Mike

Haha no lag but imagine... 😅.

Actually, there is one situation that is absolutely terrible: handling stop signs.

Law states you must come to a full stop. The car comes to a full stop.

Nobody else does that, hence anger is real from drivers behind me...

No no, you just can't see them. They apparently sit in South East Asia.

Half joking aside, my car has driven 85% of everywhere I go autonomously.

It still impresses me and it gets better with every software update.

Get in at home, enter destination, wait and observe. Get out at the destination.

I regularly search for $effect and find ways to remove it... Life is simply better that way 🥳.

It's called Thun, photo taken roughly from here: maps.app.goo.gl/qWJ4JCCBTVAk...

A view of a Swiss city with an emerald colored river, a castle, a lake and the snow-covered Swiss Alps in the background.

I often get asked why I left Switzerland...

The older I get, the less I know why 😅.

The photo is what Switzerland looks like on a sunny day. Locals consider this "normal".

{@debug ...} • Svelte Docs {@debug ...} • Svelte documentation

@debug

svelte.dev/docs/svelte/...

I did not know, glad I do now.

Also satisfying to see that I already implemented most recipes!

De-google was definitely the one that took most effort, but life is good now :).

When your dog vibe codes games, we probably reached peak hype.

I started my blog post where I share my experience vibe coding seaquel.app for the last three months.

I'm genuinely excited to start conversations based on my experience, especially with people who vibe code for life.

What you need to know is `--migrate prettier`!

months in according to an earlier post, I'd be curious to hear your opinion 4 months from now 😊

In theory it shouldn't, but I can't imagine people enjoy reviewing AI-generated code. What's the point of asking an expensive human to read code written by an inexpensive machine?

Sounds like if we ask humans to do that, we have a trust issue in what the machine generated.

Given you're only two

That's spot on how I meant it 😂

Haha I take anyone's answer :).

Your point is how I feel too. Reviewing an AI-generated PR takes extra long because of all the double checking and making sure it didn't add code that's not needed.

Interesting times regardless of how this all progresses over the next little while.

@andreasmoller.dk how do team members feel about reviewing AI-generated code?

Isn't there a feeling of "why would I spend my time reviewing code when the author didn't bother spending their time writing the PR?"

In theory, I agree that the process shouldn't change, but reviewing AI code feels odd.

In my case, the scary thing isn't that AI necessarily struggles, but it keeps adding (read: duplicating) code that already exists.

It had 13 implementations of copying a value and showing a success toast, 3 implementations of the Monaco editor, dozens of tab managers when 1 would have been enough.

Oh... "team", that's another level of "nobody every talks about it" 😅.

Lots of "we built these AI agent orchestration, our teams love it", but other than great SEO content in early 2026, I haven't found anything that had helpful content.

I'm actively trying to implement it for a client, not easy!

I'll read whatever many words you write down at whatever stage of polish they are :).

Got a list of bullet points? I'll read it :)

That's exactly where I need to spend more time, building that harness.

I make steady progress every day :). At times it feels worse than teaching a junior engineer because at least they have certain common sense :).

However, I'm not giving up just yet. There's lots of potential to get better.

Not for the first two months, until I got the first paying customer.

At that time, it felt like a liability not reading the code...

It started as a vibe coding experiment the way the cool kids talk about it. I went all in to build a first-hand opinion.

That's exactly my experience building seaquel.app.

There is no way I could have built that in three months without LLMs.

However, now people want to contribute and it's embarrassing at best, flat out a waste of people's time to try and add code.

Yeah, I continuously educate the LLMs.

I'm torn between "it'll all be worth the effort" and "all I'll ever get is average code at best".

The biggest issue I haven't figured out is how to teach LLMs to stop adding code and start using what's already in the code.

Me, "Where are the docs for that?" LLM, "Let me check the source code to find where Neon-Connection-String is actually used: I want to be honest with you here — I wasn't able to find any official Neon documentation that specifically documents the Neon-Connection-String header or the raw POST /sql endpoint pattern you're using."

Does anyone else start to get annoyed by AI hallucinations?

Slowly but surely, I think there are no actual long-term benefits in letting AI write code.

Dopamine hit? HUGE.

Benefits 6+ months later? Questionable.

Mainly for myself to feel good about the time I invest in Seaquel, I built a /metrics page.

Where possible, numbers are broken down by release. Click on cards with a 📈 icon to see details.

seaquel.app/metrics

It's the thought of me hanging in a basket below a giant balloon that could catch fire at any time that scares me 😂

Interesting, I did bobsled (when I was a teenager and had no fear) and skydiving as an adult. Both was fun.

I'd never do bungee jumping or go in a hot air balloon :)

I vibe coded a SQL client for two months and never looked at the code (that's what vibe coders do you know haha).

Two weekends ago, I launched a billing flow and got paying customers. That's when I looked at the code and I no longer vibe code the app 😱.

As for reviewing AI PRs... no thank you.

New in Svelte: Comments inside component/element tags!

Brings back memories 😅

LayerChart Composable Svelte chart components to build a large variety of visualizations

New LayerChart docs 🎉

next.layerchart.com

Massive overhaul including over 700 isolated examples with browsing / cross referencing / StackBlitz editing, AI friendly (llms.txt), component API docs, guides, improved search, showcase, improved design, and more!

Details👇

Introducing Chatto I'm building an Open Source group chat application that's simple to self-host, familiar to use, and easy to customize.

Let me summon @hmans.dev who's building www.hmans.dev/blog/chatto.