It has dropped by maybe MORE than 90%. My sons school recently asked me to build some tools for them -- I did this over a decade ago for them, for free. I did it again using AI tools (different problem though) and I had it mostly done in 30 minutes (after I got the credentials set up properly -- that took up more time than the main coding part). This was probably several days of work for me in the past.
"Building software" is a bit too general, though. I believe "Building little web apps for my son's school" has gotten at least 10x easier. But the needle has not moved much on building something like Notion, or Superhuman, or Vercel, or <insert name of any non-trivial project with more than 1000 man-hours of dev work>.
Even with perfect prompt engineering, context rot catches up to you eventually. Maybe a fundamental architecture breakthrough will change this, but I'm not holding my breath.
But in the past, you knew the codebase very well, and it was trivial to implement a fix and upgrade the software. Can the same be done with LLMs ? Well from what I see, it depends on your luck. But if the LLMs can't help you, then you gotta read the whole codebase that you've never read before and you quickly lose the initial benefits. I don't doubt someday we'll get there though.