Comment by javhu

It's less prompt engineering, and more breaking down work into chunks an llm can do. With more powerful models you can do more autonomously, and with dumber ones, you have to step in and do work more often.