xlii

I'm running different projects in ChatGPT 5.5, Claude (Opus 4.7/4.7) and GLM 5.2 is nice - worth evaluating yourself :)

If so should we also consider PostgreSQL overengineered?

It's a shame OP decided to use Elixir as base, many ecosystems don't have mature task queues (e.g. for Rust I had to roll my own: simple_queue) so the space IMO would be more welcoming.

On OTP doubt anything can even make a dent in Oban user base.

That's also the reason why I can compose at all. The analogy would be like: I don't know how pink color is named, but I can say it's pretty.

And on the output: I don't know what colors I used, but the image is pretty.

Only my music teacher tried to play and I knew it didn't work well. It's a peculiar thing.

I have aphantasia and can't memorize sounds or recall them. For decades I thought I'm deaf (Ockham say hi). But I picked up piano, play for 3 years, can't discern C from G if my life depended on it but my friends tell me I'm pretty decent composer.

Writing this so people don't get disappointed about themselves just because they can't pick ear skills.

Till now nothing could be done about them as they claimed it was dramatizing only. No law prevented streaming fake-rape (and streamers were smart and implied the act while being off camera, so anti-pornography laws didn't apply)

Such streamers were wildly popular with youth, so crackdown happened. There's now law to back state versus that, but rest assured no one will try to pick people off the street. Polish law is riddled with laws allowing easy punishment but there's no will to exercise it.

It never got to the paper but they have way more people to connect than the sample.

I really enjoy this lag. Apple with the whole DMA made iPhone completely dull to my eyes. Previously? Updated yearly. Now? 3+ years without replacement and probably will stick to it for next 2-3 years.

Sure maybe in US Apple is fun. But in EU it's.. boring (and not like a Golang boring, just boring)

- https://github.com/exlee/pikchr_pl - DiagramIDE (diagram amiga-style workbench where you can script Pikchr diagrams using TCL, Prolog, Pikchr or - recently mruby). Note: you need to navigate to actual crate for description. There are binaries built in case someone wants to try it.

- https://svg.axk.sh - semi manual SVG fitter so that I can easier vectorize AI-generated pseudovector images (who doesn't like 30kb SVG versus 1.8mb PNG?!)

- https://github.com/exlee/rik - this one makes me laugh every time I use it - it's an AI harness with text editor as an UI (i.e. it reacts to comment strings) - I gave it personality so it makes wacky comments but other than that it's very constrained agent (limited edition ;))

These are not vibe coded but AI made it much easier to slide through major friction points (e.g. for SVG fitter I really didn't want to reinvent fitting algorithms)

https://github.com/phoenixframework/phoenix/issues/6697

In case you want to see files affected I made extended writeup on my blog - for reference. https://xlii.space/eng/elixir-cycles/

As for how the problem manifests: even obvious contract violations stops being shown (making it feel like "Dialyzer is useless") but the second tell is very long check times (tens of seconds up to minutes).

I spent 3 months analyzing failures caused by - what looked like - dirty builds but was caused by unstable compilation order. Which is quite obvious.

The solution is dynamic dependency resolution but this causes problem with macros.

The problem is easy to validate. Compile application multiple time and compare hashes. I'm not sure if it's sufficiently visible in bootstrapped Phoenix but I saw it in as small as <1000 LoC toy apps.

https://xlii.space/eng/from-rust-to-ruby/

The thesis that you're making is biased. Huge tech corps can move away from Rails, but it's similar to argument of "why the most successful people in the world don't drive Toyotas". Which is true, but it doesn't mean people should stop using Toyotas and buy Koenigsegg instead.

Typed languages have consequences. Some designs are either non-ergonomic or impossible. Rust: if you want to have a swappable adapter you're in Box<dyn> world. Many apps don't have to live in Box<dyn> at all but they need to test which is the sole reason to change architecture and wrap in boilerplate.

None of these reasons matter if you're multimillion tech corporation with unlimited resources.

But these are very important reasons to consider when you have small-medium sized team and cannot afford to fight language.

Called it rik, and it's on GitHub if anyone's interested checking it.

https://github.com/exlee/rik

Having worked across many disjunctive domains the value of "expert in domain" is greatly exaggerated. It's not like search engine doesn't exist nor that it cannot be easily absorbed. Quite often becoming expert is as simple as careful reading API documentation for existing, same domain system.

The piece is heavily one sided and don't recognize the problem of low reliability software: e.g. imagine a software designed in a way that it keeps critical piece of data as a global static variable. This will always work for a single user but will leak with two. Make it semi-critical and e.g. a part of profile loading.

Just as non-expert cannot verify if result is correct from a domain perspective, expert in domain cannot verify the output based on the basic principles (data security, safety, isolation, persistence, etc.) or even know that monetary values shouldn't be kept in floats.

Nothing like article claim had changed outside of a false promise of being productive.

Software engineers without domain expertise can fake it with unreliable results and experts without swe skills can fake it with unreliable results.

https://en.wikipedia.org/wiki/Betteridge's_law_of_headlines

Imagine Internet forum/StackOverflow etc. and a objectively bad description or attempt to a solved problem.

Pleasantry and back pats won't encourage the change while harsh words will either make poster go silent or push forward and improve.

AIs aren't people. They are statistical training sets which have majority of their sets consisting in human-like communication which gives them high probability of human-like speech. But it's not kindness or rudeness in the end. It's how good the pattern can scoop data out of training set.

Weird that searching for it didn't yield any results, though I stand corrected. However it seems that select is the actual "root" of documentation. Filter sends to find_all, find_all sends to filter or select.

It's a weird skill but after seeing website, screenshot etc. I need only couple of seconds to make an opinion. I then look into history of project etc., though it rarely contradicts first impression.

I think I do this because I don't trust thus I don't want to use such projects. Not because vibe coded/vibe-code driven projects are inherently bad (they aren't). Because I think low self-investment translates to low ongoing self-interest.

Since all software is buggy to some degree, I'm certain I'll find a dealbreaker that will never be addressed, and risk is rarely worth it.

Discoverability is much worse, LSPs aren't super helpful. Using Dash (documentation app) helps a ton. Still some names are confusing after switch (took a minute to recall that 'filter' is actually 'select' in Ruby).

I won't say that dev experience is bad but it's definitely different.

While personally I'm skeptical about LLM usage (and vibe coding in general) I'm not going to pretend it doesn't exist just on principle. I wouldn't risk rewrite 6 months worth of code to Rails but it's a perfectly good case for LLM conversion.

I mentioned in other comment that I reviewed it extensively. It's not a big project so outside of a few spicy bits it's mostly a web app.

I'd say that saying "not putting thought into it" is unfair. I didn't push the button and YOLO.

I did research about the tradeoffs and consequences. Snippets of Rust code vs Rails are the real thing and testability of Rust app is something I spent 2 months on.

As LLM enthusiast would say: context matters ;)

There's Google's genkit, charmbracelet's fantasy and LangChainGo. Each has ugly hacks and omissions. Then handling slice streaming of data into Elm architecture (bubbletea) is also complex.

So in theory nothing stand against but in practice one has to get quite low to the ground to get anything done.

Also: Golang agent exist! It's called crush and is developed by charmbracelet people. It's so-so though I prefer Pi myself.