He's right about how data pipeline work is regarded by most ML scientists. It often ends up being extremely hard work and is regarded by others as useless grunt work. There is no big reward, constant overtime due to unexpected issues and lots of stress.