At least for me, writing the code // using the IDE // prompting the LLM to do the thing is not the hard part. The hard part is always understanding the actual problem // underlying assumptions // actual customer need and then architecturing the right solution to that. Actually implementing the solution is the easy part, and LLMs have made that now even easier.

But they've not really helped interpret customer requirements when they give you logically inconsistent / unimplementable business processes that need major re-vamping before they can be coded. To some extent they can help de-code poorly worded emails sent by some exec while golfing or in a meeting. But they still can't conjure information out from nothing. Nor are they that good at helping to play the political game when you have team X and team Y depending new feature Z, but feature Z requires completely changing how either team does process Ab but neither will even admit that their processes aren't compatible with each other.