Controversial opinion: Tabs (the UI concept) were a mistake! Especially in the browser, they offer no advantage over separate windows, but come with the huge downside of having to use different keyboard shortcuts than the OS already provides for navigating windows.

Progress on my journey to digital sovereignty, leaving Apple and Google behind: - Sold my AirPods Max - Switched my search engine and browser For the most part, it won't be as hard as I thought; I only fear the day when my MacBook Pro needs to be replaced. markus.oberlehner.net/blog/escapin...

So Tailwind is now showered with sponsorship money. Good. Now that leaves us with hundreds of other open source libraries, blogs, and ad supported websites that are still struggling.

By far the most useful kind of tests! Also with 100% humans in the loop.

I have a perfectly fine iPad Air lying around that is completely broken. Just because it gets no updates anymore. Providing full Linux support should be mandatory after ending support.

Apple is the champion of providing magical experiences THAT JUST DON'T WORK. Just now, my girlfriend started typing in our new Wi-Fi password on her iPhone; *magically* a pop-up appeared on my phone asking if I wanted to share the password with her. I pressed yes, but it didn't work.

I decided to spend more time on writing and building. So, this might be my only conference next year. But I heard so many great things about @madvue.es that I had to apply for this one! I am honored to be part of it!

I've completely changed my mind on this recently: **You don't need to become a software company!** (and you definitely don't need to become an AI company)! Stick to your core competency!

On the flip side: Without people like the other Steve, the one who didn't develop emotionally past the age of 13, Woz wouldn't have had the money to fund museums and arts groups. And you and I wouldn't be staring at some Apple device reading this. Would it be a net benefit after all? Maybe.

Is AI able to write code? I think we can stop asking this question: yes, it can! Actually, it is way too good at writing (a lot of) code. In a surprising number of cases, when I ask an AI to write a moderately complex piece of code, it will produce an output that works.

I’m in both screenshots. I wonder if there should be two types of commits: 1. Quicksave: part of what I’m trying to achieve works, I want to be able to safely go back there and diff new changes against this state. 2. End-of-Level Save: I’ve finished a piece of work that is fully usable.

I find it frustrating that a lot of AI critiques out there are completely off the rails, claiming that AI does not provide any value at all and hallucinates 99% of the time. The only thing this achieves is to discredit any legitimate critique.

Is it worth automating things? I'm a software developer. My job is to automate things. Automation, over the last couple of hundred years, has made many of us rich (compared to our ancestors). So the answer should be simple, and yet it isn't... markus.oberlehner.net/blog/what-we...

Can you spot the button one has to press to add this boarding pass to Apple Wallet? Liquid Glass is an accessibility nightmare even for people with A+ eyesight.

The pattern I see when I read from people I respect, who describe how they make AI coding work for them, is treating AI more like a natural language input modality: Tell it exactly what you want and it will be able to understand and do it for you.

Sitting in a hotel lobby in Amsterdam, I began chatting with a Swiss neuroscientist. At the end of our chat, I was more convinced than ever: always choose Open Source software!

Reverse JSX? Instead of HTMLish syntax in JS, JSish syntax in HTML 🤭

I have a hard time getting excited about a future of even more centralisation where all of us (developers) are merely building widgets for OpenAI. And probably having to pay a hefty 40% tax. Yay!

If you want to follow my journey to regain digital sovereignty, you can read about it here: markus.oberlehner.net/blog/escapin...

Typing this on my MacBook, I think about this picture at least once a day. It has completely changed how I see my Apple devices, from loving how tightly integrated everything is to feeling like a prisoner of the Apple ecosystem. Not a single day passes without me making plans for my escape.

In preparation for next week's @viteconf.org, I gave @vitest.dev v4.x-beta a try. Let's look at how to use the new Visual Regression Testing feature and run it in a Docker container for consistent results across environments. markus.oberlehner.net/blog/visual-...

Vercel ➝ Netlify migration made simple. ✅ Checklist ✅ Next.js tips ✅ Best practices Start here → docs.netlify.com/resources/ch...

I asked the AI to impersonate me, and it did so without hesitation, following my writing about how to write good tests for Vue applications. A couple of thoughts about this:

Underappreciated upside of AI code reviewers: you can bluntly tell it that it's wrong!

Looking for my next role! I thrive where I can help others grow, and where my Vue.js open-source work is supported with time Frontend, AI, remote-friendly, and open to (some) relocations!

