Since you mentioned me, here is my take: I would spend more time using Pharo but it often takes to much time getting things running. It happens too often that I will see a project mentioned in Pharo News, want to try it, and then have to spend too much time getting it running.

There are advantages and disadvantages to image development and I usually don’t even take advantage of image based development in SBCL Common Lisp anymore: I usually have a load file to fetch data I will need and load in my code and required libraries. In the same way, I no longer keep multiple Pharo images around.

That said, it is a large world and you don’t need too many companies and individuals using a software technology to keep the ecosystem supported and moving forward.

I was also a big fan of my Xerox 1108 Lisp Machine, but that was mostly viable because Xerox had one large customer, the CIA. The CIA had a custom work environment for analysts that took advantage of the platform. In a similar way, I can see Pharo as potentially being an integration tool for knowledge workers with interfaces to external tools like MatplotLib and TensorFlow. A highly customized Pharo environment could be effective for organizations that could support a small team maintaining a custom analysis setup.