Wow.
This is code committed by developers who presumably know HOW to read & understand code, and know how it *should* be written.
In 10-15 years will there be any of us left who can?
It'll be the movie Idiocracy with the plants all dying & no one left who knows plants need water. 😆
https://techtrenches.dev/p/the-snake-that-ate-itself-what-claude
@syntaxseed The pressure that managers are putting on their devs to integrate AI into coding/workflow will come back to bite them. Not quickly enough. I see listings now that require AI coding.
Senior devs will bow out, junior devs won't know how to work with legacy code, and AI can't build with nuance or anticipation.
The worst will be when all new code is AI product so it's the only material to continue teaching AI.
@syntaxseed Quality was never a thing in software development. Even before AI fast delivery was more important, while the consequences of that would cost a company their costumers. Fixing this approach costs a lot of fixing time and money for maintanance.
But now there's AI and it can continue that permanent fixing with higher speed.
@syntaxseed I've been thinking about this a lot recently, the reason "clean code" and "good practices" exist is for humans to be able to wrangle the code for what needs to be done. LLMs don't face the same challenges, so having well structured code makes less sense for something that excels (by design), at pattern matching (really, tell an llm to verbatim repeat to you an arbitrary random string and it will never make a mistake, even when autoregressively generating it
That and also, fuck Clean Code, it's a performance nightmare and Uncle Bob should really not have the notoriety he has
@budududuroiu This supposes a future where humans will NEVER have to read or understand the code. I really hope we never become that enslaved to a couple huge AI companies - do you want *your* healthcare, insurance, banking, investments, social interactions & everything else run by systems that ONLY Musk & friends control?
That's a mind-blowing level of trust you have there. 😬
@syntaxseed No, this supposes that Claude Code is not a critical enough piece of Anthropic IP that needs careful hand-crafting. I doubt Anthropic is vibe-coding their distributed inference engine for Claude, Claude Code is a thin client for their frontier models. There still isn't an established pattern for tools like Claude Code, so there's no point in putting in the effort to make maintainable code when you a) don't know which part of Claude Code will be made redundant by next week's research and b) you (Anthropic) have infinite tokens to spend to rewrite Claude Code.
Claude Code's sole purpose is to get people to sign up for $200/mo subscription, the 3000 line `print.ts` file is irrelevant
Sherri W (SyntaxSeed)
Joe T
Claudio Zizza 🦜
Bogdan Buduroiu