Als tourist met laag tempo zeker twee keer zwaarder!

Inderdaad dus niet álles eraan gedaan om te winnen. Z’n eergevoel is te groot.

Wat doet onze overheid nog op X? | LUBACH YouTube video by LUBACH

Massive props to @lubach.nl (Dutch TV) for speaking out against X, recommending @bsky.app and Mastodon.

youtube.com/watch?v=u4hH...

TLDR: cache

Package Managers Need to Cool Down A survey of dependency cooldown support across package managers and update tools.

If everyone uses a release cooldown, do comprised packages get noticed in time? Or is it like the joke you hear during a safari: “you don’t need to outrun the lion, you just need to outrun your friend”?

nesbitt.io/2026/03/04/p...

The topic of skills atrophy has been running through my mind quite a bit. Delegating cognitive tasks to an AI system could mean over time you lose some of those abilities yourself. Similarly to how you need to keep training a muscle or a sport: use it or lose it.
But is this always bad per se?

Yesterday I had my iPhone battery replaced after some years. I had forgotten what a great feeling it is to only have to charge once per day. It lessens the desire to replace this 13 pro, which is otherwise working perfectly fine still 👌

Wow and remember the io js drama?

The migration tool helped a ton, but the last 10% is where we have to manually verify some minor differences in rules and plugins are taking most time. I don’t blame the tools at all, we just have too many plugins and an awkward @stylistic-eslint setup we need to migrate to oxfmt.

Hell yeah, the type aware linting is absolutely the best!

I do understand skills are a bit more flexible though. Personally I use skills with a lightweight description of an CLI and let the agent discover its usage through `—help` if needed. But I’m not following the argument that MCP is inherently wasting more context doesn’t sound right?

From a token efficiency perspective I don’t fully follow why agent skills would be more efficient than MCP. Doesn’t MCP expose a short description of available tools in the same way skill markdown files have some meta details in frontmatter?

No big blockers. It’s just that it makes the already laboursome transition from eslint to oxlint less of a speed improvement. Without js plugins oxlint is way faster, which is incredible! But there are some plugins that slow down everything again.

I have always been the kind of developer who only cared about outcomes. Give me the tool, the framework, the head start. I just want my website.

A totally different kind of developer exists who cares about the *craft*. Getting that function just right. Those devs are in pain right now.

I’ll probably keep using Nuxt, but in SPA mode. More than enough. Simple static hosting 👌

But the temptations and opportunities provided by vibe coding let you go overboard as well. I choose all kinds of tech that’s kinda new to me. @nuxt.com with CloudFlare workers + D1 as db. Client-side IndexDB.
It already needs simplification.

My girlfriend and I are taking Spanish lessons. We got a traditional book and studying goes well.
So of course I’m working on a tool to optimize learning. Using spaced repetition algorithms, flash cards and good quality spoken output. Pretty much Anki, but more specialised.

Weer een usb-c apparaat wat stuk kan! Van alle handelingen tijdens het koken is het met de hand raspen van Parmazaanse kaas zo'n beetje het minste werk en ook nog het leukst toch?

macbook nano ports. Usb-c 2 vs 3

Due to technical limitations there can't be two usb 3 ports. Awkward, but fine. But then the placement is a bit counter intuitive.
When both ports are in use, it's probably most likely the port more to the back of the device is used for charging, keeping the faster port occupied.

Scraping bots are getting good. They can pass CloudFlare challenges. We’ve been fighting a 40k “random” ip botnet for a couple of days now. CF does mark them as “somewhat odd” with a lower bot score though. That helps.

Official Agent Skills are now available for React Router

github.com/remix-run/ag...

I agree about it being unclear if it’s a warning that’s not enforced. A poor justification: at work we have some rules added as warn as a “temp” measure because they’ve been added to an existing codebase and it was too much work to fix all at once. Just like marking something as `@deprecated`.

Werkt die nog/weer met een Claude code abonnement? Of gebruik je een andere LLM? Dat werd vorige week geblokkeerd door Anthropoc.

"Collision detection using style queries plus keyframes" by lee

codepen.io/leemeyer/pen...

“If unsure”

Does that work? Are they self-aware enough?

Reselect 5 switched memoization strategy from using fixed size LRU caching (1 by default) to a WeakMap strategy, relying on the GC to clean stuff up. But when one of the selectors’ arguments is a JS primitive like a string, it doesn’t get cleared.

We ran into a memory leak in a Node service due to a misunderstanding of Reselect’s createSelector memoization. Might be worth writing up, or is that too dry? 🤔

Which brand is it? I assume the others you mentioned running Android are Garmin or maybe Wahoo? I’m also a bit annoyed by my Wahoo Elements' boot time, even though it’s a minor inconvenience.

Can’t wait to try it out! 👌

Why not look at P90 load times? They should stay consistent no matter the hardware.