This was an interesting video, but this particular point in the video (starting at 7:28) was useful to me. LLMs don't think ahead, so by telling it to think ahead you improve the outputs. I told ChatGPT to do this always for me so now it does it with every answer and I can bypass it if I want to.
At a minimum, it streamlines the process to a better outcome.