It's OK to use coding assistance tools to revive the projects you never were going to finish
21 points by jmillikin
21 points by jmillikin
It's also okay to accumulate projects you are going to someday(tm) finish.
I feel like the title could've stopped after "it's OK to use coding assistance tools".
I've used LLMs to write code for production systems, internal tools, side projects for fun, and small personal tools. I'm not sure why I should feel guilty for any of these?
Is it? Plastic has its uses; plastic shopping bags were once "free", littering the landscape wherever you looked. This is a project the author "really wish[ed] existed", but didn't care enough about to make it happen; and now, after spending a "free" credit, with all the externalities of LLMs we none of us can avoid hearing about on a daily basis, he has a "Sub-standard" result (I cannot believe that the name was chosen entirely without a kind of queasy irony) that either zero or at most one person cares enough about to occasionally use. Does this apparently feeble utility outweigh the externalities? Or perhaps we are simply ignoring them here, again, as we do practically every other in our lives?
On my side, my pending projects are defined to be fun to implement, they are not missing functionality of any sort. for example: implementing a SMTP server or writing a compressor; I don't have brilliant features in mind, only want to learn to do something different. And i also have a pile of books to read, but I don't want a summary, i can read that anywhere, but i want to enjoy the details
There's no software in your life that you want fit to your needs but it would be a chore to do so?
With the software I want, I need it reliable and free of error. I don't trust LLMs to produce that kind of reliability, and given the lack of an LLM to construct a mental model of the code, I do not trust that the code would be easily reviewable. I do not trust an LLM to not simply produce bunk tests that "pass" and have interesting titles. Code review is the least interesting part of programming, why would I want to turn all my projects into code review?! Especially for a programmer that isn't working off a consistent model of the code!
Outside of that, the most difficult part of software for me isn't writing the code, it's making the decisions and architecting it, which would still need to happen if I bothered to use LLMs to generate the code! Once the architecture is set it's not that difficult, most things are just gluing in the right way these days.
Imo we gotta reframe the conversation around writing code. It's not a chore to write software by hand and consider the design descions, but part of the process. I think ai is here to stay in software development, but we need to acknowledge the tradeoffs that using ai introduces.
The title seems inconsistent with the article; whether or not it's "okay" (whatever that means) isn't really addressed aside from a small section at the end that doesn't consider any ethical factors, consisting only of the author reassuring theirself that they won't lose their own programming skills by making use of AI in this way (nothing on how their use of AI might affect anyone else). You could just as easily argue that it's not okay despite the results that were demonstrated.
What if vibecoded projects drown out the actually good traditional open source projects?
Then the companies that sell the models win, because using their product to generate software for your specicufic use case is no worse than the median open source alternative that you would have to configure or customize.
The incentives are pretty ugly.
That's not what I said though. I meant as in being available, but disappointing. Like when you search ok GitHub and you have to filter out vibecoded projects one by one, you might as well give up.
Thankfully, only the good stuff will make it through to distros, to being packaged.