You are failing to recognise all of the hard work which goes into "prompt engineering" to get AI to magically work!
Thankfully we have progressed so this time it will probably take less than 1000 years to progress to full blown chemistry :-)
Oh, and another thing, we still aren't able to quantify if AI coding is a net benefit. In my use cases the biggest benefit I get from it is as an enhanced code search, basically. Which has value, but I'm not sure it's a "$1 trillion per year business sector" kind of value.