It seems like you get downvoted, but I think you touch on something important. IMO, right now there are two limitations with AI replacing experienced developers:
a) It's not good enough at programming. It sometimes goes down rabbit holes and cannot get out. In other cases it comes up with ridiculously complicated solutions that could be solved much simpler.
b) Making assumptions instead of gathering requirements.
I suspect that a) will get better over time. I also suspect that b) can be addressed by a pre-programmed prompt-flow that uses a LLM to gather requirements from a PM and ask probing questions to get a well-defined scope and agree on how edge cases should be handled. It doesn't seem far-fetched that a AI also would be able to call out small requirement changes that might allow for much simpler/faster solutions.
That is also what I think will happen, I mean these tools right now are just capturing market share with funding rounds and hacks surly? wrapping any foundational modal and trying to scope it down is always going to suck, but once people start to truly nail just training in only the information required to do that job (I described 03-high-mini or whatever it's called as a dumb finance bro with no depth to my wife) and then couch the task LLMs with orchestration LLMs, surly things improve?
I suspect that a) will get better over time. I also suspect that b) can be addressed by a pre-programmed prompt-flow that uses a LLM to gather requirements from a PM and ask probing questions to get a well-defined scope and agree on how edge cases should be handled. It doesn't seem far-fetched that a AI also would be able to call out small requirement changes that might allow for much simpler/faster solutions.