Post

The new tools are basic with potential when used right, as usuall its the people that use them are cunts. Not everyone needs to know how C++ or whatever works just to get a simple result. Also fuck nodes. Utterly fuck them.

The commercial tech industry infected itself with the idea that computers are supposed to "just work" without deep understanding. The AI grift is just the next generation of conmen willing to take advantage of this learned helplessness.

Agree 100%

Adding an LLM to the mix is like adding a gambling layer. Maybe next time you win the jackpot. 🎰

so yeah, the tooling, the packaging, the deployment, so many patterns to context switch

I think rust is a bad example for that particular point because its use is to expose and formalise complexity that already exists, rather than create it where it otherwise wouldn't. Writing correct unmanaged code is *hard*. Scala definitely tries to elide complexity though

Seems to me you missed the Enterprise Java Stack, software has decluttered a lot compared to the 90s and 20xx

*me looking up how to open a file properly in {{some_language}}*

Note that the applications themselves largely haven’t gotten more sophisticated. 99% of apps are lists of stuff from a database with CRUD operations.

I feel increasingly like an old crank but I've been saying for decades that if you can't read and write the code in a text editor then it's too complex. I'll admit to a sliver of hyperbole, but I think it stands up practically too when taking into account abstractions that only leak in useful ways.

"ever-changing deep fried layer cakes" Too real

... which puts things that matter, like f. e. architectural skills, deep into the background.

For me the tools had a sweet spot some years ago. The languages and frameworks I use haven't evolved that much, but got a bit more convenient. But in the same time the tools got worse, slowing me down, get indeterministic UI, so I really have to concentrate and be careful. Weird.

Well put.

When they work, they correctly express the intent of a highly compressed statement as code edits. Then again, that is of course also how and where they fail ;)

For a novice to use AI to develop production software is like a novice directing the construction of a skyscraper without an architect. The workers will do what they're instructed, the way they always have; but, it is doomed to failure.

I just wish there was some tool that could make Make easier… life was so much simpler when I could just compile and then link code before making a binary… but then somebody decided to make Make so compilcated, and now they expect us to use AI????

Using something free becomes trivial when the thing you used to swear by costs money

As someone coming to dev ops at age 61 (I wanted to use Neocities with just CSS and HTML, but figured PHP was too good to miss out on) I am baffled by the vast array of nonsense that's out there. It's impossible to make a rational decision, I'm just hoping @ProcessWire comes good my hopes and dreams

Yeah. The model helps, but if the editor, CLI, file reads, or retries are rough, you’re still babysitting the loop.

Relate. I use Mistral to help me troubleshoot bad server configs because about ~75% of the time it correctly identifies the problem, but will offer solutions that don't work for my system configuration. I work faster with it, but I can't trust it too much.

Indeed, need to understand the theory first

This also sounds like AI!

It's a psychic hotline

Some are and some aren't. The ones who are using AI seem to be "producing" a lot more but the ones who aren't seem to be becoming more thoughtful.

Good for them

Personally, I think the tools are pretty great. The social institutions wrapped around them are a toxic wasteland, though. This split isn't limited to software. It's all over our professional culture, but it's extreme in software.

100% this. "LLMs are great at creating boilerplate code" yeah, but why do we need enough boilerplate code that that's a selling point? there are both individual (set up templates/snippets) and systemic (design better languages/libraries) solutions to this, and none of them are LLMs