If someone comes with a pretty decent idea and execution, it doesn’t really matter if code was written by human or llm, does it? Like if the next torvalds ships something great in the open, who cares how was it written.
In theory no, but for a successful execution you must have people that understand what's going on, otherwise the competitor that understands will eat you.
> who cares how was it written.
In the pre AI era a lot of sw was written by interns, juniors, some bigcon consultants, some offshore indian or east-european dudes, it didn't even matter before.
Everyone who wants it to work properly.
Companies used to release broken software on purpose since a very long time.
e.g. Break your code up into loosely coupled modules, don't have leaky abstractions, don't have huge God files, have good black box end-to-end testing, do the design work before the implementation, do lots of automated testing, have good documentation, etc.
Here is another one that was always true, end users only care about the observable behavior of your program. Does it solve the problem in a performant way? Is it reliable?
And another one, it was common when you inherited a large code base that was mature and written before you to not have any clue how most of it worked.
We have to accept that the art of nudging the text around is just not part of producing code for money anymore.
Now that we're seeing the quality of software we use worsening
Or do we want to be seeing that?
> Break your code up into loosely coupled modules
> don't have leaky abstractions
Not needed. The AI handles breaking up code and I mostly don't review abstractions.
> don't have huge God files
Less true than ever. The AI can read a 10k line file the same way it can read 20 files with 500 lines.
> have good black box end-to-end testing
The AI does the testing by default, and it is fine. I don't write the docs nor do I do design work.
Now what I said is certainly going to offend some users here, but that's how it is, and I know the quality of the product built with this workflow.
This workflow works well for software you develop alone, it is going to be trickier to adopt it in a team.
Still, in my opinion the writing is on the wall for our profession. No one can say with any certainty to what degree our expertise is still needed in the software development process in a year or two.
The AI handles
The AI handles
The AI does
I was reading a biography of Tovalds and one of the core things that mark the birth of Linux was the drive to learn. Same when reading “Coders at Work”, the will to learn and make things happen is a common recurrence.
I’ve never seen LLM coding that is not “I can’t bother to learn about this”. It’s always about magicking answers out of thin air, not about motivation to explore some fields.
nope. but, vibe coded slop is all we're actually seeing.