This is a great practice. I used to do it when writing complex algorithms, I'd do pseudo-code comments to outline functionality, then basically implement the comments line by line.

Now I sometimes use this practice when working with agents, if I need something done just a certain way. It's time consuming, but it produces good results.