Coding is solved? Software is not.
https://arcplane.ai/journal/software-is-not-solved
"Coding is solved" is a strange sentence.
AI agents are clearly generating a lot more code now. The more I use them, the more I feel software was never hard just because typing code was hard.
Agreed, and the gap shows up the moment the generated code hits a real edge case nobody specified. I've shipped agent output that compiled, passed lint, and still did the wrong thing, and the cleanup cost more than writing it would have. Code got cheap; deciding whether the result is worth keeping did not.
QA engineer — the "coding does not equal software development" line is exactly the gap I keep watching teams fall into.
Adding to skynet's list: ambiguity has layers. AI can clean up vague wording. It can't tell you the stakeholder doesn't know what they want yet. Most bugs aren't broken code; they're the wrong translation of someone's half-formed expectation. AI just makes us ship that translation faster.
The interesting question isn't "is coding solved." It's "what does QA become when coding is cheap." Probably less about catching defects, more about defining what "correct" means before the code exists.
Good post.
I think the interesting shift is that implementation is getting cheaper faster than ambiguity is disappearing.
AI can generate a feature very quickly now.
But teams still lose huge amounts of time on:
So coding becoming faster almost increases the importance of product clarity and QA, because more things can now be built before anyone fully agrees on what “correct” actually means.
The line that resonated most for me was:
“coding does not equal software development.”
That feels more true the better the coding models get.