The REAL speed up comes to efforts that are already well-designed, but require lots of human busy work. I've personally seen multi-day human efforts reduced to

The REAL speed up comes to efforts that are already well-designed, but require lots of human busy work. I've personally seen multi-day human efforts reduced to a 15-minute session with an LLM. In a way, LLMs are reducing implementation costs to the kolmolgorov complexity -- you can get what you prompt for, but you have to remember to prompt for everything you want to get -- which comes easiest if you already took time to consider the design.