The point that `pip` is the only "ships with python" tool can't be understated.
Most projects I've worked that have the "I need a few python tools" just generate the venv and such in a Makefile or similar, then source the venv's activate file and run whatever is needed in a make target. This is good enough in nearly all cases, and doesn't require that one of the other tools be installed at the OS level.
No matter how good the other tools get, not being in the default distribution == won't get used, no matter how good it is, because it can't be relied on always being installed.
Yeah definitely. That being said, conda competes pretty effectively in the scientific/numeric space, because pip does not solve the problems that these people have, and for quite some time would cheerfully break the environments of said people.
Very much so. As a bioinformatician, for me conda has become a central part of my workflow as a generic package manager through the conda-forge and bioconda channels. I know a few university HPC clusters that use it to handle software instead of e.g. environment modules.
Most projects I've worked that have the "I need a few python tools" just generate the venv and such in a Makefile or similar, then source the venv's activate file and run whatever is needed in a make target. This is good enough in nearly all cases, and doesn't require that one of the other tools be installed at the OS level.
No matter how good the other tools get, not being in the default distribution == won't get used, no matter how good it is, because it can't be relied on always being installed.