Comment by echelon
18 hours ago
What if vibe coding becomes 20x faster than normal coding? Are you going to stay old school and write artisanal code?
18 hours ago
What if vibe coding becomes 20x faster than normal coding? Are you going to stay old school and write artisanal code?
It may surprise you to learn that some people actually like programming, so yes I will. If AI tools are 20x faster then I guess I'll have to use them to get paid, but I'll be damned if I start letting a computer do the fun part for me on personal projects.
That said I'm not too worried. Vibe coding is currently slower due to how bad it is at writing software. In several years companies pouring billions into improving LLMs still haven't been able to make them not suck. That suggests to me that it's a fundamental limitation of the tech at present, and won't get better until another research breakthrough happens.
We've had AI assisted coding for less than half a decade.
The rapidity of development is astonishing.
These two statements can coexist. Yes, AI is amazing. And yes, it is not good enough yet to significantly speed up my work beyond research and writing tests.
Quantity was never an issue, quality is.
There's no silver bullet in software development.
Universal statements have a high burden of proof.
People used to claim we'd never fly. Shortly after we started, we reached the moon.
The entirety of the last 60 years of software may have been a low energy local optima.
The last 60 years of software gave us amazing projects, and if you go through their code, you'll see the same principles that is outlined in every good software engineering book: Good organization, hackish when needs be to resolve some accidental complexity, good comments,...
Most of those things rely on having the right mindset/philosophy first, then having a good grasp about the domain and the technologies (programming languages, platforms, libraries,...). After that you need to start thinking about the tools you used to help you (editors, test runners, static analyzers, debuggers,...). Most LLM users put the latter above all others. Like using the agent precludes knowing about the domain, the technology, and the tooling. And what philosophy? Craftmanship? Sir, here it's all about YOLO.