I certainly appreciate your comments.

I would ask this about Q, however: Isn't upgrading from one version of Java to another quite a separate problem from generating code that is bug free? THAT problem appears to be a pattern match of X => Y, whereas generating code is a different problem. AND, vis-a-vis Q, I believe that AWS folks use this as an accelerator... there most likely IS an orchestrator in the upgrader that, given compiler access, can iterate until at least the code can compile. With that, there still is risk that code may not do what was intended, e.g., if the underlying semantics of the thing that now compiles changed...

I agree with your points about proper application, but I think the nuanced contexts of application are being exposed, and LLMs do not track that type of context, e.g., where and in what subsystem or system is *this* being applied and why?

We will see the tools improve, and we may be so fortunate as to be able to constrain the risks of putting LLMs on autopilot. That should not excuse anyone from running separate tests NOT generated by an LLM (no foxes guarding the henhouses).