At the very least it let me iterate on system configuration / installation procedures more quickly. I have learned a lot just due to tweaking Dockerfiles and rebuilding them. Virtual machines are much slower in this regard.
In Data Science Docker is becoming revolutionary. People tried distributing virtual machines to let others reproduce their work. Dockerfiles are much more reproducible, and don't need a few gigabyte of seperately hosted VM images. Also these VM images usually contain tons of undocumented "state", whereas a Dockerfile is easier to reverse engineer and ultimately very reproducible.
You can include Dockerfiles in all your projects. Other developers can then go in and get started with a minimal amount of guesswork. Turns out "sane" development environments, which probably means one supposedly optimal configuration/setup/framework for all your projects, is the exception rather than then the norm.