the real shift is that throwaway code became viable for production workflows. i used to spend days writing reusable utility libraries. now i generate single-purpose scripts, run them once, and delete them. the economics of code reuse have fundamentally changed when generation is cheaper than comprehension.

the downstream bottleneck is real though. built a video production pipeline recently - generating the python glue code took maybe 10% of total project time. the other 90% was testing edge cases, tuning ffmpeg parameters, and figuring out why API responses were subtly different between providers. cheap code just means you hit the hard problems faster.