Having experienced plenty of this but also having regularly experienced Claude Code writing correct, useful programs - it is reasonably possible that coding agents are the beginning of a disruption in how we code. They'll continue to improve (and we'll continue to surround them with tools and languages that feed that improvement cycle).
Combining early, faulty "worse is better" technology with a change in cost structure (tokens vs. programers) and capability (programs in minutes not days) often wins. Even when it sucks at first.
Yes, but who will write the code that trains the AI on future stuff? Someone somewhere will have to write original code, and without writing a lot of code that new original code will be of lesser quality than what went before. Once that cycle starts there will be less money to be made in programming, so fewer high grade programmers (and we already have too little of those as it is).
Combining early, faulty "worse is better" technology with a change in cost structure (tokens vs. programers) and capability (programs in minutes not days) often wins. Even when it sucks at first.