The only way in the semi immediate future that AI coding is going to get better is with some combination of static typing, tools that can do guaranteed safe refactoring (which requires statically typed languages) and immediate feedback when a refactor causes breaking changes.
I’m imagining something like using C# where any code that you write that would cause a compilation error is immediately flagged solution wide + ReSharper that could reason over your code base and do refactors that were guaranteed safe since 2008 and an AI that calls those tools and does deterministic type checks along with running your unit tests.
I haven’t used R# in years. I don’t know how far it’s come.
I’m imagining something like using C# where any code that you write that would cause a compilation error is immediately flagged solution wide + ReSharper that could reason over your code base and do refactors that were guaranteed safe since 2008 and an AI that calls those tools and does deterministic type checks along with running your unit tests.
I haven’t used R# in years. I don’t know how far it’s come.