So @nuxt.com gets stable chunks! Seems like a small improvement, but it is huge! Big JS bundle sizes don't matter because they're downloaded once and then cached. Right? Wrong! Up until now, caching has barely done anything if we were deploying daily. Stable chunks fix that! youtu.be/eryRM-ovpyY

I get that as a tech CEO, you sometimes have to play the game, and you need a healthy dose of pragmatism. But how those tech CEOs can look themselves in the mirror kowtowing to this US president is beyond me. I find it frightening. Is there anything they won't do for financial gain?

Funny thing is: whenever I travel to North America I’m having a reverse Sputnik moment on the state of public transport. Luckily for @economist.com many Americans don’t have a passport so they’ll never know 🤫

Happy to be back talking about @vuejs.org! First time in Prague!

Controversial take: AI is merely a new I/O modality. In the past, we had structured text, graphical UIs, and "structured" voice (already AI-driven); now, with LLMs, we also have unstructured text (and voice). The unstructured text I/O modality allows us to do some impressive things, no question.

Recently I told my fiancé about my angst that I’m just not good at AI coding because I’m just not getting the results other on social media claim to get. She offered a great insight: “Maybe you’re expecting more from it than others?” I think that hits the nail on the head!

Enlightening perspective on the current state of LLM discussions.

Disabled GitHub Copilot a couple of weeks ago and forgot to turn it on again. 🤷‍♂️ Current thinking about AI coding: - Autocomplete is as useful as it is annoying - Vibe coding is fun, until it isn’t (and somewhat addictive)

If you want to have a particularly bad time, I recommend spending some time with Google Cloud (Console). How can this be a real product running significant portions of the internet? 🧐

✅ Want faster, more reliable tests without the flakiness? @markus.oberlehner.net shared how contract testing with Specmatic can replace brittle mocks and give you real confidence across your stack 🔥🧪 📽️ Watch his talk: youtube.com/watch?v=KTcF... #FEN25

I think I finally understand why abortion rights are such a huge deal in US politics. It's easy to get people agitated about the topic (because babies) and to picture political enemies who are for or against it as evil. Seeing the same thing now happening in Germany is scary!

How do we write high-quality software using AI? The answer is becoming clearer, and I'm glad to see that with Claude Code, and especially the latest release of @kiro.dev, we're moving in the direction I suggested in my article about how to build with LLMs. markus.oberlehner.net/blog/ai-enha...

100% code coverage? No! 100% feature coverage? Yes! Do not blindly chase metrics only because they are easy to measure. Instead ask yourself: What do we need to deploy with confidence doing as minimal manual testing as possible? Do whatever leads you there!

Is Vercel’s acquisition of (Nuxt)Labs a betrayal of the Nuxt and Vue.js community’s values? Fuelled by Guillermo’s Trump congratulations Tweet, Vercel and Next.js have faced growing criticism—though I’ve always found this narrative oversimplified.

What a surprising turn of events! 😳 Glad for the team! Yet I wonder how this turns out in the long term.

Revision 667: Kann eine KI Entwickler:innen ersetzen? workingdraft.de/667/ mit @markus.oberlehner.net Danke an workshops.de für das Sponsorn dieser Revision – und dafür, dass sie Teams fit machen, KI nicht nur zu verstehen, sondern auch sinnvoll einzusetzen. 🎓👏

Every few years, I increase my editor font size and line height a notch. Pro tip: add 2+4 points to your line height. Thank me later!

A question to the @vite.dev / micro frontend experts: I want to build vertical split mfs and share dependencies between individual apps (E.g. React/Vue). Is there a way to configure multiple Vite builds to use externalised dependencies? (Module federation is not an option because no SSR)

Claude 4 is my biggest AI disappointment since GPT4.5. Getting results like this is sobering. > Me: Update the rating functionality according to the test. > Claude: *Modifies the test* > Me: The test is fine! Fix the implementation > Claude: *Hardcodes (!) the expected behavior*

makes a good point, you get this by doing nothing. bsky.app/profile/andr...

I would love to know more details about this "unauthorized modification."

The perfect frontend framework features: - @vuejs.org's reactivity system - @react.dev's JSX, server components, and user base - @angular.dev's batteries included approach Do you agree?

enterJS is starting! If you’re here (alone) don’t hesitate to reach out to me or other people, let’s be conference buddies!

Great things coming up! Spoiler: I'll soon be back working more within the @vuejs.org and @nuxt.com ecosystem (among other frameworks)!

The “problem” with TDD I see most often is that it is a skill that you have to learn. You don’t just write good tests as much as you don’t just write good code. Yet bad tests offer none of the promised benefits while bad code still does something useful occasionally.