Easily 80 percent of my time at work is now spent in ArcGIS Python API, either the stable branch, or much of the time, the daily build. Frequently I switch between these environments, but also need a few other Conda packages to support machine learning. To support flexibly using both environments I set up these configurations using Docker, and shared these resources via a GitHub repo - ArcGIS-ML-Docker.
I defintely can vouch for them. Since creating these configurations about four months ago, they have become my default working environment. I use them in both a Windows 10, and two Macintosh environments. Still, my specialty is definitely GIS, not Docker and Dev-Ops. Consequently, I welcome any suggestions for improvement, especially if they are submitted as pull-requests (hint-HINT).
Still, having used them daily for over three months, they have proved extrmely reliable. If you are in a similar situation, needing a reliable, repeatable, and flexible environment for working with the ArcGIS Python API with the option to take advantage of Machine Learning if necessary, hopefully this helps you out!