTHING to sell you their narrative

@GrahamJCampbell @NoContextHumans You mean the best at causing rage, right?

@NoContextHumans UK one is best.

@idnan_se @kamrify You gotta agree that it seems impressive at first glance. The ‘before’ was a nice little website and the ‘after’ looks awfully generalized like a lot of successful businesses 🤷‍♂️

@kamrify Before was definitely better than the after 😀

before vs after brainstormed + redesigned with claude to optimize for conversions. let's see how it performs

@cdslash @adamdotdev God, I hate that shit

@adamdotdev At least they didn't go the Linux route, I still have no idea whether bookworm is better than buster, jammie or trixie

- 5.3 - 4.6 - 3.0 Kind of interesting that the big labs all have similar versioning, Google could have gone the chrome route and released Gemini 16 by now

@nicolaibaaring @calebporzio That way I can test my dashboard has the expected components and test each widget individually, with data and everything. That way my dashboard can never send the wrong data because it never passes data to widgets

@nicolaibaaring @calebporzio I’m working around it by having each widget reach out to a “Repository” class that compute the data, but implement memorization so that when two widgets need the same data, it still only compute once. I noticed I can have a dummy test on the dashboard to “see” widgets renders

I never really liked Livewire, it always felt weird and disjointed. But Livewire 4 blows it all away because SFC are like Vue components without switching languages or needing something like Inertia. I haven't found how I want to structure Livewire yet, but I love it @calebporzio

@nicolaibaaring @calebporzio In inertia I like how my backend can compute all data at once and inject in the page and I can test the backend calculations. In the frontend my page can pass down metrics down to components and I can have a test where I mount the page and assert the DOM of the final page

@nicolaibaaring @calebporzio I don’t like that level of isolation because it gives too much room for bugs to sneak in uncaught. My first Livewire project is a dashboard page (report) with 10 widgets. Testing the page component doesn’t work because each widget is a child component

@nicolaibaaring @calebporzio I would have liked to: If I have these records in the DB When I visit this url, it’s a livewire component that has these children and assert what I want to see on each children

@nicolaibaaring @calebporzio Livewire sort of bring back that problem, but at least writing test for it is somewhat more straightforward; we can just seed the DB and test the component. We can’t do that with Vue Testing (with or without inertia) unless we orchestrate a backend as well

@nicolaibaaring @calebporzio Vue and Inertia makes components have data flowing through “dependency injection”. Frontend testing you can just mount the component with props. It solves a long frontend issue of: mount, use a service locator, make api call, data arrives, set context

@tobias_petry @PovilasKorop I’m being forced into that train and I hate it 🫠 Quite sad that MySQL seems to really be on it’s deathbed

@PovilasKorop Preaching that for years.

It’s 2026, Just Use Postgres tigerdata.com/blog/its-2026-…

@MiPnamic @fedearg79 @taylorotwell @laravelphp There’s nuance here. It’s not slop by default. Laravel (and me as well) entered the world of programming at a time that “best practice” was the excuse given to force something down everyone else’s throat. So we question everything. If there is a reason, there is a fix.

“I started working in an era where caring about payload size was considered a best practice” That era never ended. As someone pointed out, enforcing JSON would never have shown that image, but I stand with you. Every single unnecessary thing shouldn't impact the size by much. Everything else is just slop.

I think it's time to explain how I came to write this post. First, I want to thank @taylorotwell and the @laravelphp team for taking the time to respond. I was working (well, actually Claude was working) on building a simple API with Laravel for a PoC, when it suddenly stopped responding. Weird. I killed the process, asked it to continue, and it hung again. Very weird. I checked where it was getting stuck — it was doing a curl to the API. OK, I tested it manually and saw a wall of HTML fly by. Alright, probably an error message... let me see what happened. But when I tried to scroll up to find the actual error, it was truncated. The HTML was larger than my terminal buffer. Definitely very weird. So I opened the same URL in the browser and saw the error page. The error itself was just two lines — a missing table in a query (I hadn't run the migrations). Note that I didn't need to scroll, so I never even saw the infamous SVG. But curiosity got the better of me. Why was such a seemingly simple page so heavy? And how heavy exactly? I piped the curl output to a file and checked the size: 1,844,013 bytes. A bit much for a 2-line error. That's when I found the SVG in the source, looked it up on the page, and decided to make the post. I stand by what I said in the original post: I think it's a bad idea to include such heavy decorative elements in an error page, especially one that sits at the bottom where most people will never even see it. As some comments pointed out, the SVG itself is actually ~0.5 MB — but it's included twice to create a flashlight effect, which makes it even more unnecessary. I started working in an era where caring about payload size was considered a best practice — where you'd use external CSS, images and scripts so they could be cached by the CDN or the browser. I understand the benefits of inlining assets, but in this case, it adds nothing of value.

@JackEllis I have practically the opposite mindset. To be young, full of energy and fully able to learn how to code in this new age using the best tools must be so exciting and full of potentials. They don't get hold back by being stuck in their own ways

I can’t imagine using AI to code without already knowing how to code myself. An absolute world of danger. With great power comes great responsibility.

@praneetaprakash @QuinnyPig @awscloud Who’s not scared to give AI credentials to AWS? 🙋‍♂️🙋‍♂️🙋‍♂️

@QuinnyPig @awscloud We have the managed AWS MCP server in preview! Check it out docs.aws.amazon.com/aws-mcp/

Honestly, watching Claude Code struggle to figure out how the FUCK the @awscloud CLI parameters and arguments work is incredibly validating.

@mattpocockuk I know a company founded in 2019 that is full JS and all my attempts to help them adopt TS was shot down. They’re very open to AI and AI use for development.

Quick check-in: Is ANYONE using untyped JavaScript with coding agents? Or is JS vs TS now a dead debate?