Hacker News new | ask | show | jobs
by danielvaughn 16 days ago
To be fair, "excited by the promise only to be disappointed by the reality of the implementation" describes ~95% of my experiences with all software over the last 20 years. In fact only a few exceptions really come to mind - git, treesitter, ffmpeg, and sqlite.
2 comments

Yeah, maybe it's rose coloured glasses on my behalf. Those examples you mentioned, I would 100% agree with. It's some of the best software out there. And yeah, there's probably always been rubbish about.

I guess I hope that the good stuff keeps coming and the dross falls away. More signal, less noise.

I do agree with you though. It feels like the industry has steadily been getting worse, AI is just like pouring kerosine on the fire. I'm almost embarrassed to call myself a software engineer now.

On a small bright note, I've gotten AI to help me produce some of my best work over the last couple of months. It may enable sloppy behavior, but it doesn't require it. I have hope that serious work will win out in the end, and that sheer human effort is still the differentiator.

Yeah, there was a good article on here the other day where the author suggested going slower with AI and using it to help produce higher quality output. I think the idea is to be quite "hands on", coding much in the old way, but to use AI to help with, for example, test coverage, error mode detection and handling, refinement of the solution/feature, etc.

At least that's how I read it. :-) I'm learning that there's a place for the LLM but it's the sandpaper, not the chisel.

Hah, that describes my relationship with git itself, actually.
lol I'm more speaking to reliability than the quality of the interface. git and ffmpeg are not exactly known for the most intuitive API surface, but I don't think I've ever encountered a bug with them in my 17 years. That's a pretty extraordinary thing when you think about it.
Fair point. My problem with git is actually mostly the flaw in the object model itself, more than the dismal API. The fundamental mismatch between "to get a clean history you have to edit/destroy history with squashes and rebases and whatnot" and "editing history destroys the ability to do comparisons of two branches, which basically ruins half of git's functionality from top to bottom when you encounter that problem".

Like even the basic question of "hey did I already merge this branch?" becoming unknowable if you autosquash-on-merge is just nasty.

I've got a million ideas on what the "correct" fix for that problem might be, but imho it's a flaw deep in the heart of git that creates a massive amount of pain.

But I'll give it credit for being rock solid and blazing fast, as you say.